PrompTom

Mises à jour du catalogue

Ajouté récemment

Le catalogue s'enrichit par lots. Voici les derniers, pour ne pas tout relire à la recherche du neuf.

14 août 2026

91 prompts

Two or three brand colours turned into a working set: success, error, hover, dark mode.

Claude / ChatGPTpalettedesign systemdark mode
You are a product designer. Expand a set of brand colours into a working palette.

Brand colours: {hex codes}
Product: {what it is}
Primary theme: {light or dark}

Deliver:
1. A neutral ramp: background, surface, sunken surface, two border weights, three text levels. The neutral is not grey — bias it toward the accent hue, or the interface looks assembled from other people's parts.
2. Semantic colours: success, warning, error, info — and why these rather than their neighbours on the wheel.
3. Accent states: hover, pressed, disabled, focus.
4. The same roles again for the opposite theme. Not an inversion: the same colours read brighter on a dark ground, so saturation has to come down.
5. Text-to-background contrast for every pair, as a number.

Name each colour by its role, not its look: violet-500 will be blue in six months and the name will start lying.
Voir un exemple de résultat

Background #09090F is not pure black: it carries 4% of the accent's violet. Error #F0616D rather than pure red: next to violet, pure red vibrates. Dark theme: the accent lightens from #6D28D9 to #A78BFA — dark violet on black reads as grey. Text on background: 14.8:1. Muted text: 5.2:1 — never go below 4.5.

What cannot simply be inverted: shadows, images, illustrations, borders.

Claude / ChatGPTdark modeinterfaceporting

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

One paragraph appended to every prompt — and the images stop drifting apart.

Midjourney / ChatGPT (vision)generationbrandconsistency
You are an art director. Build a standing prefix for image-generation prompts.

Brand: {name and what it does}
Mood: {three or four words}
Brand colours: {hex or names}
Examples you like: {links or descriptions}
What must never appear: {what grates}

Deliver:
1. A 40-60 word prefix: light, materials, palette, optics, composition. It gets appended to any request.
2. A "never" list of explicit exclusions. Without one the model folds in stock cliches — lens flare, perfect smiles, glass spheres.
3. Three complete example prompts: a cover, a product, a background.
4. What changes from image to image and what is never touched.

The prefix must work without the brand name inside it: the generator does not know the brand and will start guessing from the sound of it.
Voir un exemple de résultat

Prefix: soft diffused daylight, matte surfaces, deep violet and near-black palette, 50mm lens, shallow depth, generous negative space, calm composition, no text. Never: lens flare, glossy 3D spheres, stock smiles, neon grid, cluttered background. Never touch: light and palette. Change: subject and angle.

The text Figma Make or Stitch needs to build a coherent flow, not a pile of pictures.

Claude / ChatGPTinterfacefigmaflow
You are a product designer. Describe a screen flow so a generator builds it as one connected thing.

Product: {what it is}
User's goal: {what they came to do}
Platform: {web, iOS, Android}
Style: {two or three words}

For every screen give:
1. A name and a one-line job: what the person does right here.
2. The blocks top to bottom with real content — not "heading" and "text".
3. Exactly one primary action. Secondary ones sit apart and quieter.
4. States: empty, loading, error. Generators draw the happy path only; the rest has to be asked for outright.
5. What carries over to the next screen.

Finish with the shared parts: header, spacing, radii, type. Without them the screens come out of different products.
Voir un exemple de résultat

Screen 2, "Choose a plan". Job: see the difference and pick. Blocks: heading "Two plans, no asterisks" then two cards (Free / PRO $6 a month) then a line reading "Cancel any time". Primary action: "Get PRO". Secondary: "Stay on Free", as a text link. No empty state exists here. Error: payment declined — the card stays, a bar above it gives the reason.

FREE

What to show when there is nothing to show — and how to explain a failure without apologising.

ChatGPT / Claudeinterfacecopystates
You are a product designer. Write the empty states and errors.

Product: {what it is}
Screen: {where it is empty}
Why it is empty: {not started yet / deleted everything / search found nothing}
What the person can do next: {available actions}

For each case give:
1. A heading under six words: what is happening, with no "oops".
2. One line of explanation — why it is empty right now.
3. One action with the exact button text.
4. Whether an illustration is needed. Usually not: a drawing on an empty screen substitutes mood for an answer.
5. Separately, the error: what happened, what to do, whose fault it is. Own faults are stated plainly.

A newcomer's empty screen and a post-deletion empty screen are different screens. The first teaches; the second confirms it worked.
Voir un exemple de résultat

First run: "Your prompts will live here" / "Save the ones you like and they collect here" / button "Open the catalogue". After search: "Nothing for 'carousel'" / "There are 4 prompts containing 'card'" / button "Show those". Error: "Could not save" / "This is on our side and we are fixing it. Your text is still in the field" / button "Try again".

Pour designers

Interface microcopy

PRO

Buttons, hints, confirmations — the short lines a product is actually felt through.

Claude / ChatGPTcopyinterfacebuttons

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Eight conversations turned into decisions you can make tomorrow, not a retelling.

Clauderesearchinterviewssynthesis

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
PRO

How to push back on the bigger-logo note and be heard rather than filed as difficult.

Claude / ChatGPTnegotiationfeedbackclients

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour designers

Design handoff spec

PRO

Everything a developer will ask later, written down before they start.

Claude / ChatGPThandoffengineeringspec

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Rules that make ten drawings read as one set rather than ten discoveries.

Midjourney / ChatGPT (vision)illustrationstyleset
You are an art director. Define an illustration style for a product.

Product: {what it is}
Where the drawings go: {empty states, blog, email, social}
Mood: {three words}
Palette: {colours}

Set the rules:
1. Stroke weight, whether there is an outline, fill — flat or textured.
2. People: present or not. If present, how abstracted. Realistic faces age fastest and date the work first.
3. Perspective: front-on, isometric, top-down. One for the whole set.
4. How many colours per drawing — as a ceiling, not a preference.
5. What the drawing does: explains, decorates, or leads the eye. Detail level follows from that.
6. Three ready prompts showing the range: simple, medium, complex scene.
7. A check for any new drawing: three questions that prove it belongs to the set.

Another person must be able to apply these rules. "Friendly and modern" cannot be applied.
Voir un exemple de résultat

2px stroke, no outline, flat fill with 6% grain. People abstracted: the head is a circle with no features. Realistic faces look like clip art within a year. Isometric, 30 degrees, one angle for the set. No more than four colours plus the ground. Check: one stroke weight? Same angle? Four colours or fewer?

FREE

Five layouts that make posts assemble in minutes and stay recognisable.

ChatGPT / Claudesocialtemplatesbrand
You are a designer. Build a set of post templates for a brand.

Brand: {name and subject}
Platforms: {where you publish}
What you post most: {post types}
Brand assets: {colours, type, mark}

For each of five templates describe:
1. Its job: quote, number, list, announcement, before-after.
2. The layout: what sits where, how many lines of text fit, where the mark goes.
3. A character limit. A template that fits "as much as needed" becomes small type within a month.
4. What changes post to post and what is fixed forever.
5. How it reads in a feed beside its neighbours — at thumbnail size only the large thing survives.

Five is the ceiling. More will not be remembered, and people start laying out every post from scratch.
Voir un exemple de résultat

"Number" template: the figure fills the frame, a line under it up to 40 characters, mark bottom right. Limit: 40 characters, not "whatever fits". At 60 the type halves and dies in the feed. Fixed: mark position and 48px margin. Variable: the figure, the caption, one of three prepared grounds. In feed: only the figure reads — that is enough; the rest gets read by whoever opened it.

PRO

An email layout that accounts for mail clients ignoring half your CSS.

Claude / ChatGPTemailbuildnewsletter

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Twenty minutes down a list, and half the client's notes never happen.

ChatGPT (vision) / Claudereviewlayoutquality
You are an art director. Review a layout the way you would before showing a client.

Layout: {describe or attach}
What it is for: {the goal}
Who sees it: {client, team, public}

Walk the list and say where it does not hold:
1. What is seen first. Whether that matches what should be primary.
2. Alignment: what is almost on the grid — almost is worse than plainly off.
3. Spacing: how many distinct values are in use. More than five means some are accidental.
4. Type sizes: how many. More than six means there is no scale.
5. Legibility: where text sits on an image with no plate, where contrast drops under 4.5.
6. What happens to the longest name, the biggest number, the missing image.
7. The one place where you chose pretty over clear. There always is one.

Do not compliment. The job is to find what the client will catch, before they do.
Voir un exemple de résultat

The image is seen first, but the price is the point. The eye goes the wrong way. Nine spacing values: 12, 14, 16, 20, 24, 28, 32, 40, 48. Fourteen and twenty-eight are accidents — fold them into the scale. Text over the photo at the bottom: contrast 3.1. On a light frame it will not read at all. Pretty over clear: the caption is letter-spaced and reads as a heading.

What to show on each slide so an investor gets the business without your voice.

Claude / ChatGPTdeckinvestorsslides

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

One product, five different promises — to find out which one lands.

Claude / ChatGPTlandingherotesting

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

An audit of a designer's routine: what goes to a script, what to a model, what stays manual.

Claudeautomationroutineprocess

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

How to make ChatGPT and Perplexity point at you instead of a competitor.

Claude / ChatGPTgeovisibilityai search

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour marketeurs

A win-back sequence

FREE

Three emails to people who drifted away — no guilt, no begging.

ChatGPT / Claudeemailwin-backretention
You are an email marketer. Write a win-back sequence.

Product: {what it is}
Who left: {stopped visiting, did not renew, dropped off mid-way}
How long ago: {timeframe}
What changed since: {new in the product}
What you can offer: {discount, extension, nothing}

Three emails:
1. The first carries no offer. Only a reminder of what they came for and one honest line: we noticed you are gone.
2. The second is what changed. Specifically, not "we got better": two or three changes that touch this person.
3. The third is the offer and a direct question: come back or unsubscribe. Keeping someone on the list in silence is worse than losing them.

For each give a subject under 40 characters, a preheader, under 120 words of body, and one button. Plus how many days apart to send.

Do not apologise for their leaving. They owe you nothing, and an apology makes the email cling.
Voir un exemple de résultat

Email 1, subject "You were after listing prompts". Body: "Three weeks ago you took two Wildberries prompts. Nothing since — maybe they did not fit. If so, tell me how and I will answer personally." Button "Write back". Email 3, subject "Stay or unsubscribe". A direct question, two buttons, no discount on the third email — it would cheapen the first two.

FREE

A real case told so the sale does not stick out.

ChatGPT / Claudesocialstorytellinglinkedin
You are an editor. Build a story post around a real piece of work.

What happened: {the case, in your words}
Who was involved: {you, client, team}
How it ended: {result, better with a number}
What you sell: {service or product}
Platform: {LinkedIn, Facebook, Telegram}

Build:
1. A first line that is a moment, not a conclusion. "At 7:40pm on Friday the client wrote that launch was tomorrow" works; "I want to share a case study" does not.
2. The middle: what went wrong, what you tried, where you were wrong. Without the mistake there is no story, only a report.
3. The result, with a number and a timeframe.
4. A one-line lesson that applies beyond your case.
5. A soft ask on the last line: a question to the reader or an invitation to write. A hard sell kills both halves of a story post.

Length: under 200 words. Paragraphs of one or two lines. No bullet lists — they break the narrative.
Voir un exemple de résultat

First line: "At 7:40pm on Friday the client wrote: launch is Monday, there is no copy." Middle: wrote it over the weekend, realised on Sunday it was aimed at the wrong audience, rewrote half. Result: launch held, 34 enquiries in week one. Lesson: urgency is rarely about the deadline; it is about the task being set late. Ask: "Had one of these? Tell me how you got out of it.

PRO

A guide everyone writes by, models included — and the copy stops sounding like strangers.

Claudebrand voicecopyguidelines

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

What works on their page and what you can take without copying it.

ChatGPT (vision) / Claudecompetitorslandingteardown

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour marketeurs

Building the offer

FREE

Turning a list of what you do into a proposition people answer.

ChatGPT / Claudeoffersalesproposition
You are a marketer. Build the offer.

What you do: {service or product}
For whom: {segment}
What they end up with: {the result}
Price: {amount}
What they fear: {the main doubt}

Build:
1. The offer in one sentence: for whom, what they get, in what time, for how much. No adjectives.
2. Three versions for different segments — they fear different things.
3. What strengthens it: a deadline, a guarantee, a limit, a bonus. Take one, not all — four amplifiers read as a clearance sale.
4. What to cut from the current description: anything the buyer cannot verify before paying.
5. A test: retell the offer in the customer's words. If it runs longer than one sentence, it will not be remembered or repeated to a friend.

Do not promise what you do not control. "We will get you to the top" promises someone else's decision; "we will build the keyword set and rewrite your listings in 5 days" promises yours.
Voir un exemple de résultat

Offer: "For marketplace sellers — 20 listing prompts, 9 free, one-click copy." For a beginner: "Your first listing done in an evening, without a copywriter." One amplifier: half the catalogue is free — you can check before paying. Cut: "professional approach" and "individual attention" — neither can be verified.

PRO

Not a schedule, but an email answering what the person did or failed to do.

Claude / ChatGPTemailautomationretention

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour marketeurs

A referral mechanic

PRO

What to pay referrers and how much, without eating the revenue.

Claude / ChatGPTreferralsgrowthpartners

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

Half a page that shows what to do next, not how much happened.

Claude / ChatGPTreportinganalyticsweekly
You are a marketer. Write the weekly report.

This week's numbers: {paste}
Last week's: {for comparison}
What you did: {launches, posts, ads}
Who reads it: {yourself, your boss, a client}

Build:
1. A first line naming the week's main change and its cause. Not a metric list: those get read later, if at all.
2. Three numbers that moved and why. If the cause is unknown, say so — a guess is worse than a gap.
3. What worked and what did not, with timeframes: an article's effect is not visible in a week, and writing it off is premature.
4. One decision for next week. One, not six: six produces none.
5. What stopped making sense: where sources disagree.

Do not compare week to week blindly: holidays, a competitor's launch and plain weekends move numbers more than your work does. Mark those weeks in the report itself.
Voir un exemple de résultat

Main thing: visits up a third, cause was a mention in someone else's newsletter on Tuesday. Not our doing, not repeatable. Did not work: ads for the new section, 40 clicks, 0 payments. Stopping them. Too early: the article went up Thursday, search has not seen it yet. Decision: rewrite the skills section hero, 70% leave there.

Pour marketeurs

An objection bank

FREE

What you say to too expensive, let me think, and we already have one — written once.

ChatGPT / Claudesalesobjectionsscripts
You lead sales. Build an objection bank.

What you sell: {product}
To whom: {segment}
Price: {amount}
What you hear most: {objections in your words}

For each objection give:
1. What is actually behind it. "Too expensive" means either "I do not see the value", or "no money right now", or "I found it cheaper" — three different answers.
2. The clarifying question that tells those apart. Answering before asking is firing blind.
3. An answer for each case, under five lines.
4. What not to say: how bad the competitors are, "but we have", "it is only twenty cents a day".
5. When to agree and let go. Not every objection dissolves, and trying to dissolve them all costs you your reputation.

Add the objection nobody says out loud but everyone thinks. Usually it is "will you still exist next year".
Voir un exemple de résultat

"Let me think" usually means "it is not my call". Question: "Who else looks at this decision?" If it is not their call: give them something forwardable — one page with price and outcome. Do not say: "what is there to think about". Let go: if the quarter's budget is closed, agree on a reminder date instead of pushing. Unspoken: "you are new". Answer: half the catalogue is free, checkable without paying.

FREE

A first message to a stranger's project that gets answered instead of deleted.

ChatGPT / Claudepartnershipsoutreachemail
You handle partnerships. Write the first email.

Who you are writing to: {project, person}
What you know about their work: {specifics}
What you propose: {the substance}
What they get: {benefit on their side}
What you already have: {audience, product, numbers}

The email:
1. A first line about them, and not a compliment. "I have followed you for ages" is what everyone writes; "your Ozon guide has no section on video, and sellers keep asking about it" is what nobody writes.
2. The proposal in two lines: what you do and what they do.
3. Their benefit, as a number or a timeframe. Without one it reads as a favour request.
4. The smallest possible first step. Not "let us jump on a call" — a call is already a commitment.
5. A graceful out: one line permitting them not to reply.

The whole email under 120 words. Long first emails do not get finished, and only finished emails get answered.
Voir un exemple de résultat

First line: "Your Ozon guide has no section on listing video — it is asked three times in the comments." Proposal: "We have 20 listing prompts; I can hand over the video part as a ready insert. Your name on it, a link to us at the bottom." Benefit: "Closes a question you keep getting, without costing you time." First step: "Want the draft insert? Five minutes to read." Out: "If it is not your thing, just leave this — no hard feelings.

PRO

Questions people answer honestly instead of pleasantly.

Claudesurveyresearchquestions

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

The same product explained to a beginner, a professional, and the person paying.

ChatGPT / Claudecopysegmentsrewriting
You are an editor. Rewrite a text for different segments.

Original text: {paste}
Product: {what it is}
Segments: {three groups, who they are}
What matters to each: {if you know}

For each segment deliver:
1. The rewritten text at the same length.
2. What changed: which word became which, and why. Not "made simpler" — which word.
3. Words this segment does not know, and what replaced them.
4. The objection this version answers.
5. What stayed the same across all three. The shared part is your product; if nothing is shared, you have described three.

Do not substitute tone for segmentation. "Friendly for beginners, serious for professionals" is intonation, not segments. What changes is what you talk about, not how.
Voir un exemple de résultat

Beginner: "Not sure where to start — take a ready prompt and change only what is in braces." Professional: "Role-input-format structure, adapted to your process in a minute." The person paying for a team: "Five people stop writing prompts five different ways." Shared by all three: a ready structure nobody has to invent.

Pour marketeurs

A launch plan

PRO

What to do two weeks before, on the day, and after — by day, not in generalities.

Claude / ChatGPTlaunchplanproduct

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour marketeurs

An ad creative brief

PRO

A brief for a designer or editor that does not end in a redo.

Claude / ChatGPTadvertisingbriefcreative

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Where viewers leave and what to change in the edit so they finish.

Claude / ChatGPTeditingretentionreels

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

The opening line that stops the scroll, in eight versions.

ChatGPT / Claudehookopening linereels
You write short video. Rewrite the opening line.

Current hook: {as it sounds now}
What the clip is about: {the substance}
For whom: {audience}
What is on screen in second one: {describe}

Deliver:
1. Why the current one fails, in one line, without generalities.
2. Eight versions on different mechanics: question, number, mistake, disagreement, confession, before-after, warning, cut-off sentence.
3. For each, what must be on screen in that same second. A hook is not only words: a strong line over an empty frame holds nobody.
4. Two versions that work with no sound at all, as on-screen text.
5. Which to try first and why.

Every hook under 8 words. Longer and they are already scrolling.
Voir un exemple de résultat

Fails: "Let me tell you how I shoot reels" — promises a talk, not a result. Mistake: "I shot vertical for three years. Wrongly." Number: "40 clips, two worked. One difference." Cut-off: "Never open a video with…" On screen: a hand covering the lens — movement in second one holds better than a face.

A scene description for Sora, Veo or Kling that yields a clip rather than loose frames.

ChatGPT / Claudegenerationvideoproduct

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Showing how the thing works, without the words quality and convenient.

ChatGPT / Claudedemoproductscript
You write product films. Write a demo.

Product: {what it is}
Key property: {what sets it apart}
Buyer: {who}
The doubt: {what stops the purchase}

Script it by second:
1. 0-2: the property in action, with no introduction. Not "today I will show you a flask" but a flask being dropped.
2. 3-8: how it works, in one unbroken take. A cut here reads as a swap.
3. 9-14: the doubt and the answer to it. The exact reason people do not buy.
4. 15-18: scale or comparison — next to a hand, a coin, a familiar object.
5. 19-20: what to do next.

For each second: what is in frame, what is said, what is on screen.

The words quality, convenient, stylish and unique are banned. Each has to become an action: convenient means it opens one-handed, and that is visible.
Voir un exemple de résultat

0-2: the flask falls off a table onto tile, the lid holds, nothing spills. 3-8: one take — open one-handed, pour, close. No cuts. 9-14: the doubt is "it will taste of plastic". Pour water, smell it, cut to the steel interior. 15-18: beside a 500ml bottle — it fits the cup holder. On screen: "drops, does not leak", four words.

One page that makes brands want to work with you and know the price.

Claude / ChatGPTmedia kitbrandscollaboration

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

What the brand actually wants and what to ask before you shoot.

Claude / ChatGPTbriefbrandsprocess
You are a producer. Decode a brand brief.

The brief: {paste it as sent}
The product: {if the brief does not say}
What is on offer: {format and money}

Work through:
1. What the brand actually wants: awareness, sales, or footage for their own ads. Everything follows from that, and briefs rarely say it.
2. What contradicts itself inside the brief. "Natural, like you would for yourself" and "must say these five points" do not coexist.
3. What is unstated but will surface: rights, usage period, revisions, who approves.
4. Five questions for the brand before you start. Each one asked now is one redo that never happens.
5. What is unrealistic and what to offer instead.
6. Red flags: unlimited revisions, payment after publication, perpetual rights at the same price.

If the brand says "however you like", ask what specifically they liked in your work: usually one clip, and that is the logic to shoot in.
Voir un exemple de résultat

Actually want: footage for their own ads — hence the horizontal requirement. Contradiction: "lively and natural" plus an approved script word for word. Unstated: the usage period. Ask — "perpetual" costs double. Question: "Which of our clips did you like?" The answer sets the tone better than any brief.

How to say this is the third redo without losing the client.

Claude / ChatGPTnegotiationrevisionsmoney

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

A two-minute check: will it work for you or look forced.

ChatGPT / Claudetrendsdecisioncontent
You are a short-form strategist. Judge whether to take a trend.

The trend: {describe or link}
Your subject: {what the channel is about}
Audience: {who watches}
What you posted last month: {examples}

Check:
1. How old the trend is. Past two weeks and you arrive at an empty table.
2. What actually works in it: the sound, the edit, the phrasing, or the situation. That is what you copy, not the whole clip.
3. Whether it fits your subject without strain. Strain shows instantly and costs more trust than it earns reach.
4. What must change so it is your clip rather than a repeat.
5. Who in your niche already did it. Five means late; zero means check whether there is a reason.
6. The call: take it, take it changed, skip. With one line of why.

Skip is a normal answer. A channel that takes every trend is about nothing within a month.
Voir un exemple de résultat

Five days old — in time. What works is not the sound but the phrasing: "things I stopped doing". Fits: "prompts I stopped writing by hand". Two in the niche did it, both on other subjects. Call: take it changed — keep the phrasing, drop the sound, it is not ours.

What to film back to back so the rest of the month is only editing.

Claude / ChatGPTshootingplanningbatching

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Not yes/no for engagement, but a question whose answer changes something.

ChatGPT / Claudestoriespollsales
You produce social. Build a story series around a poll.

What you sell: {product}
What you want to learn or nudge: {goal}
Audience: {who watches}
How many stories you will shoot: {number}

Deliver:
1. The chain: the story before the poll, the poll, what shows after each answer. A poll with no follow-up is a wasted contact.
2. The question with two options people genuinely choose between. "Do you like it?" with "yes" and "very" tells you nothing.
3. What to show the people who picked option one and what for option two. Different answers mean different doubts.
4. Where the link goes and why there: right after the answer, while the person is in contact.
5. How to use the result in the next post. A number from your own poll is the strongest argument: it is about these very people.
6. What not to do: polls for view counts, three polls in a row, questions with an obvious right answer.

The question must be one whose result you actually want to see. If you already know the answer, it is decoration, not a poll.
Voir un exemple de résultat

Before: a frame with two listings, one with video, one without. Poll: "What do you look at first in a listing?" — "photos" / "video". Photo voters: a story on building the infographic. Video voters: the generation prompt. Link right after the answer, not at the end of the series. Next post: "68% of you look at photos first — so video at the bottom of the listing will not save it.

What to do on camera when you have no product and no pretty location.

ChatGPT / Claudeshootingcameradelivery
You are a director. Help shoot a clip that is just a person on camera.

Subject: {what you talk about}
Length: {seconds}
Location: {where}
What gets in the way: {stiffness, reading off paper, monotone}

Deliver:
1. The script broken into 6-8 second chunks. Long paragraphs cannot be held in the head, and reading off a page shows in the eyes.
2. What to do between chunks: change framing, step, turn, gesture. A still face holds nobody for 30 seconds.
3. Where to place a pause. A pause before the main point beats volume.
4. What to cut from the speech: filler, "actually", "sort of", apologies for the recording quality.
5. Where to stand and where to look. Into the lens, not the screen — otherwise the gaze drops and reads as uncertainty.
6. What to put behind you and what to remove. The background must not be more interesting than you.
7. Three takes with different delivery: calm, fast, with pauses. Cut the best one.

Do not memorise the words. Memorise the order of the thoughts — memorised text is audible and sounds like an advert.
Voir un exemple de résultat

Chunk 1 (7s): "I shot forty clips. Two worked." Pause. Change framing. Chunk 2 (8s): "The difference was not the lighting or the edit." Cut: "so", "basically", "sorry about the audio". Background: close the open wardrobe — the eye goes there.

What exactly you sell a brand, and what the word perpetual costs.

Clauderightscontractmoney

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Opening lines written not by you but by people who already bought.

ChatGPT / Claudereviewshookscontent
You write short video. Mine hooks out of reviews.

Reviews: {paste, unsorted is fine}
Product: {what it is}
Platform: {where you post}

Deliver:
1. Ten lines from the reviews that work as a first second. Verbatim: a buyer phrases it better than marketing, because they use their own words.
2. Five doubts that repeat. Each is a clip: the doubt in second one, the answer in second two.
3. Three negative reviews that make good clips. An objection you name yourself lands better than one you avoid.
4. The words buyers use for the product. Often not yours — and they will search with theirs.
5. What reviews praise that you never mention. Usually that is the main thing.

Do not tidy the quotes. A review hook works precisely because it sounds like a person rather than copy.
Voir un exemple de résultat

From a review: "bought it for the cottage, turns out I use it more at home" — a ready first line. Repeating doubt: "it is probably heavy". Clip: put it on the kitchen scale. Negative review: "the instructions are useless" — shoot your own in 20 seconds. They call it something else: you write "insulated tumbler", buyers write "cup with a lid".

Why it did not land, in order, rather than by feel.

Claude / ChatGPTanalysisstatsmistakes

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

A list of the ones you genuinely fit, instead of a blast to everybody.

Claude / ChatGPTbrandsprospectingcollaboration
You handle partnerships. Build a brand list for a creator.

What you film: {subject}
Who watches: {audience in detail}
What you have advertised: {if anything}
Your format: {clips, stories, reviews}
Geography: {country, city}

Deliver:
1. Twenty brands in three groups: obvious, non-obvious, and those already working with creators your size.
2. Why each fits — one line about audience overlap, not about the product being good.
3. Who decides inside those companies and how to reach them: a specific role, not a general inbox.
4. What to say in the first message to that particular brand: a hook from their current work.
5. Who not to write to and why: too large, agency-only, recently burned by creators.
6. The order: who to start with so the first reply arrives sooner. Starting with your dream brand is a reliable way to be disheartened by week two.

Twenty specific emails beat two hundred identical ones. Two hundred identical ones do not work at all.
Voir un exemple de résultat

Non-obvious: analytics tools for sellers. Same audience, and they barely hire creators — no competition. Decides: the marketer, not procurement. Search the company name plus "marketing". Hook: they published a teardown last month that is missing video. Do not write: large marketplaces directly — their vendor is booked a year out. Start: with the non-obvious ones. They reply faster, and then you have something to show the obvious ones.

PRO

A Sora, Veo or Kling prompt for your product — and what to film on a phone.

ChatGPT / Claudevideogenerationlisting

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Resolution, background, margins and shot order — before upload, not after rejection.

ChatGPT / Claudephotosrequirementslisting
You manage marketplace content. Build the photo requirements.

Product: {what it is}
Platforms: {where you list}
What you have: {a shoot, supplier photos, renders}
Category: {clothing, electronics, cosmetics, other}

Deliver:
1. Each platform's technical specs: size, aspect ratio, weight, format, background. One photo does not fit all — some demand white, some do not.
2. Shot order: what comes first, second, third. The first frame decides whether the listing is opened at all.
3. The minimum number of shots and what each shows. Empty slots damage trust more than mediocre photos do.
4. What to add: scale beside a familiar object, the underside, what is in the box, the product in use.
5. What is not allowed: other brands' logos, price tags, text collages where text is banned, watermarks.
6. What to do with supplier photos: where they are usable, and where a hundred other sellers already run them and they work against you.
7. A pre-upload check: how the first frame looks as a 200-pixel thumbnail. That is how it will be seen.

Point 7 outweighs the rest: listings are chosen in a grid where the photo is the size of a fingernail.
Voir un exemple de résultat

First frame: the whole product, 10% margins, white ground. The silhouette reads at thumbnail size — that is enough. Second: scale beside a hand. "How big is it" is the most common question. Do not use the supplier photo as the first frame: they currently sit on 40 other listings. Add: the underside and the box contents — those cut returns most.

The frame you get chosen by, in a grid of forty near-identical products.

ChatGPT (vision) / Claudemain imagesearch resultslisting
You design listings. Review or design the main image.

Product: {what it is}
What the photo shows now: {describe or attach}
Competitors in the grid: {how their photos look}
What sets you apart: {the property}

Work through:
1. What survives at 200 pixels: the silhouette, the colour, one detail. Nothing more fits, and that is where the choice happens.
2. How your frame differs from its neighbours in the grid. If everyone is on white from the same angle, difference beats beauty.
3. Whether the product fills the frame. A small object in a large field looks cheaper than it is.
4. Whether text belongs on the photo and which text. One word or one number at most, and only where the platform allows it.
5. What to remove: props, shadows, a distracting background, a second product in frame.
6. Three versions to test: different angle, different scale, with and without text.
7. How to compare honestly: screenshot the grid, paste your versions into it, and look from across the room.

Point 7 is mandatory. A photo judged alone always looks fine; it has to be judged in a row of other people's.
Voir un exemple de résultat

At thumbnail size it is a dark rectangle — no silhouette. Rotate 20 degrees and it gains volume. Everyone in the grid is on white. Use light grey: it separates without looking foreign. Remove: the branch and the stones. Props become dirt at thumbnail size. Text: "500ml" in the corner — that is the question asked most often in chat.

Pour marketplaces

A wave of bad reviews

PRO

What to do when the rating slides: in order, without panic.

Claude / ChatGPTreviewsratingcrisis

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Sizing a batch between two disasters: out of stock and dead on the shelf.

Claude / ChatGPTpurchasinginventorymaths

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour marketplaces

Join the promotion or not

PRO

The maths before you click: what survives the discount, the fees and the return spike.

Claude / ChatGPTpromotionsmathsmargin

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

A check in order: what actually happened and which of it is fixable.

Claude / ChatGPTrankingdropdiagnosis

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Porting without copy-paste: different rules, different buyers, different titles.

ChatGPT / Claudeportingplatformslisting
You manage listings. Port a listing to another platform.

Current listing: {title, description, attributes}
From: {platform}
To: {platform}
Product: {what it is}

Deliver:
1. A title under the new platform's rules: some weight the first 60 characters, some expect "brand + type + attribute". A copied title almost always breaks the new rules.
2. The description: what to rewrite, what to delete. Links to the other platform, its promotions and its terminology have to go — listings get hidden for those.
3. Attributes: which fields are mandatory here that did not exist there. Empty fields cost position.
4. Category: where it belongs. The same object sits on different branches per platform, and the whole grid depends on it.
5. Photos: what to reshoot or recrop for a different aspect ratio and background rule.
6. Price: how to absorb different commission and logistics so margin holds.
7. What is forbidden here that was allowed there: text on images, claims, comparisons.

Do not try to port reviews and rating — they cannot move. Assume the first weeks carry no social proof, and decide what replaces it.
Voir un exemple de résultat

Title was "Thermos 500ml black matte stainless for tea and coffee". Here the first 60 characters decide: "Thermos 500ml stainless steel, matte black". Delete from the description: "on promotion" and the other platform's name. Fill the mandatory fields: capacity, liner material, heat retention hours. Absent there, required here. Photos: recrop the main one to 3:4 and widen the margins.

FREE

The text people read before buying from a shop they have never heard of.

ChatGPT / Claudebrandshoptrust
You are an editor. Write the brand blurb for a seller page.

What you sell: {range}
Who you are: {manufacturer, distributor, reseller}
How long you have traded: {duration}
What sets you apart: {actually, not as a slogan}
What matters to the buyer: {delivery, warranty, sizing}

Deliver:
1. A first paragraph under three lines: who you are and what you sell. No founding story, no "mission".
2. What answers the main doubt about an unknown seller: where it is made, what the warranty is, how you handle defects.
3. Three concrete facts with numbers. "Trading since 2019" is checkable; "years of experience" is not.
4. The range in one line: what else you carry. Second purchases come from there.
5. How to reach you and when you answer. A stated response time removes more doubt than a promise of quality.
6. What not to write: "dynamically developing company", "team of professionals", "individual approach".

Write it so it could not be pasted onto another seller's page. If it could, you described anyone, not yourself.
Voir un exemple de résultat

First paragraph: "We sew bags in Tver, since 2019. We sell only our own — what you see in the photos is in our warehouse." Doubt: "One-year warranty on hardware. Zip broken? Send a photo in chat, we ship a new bag and you keep the old one." Facts: trading since 2019, 11 models, average chat reply 40 minutes on weekdays. Do not write: "we value every customer".

Public answers read not by the asker but by the hundred people after them.

ChatGPT / Claudequestionslistingconversion
You manage a marketplace shop. Answer the questions on a listing.

Product: {what it is}
Questions: {paste}
What the listing is missing: {if you know}

For each question:
1. An answer under four lines, with a number or a fact. "Yes, it fits" is not an answer; "It fits, inner diameter 74mm" is.
2. What to add to the listing so the question stops being asked. A repeated question is a hole in the description, not curiosity.
3. If the answer is no, say so plainly and say who the product does suit. An evasive "depends" produces a return.
4. Separately: the three most frequent questions and exactly where the answers belong — attributes, description, photos.
5. One question nobody asks but should: the thing that most often causes returns.

Write answers that still read in six months. "Currently in stock" expires in a week and starts lying.
Voir un exemple de résultat

Question: "Will it fit a cup holder?" Answer: "Base diameter 68mm, fits a standard holder. Narrow car holders at 65mm, no." Add to the listing: base diameter in the attributes and on photo five beside a ruler. Not asked but should be: dishwasher safe. It causes a third of returns.

A range review by money: what to keep, what to reprice, what to drop.

Claude / ChatGPTrangeanalysisprofit

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

When to follow down, when to hold, when to raise — by rules rather than nerves.

Claude / ChatGPTpricingcompetitorsrules

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour marketplaces

Launching a new product

PRO

The first thirty days, week by week, so the listing does not die without reviews.

Claude / ChatGPTlaunchnew productplan

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

Building a name that search finds and a person understands.

ChatGPT / Claudetitleseolisting
You build marketplace listings. Write the title.

Product: {what it is}
Platform: {which}
Keywords: {if collected}
What distinguishes it: {size, colour, material, purpose}
Brand: {yes or no}

Deliver:
1. The formula for this platform: the order of type, brand and attributes. It differs by platform, and both search and legibility depend on it.
2. Three titles of different lengths: short, medium, at the limit.
3. What lands in the first 60 characters. Those are visible in the grid and weigh more than the rest.
4. Which words to drop: "buy", "cheap", "bestseller", repeated stems. They do not help search and eat space.
5. How to fit secondary keywords without turning the title into soup. Platforms demote soup.
6. A check: read it aloud. If it is a word list rather than the name of an object, search may find it but a person scrolls past.

A title is written for two readers at once: the search engine and the human. The second is usually sacrificed — which produces impressions without orders.
Voir un exemple de résultat

Formula: type + capacity + material + colour + purpose. Short: "Thermos 500ml stainless steel black". First 60 characters: type, capacity, material — that is the click decision. Drop: "buy", "thermos thermos", "for tea coffee water drinks" — four words for one thing. Aloud: reads as the name of an object, not a list. It passes.

What to ask before paying and what to check once the boxes arrive.

Claude / ChatGPTsuppliersreceivingquality
You buy goods. Build a supplier check and a receiving procedure.

Product: {what it is}
Supplier: {where from, how found}
Batch size: {units and value}
What you already know: {correspondence, samples}

Deliver:
1. Ten questions before paying. Production, lead times, defects, packaging, documents, what happens if they are late.
2. What to check in the sample and how: what to measure, what to pull, what to wash, what to leave for a week.
3. How to agree on defects in advance: what rate counts as normal, who pays for returns, how it is recorded.
4. Receiving: how many units to open from a batch of 100, 500, 1,000. Checking everything is impossible; checking three is pointless.
5. What to inspect, in order: contents, dimensions, function, packaging, labelling.
6. What to photograph immediately. A claim without photos taken at opening is refused almost everywhere.
7. Red flags in correspondence: dodging defect questions, rushing payment, refusing a sample, changing price after agreement.

A sample from the batch and a sample "to look at" are different things. Ask for one from the batch being shipped to you.
Voir un exemple de résultat

Before paying: "What defect rate do you consider normal, and what happens above it?" Dodging is itself an answer. Sample: fill with boiling water, leave 24 hours, then check the smell and the gasket. Receiving 500 units: open 20, from different cartons, not off the top. Photograph: carton labels, the packing layer, and every defect with a ruler in frame.

Find the action after which users stop leaving, and start counting it.

Claudeactivationmetricsretention

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
PRO

How to ask why and offer a way out without holding people hostage.

Claude / ChatGPTchurncancellationretention

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Half of lost subscriptions are not a decision to leave but an expired card.

Claude / ChatGPTpaymentssubscriptionsrevenue

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Help at the moment someone is stuck, not seven modals on first login.

Claude / ChatGPTonboardinghintsinterface
You are a product designer. Replace the product tour with contextual hints.

Product: {what it is}
Current onboarding: {tour, video, nothing}
Where people get stuck: {if known}
The key action: {what they must do}

Deliver:
1. Three or four moments that need a hint. A moment is not a screen but a state: empty, first time, after an error, before something irreversible.
2. For each: what to show, in one sentence, and where. A hint lives beside its element, not in the middle of the screen.
3. The show condition and the condition for never showing again. A hint that returns on the tenth visit is an irritant.
4. What the interface should solve instead of a hint. If a field needs explaining, renaming the field is usually cheaper.
5. What to delete from the tour entirely. First-login tours are watched by 20% and remembered by nobody: the person does not yet know why they would care.
6. How not to obstruct experienced users: one way to dismiss everything.
7. What to measure: share reaching the key action, before and after.

Rule: a hint appears once the person is already trying to do something. Before the attempt there is nothing to explain — they do not need it yet.
Voir un exemple de résultat

Moment: an empty favourites list. Hint by the heart button: "Tap to keep a prompt here." Gone for good: after the first save. Solved by interface, not a hint: rename the "Slug" field to "Page address" and the hint becomes unnecessary. Cut from the tour: 5 of 7 steps. Keep one — where the sections are. Measure: share who copy their first prompt within the session.

Most churn happens in the first month — here is what happens week by week.

Claudeonboardingretentionplan

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Who to tell, when, and in what words, so a handful leaves rather than half.

Claudepricingsubscriptionscommunication

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

The scores are in — now turn them into decisions rather than a slide with a number.

ChatGPT / Claudesurveyfeedbackdecisions
You are a product analyst. Work through survey responses.

What you asked: {the wording}
Scores and comments: {paste}
How many answered, out of how many: {numbers}
Product: {what it is}

Work through:
1. Who answered. Extremes answer — the delighted and the furious. The middle stays silent and is absent from this data.
2. Themes in the comments with mention counts. One mention is not a theme.
3. What you actually intend to fix and what you do not. A list nobody intends to build is worse than no list.
4. Who to reply to personally and what to say: a low score with a coherent comment is a free interview, if you write the same day.
5. What to do with high scores: ask for a review or a referral, but not in the same email where you thank them.
6. One product change for the coming month.
7. What to ask next time so the data is more useful.

The score itself decides nothing. The comments decide, and so does who wrote them: a complaint from a large customer and one from a free plan are worth different amounts.
Voir un exemple de résultat

84 of 1,200 answered — 7%. Almost no middling scores; this is not a picture of the base. Themes: slow search (19 mentions), no dark mode (11), plan is unclear (9). Fixing: search. Dark mode already exists — so it is not being found, which is a different problem. Reply personally: the 6 people scoring 3-5 with a coherent comment, today. Next time ask: what they did in the product this week.

Pour le SaaS

A switch-from page

FREE

Copy for people already using something else and wondering whether to move.

Claude / ChatGPTcompetitorslandingmigration
You do product marketing. Write a page for people switching from a competitor.

Your product: {what it is}
Competitor: {who}
Why you are better for their users specifically: {real differences}
Where they are better: {honestly}
What blocks a move: {data, habit, contract}

Write:
1. A headline with no swipes at the competitor. The reader uses it — criticism reads as calling their choice stupid.
2. Three differences that matter to their users. Not a feature list but the things their community complains about.
3. An admission of where they are better. Without it the page is not credible: people know the product they use.
4. What happens to data in a move: can it transfer, how long, who does it.
5. The first 15 minutes after switching: what they see and what they get done.
6. Who should not switch. That filters out the people who would return unhappy and write a review.
7. A comparison table, if it is honest. A table where you have every tick reads as advertising and does not work.

Do not make price the only argument. Whoever moved for price will move away from you for the same reason.
Voir un exemple de résultat

Headline: "Moving from X: we transfer your data in a day." Differences: works on a phone, you pay for use rather than seats, search across the whole archive. Honestly: X has stronger integrations. If you live in their ecosystem, stay. Data: CSV export, import on our side, typically 4 hours, we do it. Do not switch: if you have 50+ people and configured roles — we do not have those yet.

Pour le SaaS

Error messages

FREE

What happened, what to do, and whose fault it is — three lines instead of a 500.

Claude / ChatGPTerrorscopyinterface
You are a UX writer. Rewrite the error messages.

Product: {what it is}
The errors: {how they read now}
Who uses it: {audience}

For each error:
1. What happened — with no technical words, if the reader is not technical.
2. What to do right now. One action. "Contact support" is not an action, it is a hand-off.
3. Whose fault it is. Own yours plainly: "this is on our side" defuses more than an apology.
4. What happened to the data they already entered. The most common fear is that it is gone.
5. Where to show it: at the field, as a top bar, as a dialog. Dialogs are for irreversible things only.
6. What goes in the log and what goes to the person. An error code helps support, not the reader — put it small at the bottom.
7. Three errors to remove rather than rewrite: the interface can prevent them.

Do not write "something went wrong". It is the most useless sentence in interfaces: it says nothing and sounds like an excuse.
Voir un exemple de résultat

Was: "Error 422". Now: "We could not accept that email — it looks like there is a stray space. Check it and send again." Our fault: "Could not save, this is on our side. Your text is still in the field, try again in a minute." Data: say so always, even when it is safe — otherwise people check themselves and lose it. Remove entirely: "file too large" — show the limit before the file picker.

How to build one that solves a job rather than filling space on the landing page.

Claudeaifeaturesproduct

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
PRO

Where to draw the line so it nudges an upgrade rather than reading as punishment.

Claude / ChatGPTplanslimitsmonetisation

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour le SaaS

A quickstart page

FREE

The page after which something actually works, in ten minutes.

Claude / ChatGPTdocumentationquickstartdevelopers
You are a technical writer. Write the quickstart page.

Product: {what it is}
For whom: {developer, analyst, ordinary user}
What they end up with: {the result}
What is needed first: {key, account, install}

Write:
1. What they will have at the end — the first line, before any preamble. They must be able to judge whether ten minutes is worth it.
2. Prerequisites, as a list. Not buried where they surface at step four.
3. The steps. Each is one action and one check: how to know it worked. A step without a check ends with someone continuing from a broken state.
4. Code or values that can be copied whole and run. Placeholders like your_key are the only thing they change.
5. What goes wrong for half of them: three common errors and what to do, on this page, not in a troubleshooting section.
6. What next — three links, no more.
7. What does not belong here: architecture, explanations of why things are built this way, a full parameter reference.

To test the page: hand it to someone who has never seen the product and watch in silence. Every place they stop is an edit.
Voir un exemple de résultat

First line: "In 10 minutes you will send your first request and get a prompt back." Step 2, check: "The response should be 200 with an id field. A 401 means the key was truncated — it is 40 characters." Common error: a test-environment key against the production URL. Everyone does it, including us. Next: three links — reference, limits, examples.

Fifteen minutes that explain churn better than any dashboard.

Claude / ChatGPTchurninterviewscustomers
You are a researcher. Prepare a call with a churned customer.

Product: {what it is}
Who left: {segment, how much they paid, how long they stayed}
What the data says: {their last actions}
What you want to learn: {the decision that depends on it}

Prepare:
1. The invitation: a three-line email, with no gift attached. A gift produces polite answers instead of honest ones.
2. A first question that is not "why did you leave". That gets a rehearsed answer. Start with what they did in the product last month.
3. Eight questions about facts rather than opinions: what they did, what replaced it, who decided, what the last straw was.
4. The alternative question: what they use now and what is better there. That answer is worth more than the rest of the call.
5. What not to ask: will you come back, what should we improve. Both produce polite untruths.
6. How to take criticism: do not defend, do not fix it live. One objection and they stop talking.
7. What to do after: what not to promise, and what to write in the thank-you.

Note separately: a churned customer owes you nothing. Fifteen minutes, and not one more without their consent.
Voir un exemple de résultat

Invitation: "You stopped using us in July. I am not selling anything — I want to understand what went wrong. 15 minutes?" First question: "Walk me through how you used it in your last month." Last straw: "What happened on the day you decided not to renew?" Do not ask: "what should we improve" — the answer will be about buttons, and buttons are not why they left.

The dull emails that cause most of the card disputes.

ChatGPT / Claudeemailbillingsubscriptions
You are an editor. Write the money emails.

Product and price: {what and how much}
How the subscription works: {period, auto-renewal}
What you send today: {as it is}

Write five emails:
1. Payment succeeded: what was bought, for how long, the amount, where the receipt is. Short but complete — people keep it.
2. Renewal coming: 3 to 5 days ahead, with the amount and date. This email prevents more card disputes than any terms page.
3. Renewal succeeded: same as the first, without congratulations.
4. Subscription ending and not renewing: what stops working and what remains.
5. Refund: confirmation, amount, when it lands, what happens to access.

For each: subject under 40 characters, body under 100 words, one button.

Rules: amount and date in the first two lines; a cancellation link in every money email; no upsells in a charge notification — it reads as trading on someone else's wallet.

The upcoming-charge email matters most. It gets skipped so as not to "scare people off", and then come the refunds and disputes.
Voir un exemple de résultat

Subject: "Renewing 14 September, $6". Body: "Your PRO subscription renews automatically on 14 September and we will charge $6 to card ending 4417. If you would rather not renew, cancel and access stays until the 14th." Button: "Manage subscription". Not included: the annual plan offer. Not in this email.

Pour le SaaS

A demo script

PRO

Twenty minutes of showing that leaves a decision rather than an impression.

Claude / ChatGPTsalesdemob2b

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour Threads

Profile topics

FREE

The topics on your profile decide which strangers get shown your posts.

ChatGPT / Claudeprofiletopicsreach
You do social strategy. Pick profile topics for Threads.

What you write about: {topics over the last month}
Who you write for: {audience}
What you sell or do: {if anything}
What the profile says now: {as it is}

Deliver:
1. Three to five topics you should be found under. Not what interests you, but what strangers will look for you under.
2. How each topic overlaps with what you actually post. A topic with no posts works against you: arrivals leave.
3. What to remove: the too broad ("marketing") and the too narrow that nobody searches.
4. How this fits the first line of the bio. Topic and bio must agree, or the profile reads as somebody else's.
5. Three opening posts per topic, so an arrival sees this place is about that.
6. When to check and what to look at: the share of non-followers in your reach.

Topics are not hashtags. They are a claim about which feed you belong in, and the claim is backed by posts, not words.
Voir un exemple de résultat

Topics: prompts for work, marketplaces, seller tools. Remove "AI" — too broad, you are two hundredth there. Bio first line: "I break down prompts that save sellers time." Matches the topics. Check in 3 weeks: share of impressions from non-followers. Was 40%, should rise.

The fastest growth on Threads is not your posts but your replies in your niche.

Claude / ChatGPTrepliesgrowthniche

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

How to take the format going round the feed and not look like a tourist.

ChatGPT / Claudetrendsformatfeed
You write on Threads. Adapt a trending format to your subject.

The format: {describe it — city lists, five jobs, unpopular opinion, whatever}
Your subject: {what you write about}
Audience: {who reads}
Your last month of posts: {examples}

Deliver:
1. What works mechanically in the format: enumeration, recognition, disagreement, nostalgia. Copy the mechanism, not the words.
2. Three versions for your subject. Each must stay yours: if the format would fit any account, it adds nothing to yours.
3. What to drop in the transfer: whatever was funny only in the original.
4. The first line of each version. The feed shows that and half of the second.
5. What to reply in the comments so the discussion continues. For list formats this is the main thing: people come to add their own.
6. When not to take the format: if your subject offers no recognition, an enumeration reads as a list.

Do not take a format after day seven. By then the feed is tired of it and your post reads as an echo.
Voir un exemple de résultat

Mechanism: recognition through enumeration. People add their own in the comments. Version: "Five prompts I stopped writing by hand." Enumeration plus a confession. Drop: the office joke — our audience works from home. First line: "Five prompts I no longer write by hand" — enough to open it.

An enumeration people finish and then add to in the comments.

ChatGPT / Claudelistformatpost
You write short. Build a list post.

Subject: {what about}
What you enumerate: {mistakes, tools, rules, observations}
How many items: {number}
Audience: {who reads}

Build:
1. A headline with a number and a promise. The number sets the length — the reader knows what they are getting into.
2. One line per item. Two sentences per item turns it into an article, and articles do not get read in a feed.
3. Order: the most surprising first, the most useful second. The last item is remembered, but they have to reach it.
4. One item worth arguing with. Without it there are no comments.
5. A last line inviting additions: "what would you add". A list without one ends rather than continues.
6. What to cut: items everyone has, and items you have not tried yourself.

Five to seven items. Ten does not get finished; three looks like a draft of something larger.
Voir un exemple de résultat

Headline: "6 prompts I deleted after a month of use." Argument item: "A prompt longer than one screen is always worse than a short one." People argue, which is the point. One line per item, no explanations — those go into the comment replies. Last line: "What did you delete?

PRO

A routine that survives months, instead of weekend sprints.

Claude / ChatGPTroutinehabitgrowth

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

When a picture helps, when it hurts, and what has to be on it.

ChatGPT / Claudeimagepostattention
You edit social. Build a post with an image.

What the post says: {the substance}
The image: {screenshot, photo, chart, cover}
Audience: {who reads}

Deliver:
1. Whether an image belongs here at all. On Threads text works on its own, and a spare image pulls attention off the first line — say plainly if it is better without.
2. What must be visible in the preview. Feed images are small: small type on them is never read.
3. How text and image connect: the text does not narrate the picture, it adds what is not in it.
4. Screenshots: what to crop and what to redact. Personal data, other people's names, spare interface.
5. Charts: one idea, axes labelled in words, one highlighted point. A chart with no conclusion in the caption goes unread.
6. The first line of the post — it still decides. An image will not save a weak first line, and a strong one works without an image.

Do not add a picture "so people notice". They will notice and scroll: the feed has too many images and too few reasons to stop.
Voir un exemple de résultat

Image belongs: showing the difference between two model outputs — in words it would take three paragraphs. Visible in preview: two text blocks side by side, the left one shorter. Nothing else is needed. The text adds: what exactly changed in the prompt. That is not in the picture. Redact: the client's name in the screenshot header.

How to say it so people argue with the idea rather than with you.

Claudeopiniondebatereputation

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Collect questions and answer them so it becomes a week of posts.

ChatGPT / Claudequestionsaudiencecontent
You write on Threads. Run a question round.

Subject: {what you cover}
Followers: {number}
What has been asked in comments: {examples}
How many you will answer: {number}

Deliver:
1. How to invite questions. Not "ask me anything" — a narrow frame produces more questions than a wide one. People do not know what to ask an open door.
2. What to do with few followers: where else to get questions — a month of comments, your inbox, search on your subject.
3. How to select: questions interesting to more than the asker. Answer the rest privately.
4. Answer format: one post per question, not a ten-part thread. Each answer lives its own life in the feed.
5. What a good answer looks like: the question on line one, the answer on line two, detail after.
6. What to do with an awkward question: answer plainly or say you will not — but do not ignore it in public.
7. How this becomes a week of posts, and what to do with questions that return every month.

A repeating question is not boredom, it is a signal: the answer exists nowhere. That answer belongs pinned.
Voir un exemple de résultat

Invite: "Ask me about prompts — what is not working" rather than "ask me anything". Select: "how do I know a prompt is bad" interests everyone. "Why will the PDF not open" gets a private reply. Format: the post opens with the question itself, verbatim. That way people who did not ask recognise it. Returns monthly: "where do I start". That needs a pinned post.

A circle that moves an account more than any twenty posts do.

Claude / ChatGPTcommunityrelationshipsgrowth

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

When a thought fits one screen, and when it genuinely needs unpacking.

ChatGPT / Claudeformatthreadlength
You are an editor. Decide whether a text is one post or a thread.

The text: {paste}
Subject: {what it is}
What the reader should take away: {one thought}

Work through:
1. How many thoughts are in it. One thought is one post, always. A thread caused by length rather than by count reads as padded.
2. If there are several: are they sequential. A thread works when point two makes no sense without point one. If the points are independent, that is a list in one post.
3. How to compress to one screen: what to cut. Usually the preamble, the repetitions, and explaining what was already clear.
4. If a thread is warranted: how to build the first post. It must work alone — most people read only that.
5. How to end each node so people continue.
6. The maximum number of nodes. After the fifth almost nobody follows, and a second thread the next day works better.
7. What a thread costs: discussion smears across the nodes and the comments stay empty.

The default is one post. A thread has to be justified, not chosen by habit.
Voir un exemple de résultat

One thought: "short prompts work better." So, one post. Cut: two paragraphs on how you got there. They do not fit, and the comments are where you tell it. Were it a thread: node one is the conclusion, not the preamble. Everyone reads that, half read the rest. Maximum: 4 nodes.

FREE

How to bring back what worked without looking like a repeat.

ChatGPT / Clauderepostingarchivecontent
You are an editor. Prepare a repost of old material.

Best posts: {paste several}
How long ago: {timeframe}
Followers then and now: {numbers}
What has changed since: {in the subject, for you}

Deliver:
1. Which posts are worth repeating: those whose idea has not aged while the audience grew. A post everyone saw is not worth repeating.
2. What to change first: a new opening line, always. The same run-up over the same text is recognised and scrolled past.
3. Three ways to repost: with a new occasion, with a correction, with an admission. The last works best: "a year ago I wrote this, now I think otherwise".
4. What to do with numbers and examples: outdated ones must be refreshed, or the repost reads as carelessness.
5. How long to wait. Under three months is too soon.
6. How many such posts to keep in reserve and how to pick them without trawling the archive by hand.
7. What not to do: repost verbatim, repost more than monthly, repost what caused a fight.

A good repost is not a copy but a continuation. It adds what you did not know a year ago.
Voir un exemple de résultat

Repost: the short-prompts piece — 900 people saw it, there are 4,000 followers now. New opening line: "I wrote this a year ago. A third of my view has changed since." Refresh: was "of 69 prompts", now "of 207". Do not repost: the pricing argument — a second run collects the same fight, angrier.

Two weeks of posts that end in sales — with no post saying buy this.

Claude / ChatGPTlaunchsalesplan

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour Threads

Reach dropped

FREE

What to check and what not to do in the week everything went quiet.

Claude / ChatGPTreachdeclinediagnosis
You analyse social. Diagnose a drop in reach.

Before and after: {numbers over two or three weeks}
What you posted those days: {topics and formats}
What you did differently: {if anything}
Followers: {number}

Check in order:
1. How many days the drop has lasted. Two days is not a drop, it is two days. Reacting before a week means fixing something that is not there.
2. What changed on your side: posted less often, stopped replying, changed subject, started adding links.
3. Outbound links: almost everywhere they suppress reach. Check whether the start coincides with their appearance.
4. Format: whether you drifted into one type of post. Feeds tire of sameness faster than of bad.
5. The calendar: holidays, summer, major events. They move numbers more than your work does.
6. What to do this week: one change, not five. With five you will not know which worked.
7. What not to do: delete posts, double your posting rate, buy anything, change subject.

A two or three week dip is normal, not a fault. The problem starts when it lasts a month with your work unchanged.
Voir un exemple de résultat

Nine days — worth looking at. Coincidence: from the 3rd every post carries a link to the site. There were no links before. Format: 8 list posts in a row. The feed is tired. One change this week: links move to the first comment. Watch for 7 days. Not doing: posting more often, deleting anything.

Find your angle before you are mistaken for three identical accounts.

Claudevoicepositioningdifference

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

8 août 2026

48 prompts

Block order and what goes in each — before you open Figma.

ChatGPT / Claudelandingstructureux
You are a product designer. Build the wireframe for a landing page.

Product: {what it is}
For whom: {audience}
Target action: {what the visitor should do}
Main objection: {what the person fears or doubts}

Deliver:
1. Block order top to bottom, each with its job in one line.
2. For the first screen: headline, subheadline, button label.
3. Where the main objection is answered and with what exactly.
4. Three blocks that are not needed in this case, and why.

No vague labels like "benefits block" — write what will actually be in it.
Voir un exemple de résultat

1. First screen — name the problem in the customer's own words. 2. How it works, 3 steps. 3. Proof: a number or a teardown. 4. The "too expensive" objection → compared with the cost of the mistake… Not needed: unnamed "trusted by" logos, an "about us" block, a testimonial slider.

Turns colours, spacing and type into named variables a developer can use.

Claude / ChatGPTdesign systemtokensfigma

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

The idea, the front-panel composition, and what the law requires on it.

ChatGPT / Midjourneypackagingprintbrand
You are a packaging designer. Propose a concept.

Product: {what is inside}
Shelf: {where it sells — marketplace, shop, showroom}
Neighbours on the shelf: {what competitors look like}
Print run: {roughly}

Deliver:
1. The idea in one line — what makes this pack stand out on the shelf.
2. Front-panel composition: what is large, what is small, where the eye goes.
3. Palette and material, with this print run in mind.
4. Mandatory elements: contents, weight, barcode, expiry — where to place them.
5. What most often ruins packaging for this kind of product, and how to avoid it.
Voir un exemple de résultat

Idea: white space where every competitor screams with busy print — on the shelf it reads as a pause. Flavour large, brand small: on a marketplace people decide from the thumbnail, not the logo…

PRO

What moves, for how many milliseconds and on what curve — precise enough to rebuild.

Claude / ChatGPTmotioninterfacebrief

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Contrast, tap sizes, focus order — the things most often missed.

Claude / ChatGPTaccessibilitycontrastreview
You are an interface accessibility specialist. Review a screen from its description.

Screen: {what is on it}
Colours: {text and background in pairs, HEX}
Sizes: {font sizes, button sizes}
What is interactive: {list them}

Check and deliver as a list:
1. Contrast of every text-on-background pair: compute the ratio and say whether it passes 4.5:1 for body text and 3:1 for large text.
2. Tap target sizes: name anything below 44×44 points.
3. Keyboard tab order: is it logical, where are the traps.
4. Where meaning is carried by colour alone, and what to add alongside it.
5. What needs a label for a screen reader.

For each point say exactly what to change, not "improve contrast".
Voir un exemple de résultat

#8A8F9C on #14151A → 4.8:1, passes. #6B6E78 on #14151A → 3.1:1, fails at 14px, raise to #8A8F9C. The "More" button is 32×32 — too small, bring it to 44 with padding without enlarging the icon…

How to explain decisions so feedback lands on the goal, not on taste.

ChatGPT / Claudeclientpresentationnegotiation

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
Pour designers

Type scale and grid

FREE

Sizes, leading and columns derived from one step instead of picked by eye.

ChatGPT / Claudetypographygridlayout
You are a typographer. Build a scale and grid for a project.

What we are setting: {website / app / document}
Base body size: {e.g. 16px}
Fonts: {display and text}
Content width: {e.g. 1120px}

Deliver:
1. A size scale from the base — the ratio and the resulting values for h1–h4, body, caption.
2. Leading for each level, plus the rule: the larger the size, the tighter the lines.
3. The grid: column count, gutter, margins — separately for wide screens and for 360px.
4. Line length in characters for body text and how to hold it.
5. Three combinations that will turn to mush in this particular font pairing.
Voir un exemple de résultat

Ratio 1.25: 16 · 20 · 25 · 31 · 39 · 49 h1 49/1.05 · h2 31/1.15 · body 16/1.6 · caption 13/1.5 Line of 60–75 characters → max-width 34rem…

PRO

What to fix, what not to touch, and how to tell good work from bad.

ChatGPT / Clauderetouchingphotobrief

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

What to give away so people leave a contact — and later buy.

ChatGPT / Claudelead magnetfunnelemail
You are a marketer who builds funnels. Propose five lead magnets.

Product: {what you sell}
Audience: {who buys}
Main pain: {what they come with}
Paid product: {what we lead to}

For each option:
1. A name that promises a result, not "a useful guide".
2. Format: checklist / template / teardown / mini-course / calculator.
3. Which pain it closes and in how many minutes.
4. How it connects to the paid product: what the person realises after using it.
5. How long it takes to make.

The options must differ in depth: one for five minutes, one for an evening, one that requires effort.
Voir un exemple de résultat

"Moving budget in 10 minutes" — a spreadsheet with formulas. Pain: they cannot size the budget. Using it, they hit the contractor question — which is the paid service. Not "A guide to renovation": the promise is vague, they download it and never open it.

A minute-by-minute plan: where to hold attention and where to name the price.

Claude / ChatGPTwebinarsalesscript

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

A result story with numbers and an honest account of what failed.

ChatGPT / Claudecase studycontenttrust
You are an editor who writes case studies. Build one from the facts.

Client: {who, the name can be withheld}
Before: {situation and numbers}
What was done: {list the steps}
After: {numbers and over what period}
What did not work: {honestly}

Structure:
1. A headline with the result and the timeframe, no superlatives.
2. The starting point: why it was that way, not "the client did no marketing".
3. What was done, step by step, with the decisions that had to be made.
4. Numbers: before, after, over what period, and what else influenced them.
5. What did not work and what would be done differently.
6. Who this case does not apply to.

Points 5 and 6 are mandatory: a case study without a single failure reads as an ad and does not work.
Voir un exemple de résultat

"From 4 to 11 enquiries a week in two months" — not "explosive growth". Did not work: a blast to the old list, 6% open rate. We dropped contacts older than a year… Does not apply if your enquiry flow is limited by production capacity rather than by advertising.

Pour marketeurs

Pricing page copy

PRO

Plan names, feature order, and answers to the question of why it costs that.

Claude / ChatGPTpricingcopywritingconversion

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Options of different kinds, each with its preheader.

ChatGPT / Claudeemailheadlinesnewsletter
You are an email marketer. Write subject lines.

What the email is about: {content in two sentences}
To whom: {segment and what they know about you}
Action: {what the reader should do}

Give 20 subjects in four groups of five: a question, a number, intrigue without deceit, plain usefulness.

For each:
1. The subject itself — under 45 characters so it is not cut off on a phone.
2. A preheader that continues the subject rather than repeating it.
3. A note on who this subject suits less well.

Separately: three subjects that will certainly land in spam, and exactly what is wrong with them.
Voir un exemple de résultat

Question: "Why aren't your emails opened?" · preheader: "It isn't the subject — check this" Spam: "URGENT! Today only −70%" — capitals, exclamation mark, urgency with no reason…

Contacts split by behaviour, with a message for every segment.

Claude / ChatGPTcrmsegmentationemail

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

The news in the first paragraph, facts instead of adjectives, a usable quote.

ChatGPT / Claudeprpress releasemedia
You are a news editor, not a press office. Write the release so it gets picked up.

Event: {what happened}
Why it matters outside the company: {who the news is for}
Facts and numbers: {list them}
Who can give a quote: {job title}
Date: {when}

Structure:
1. A headline stating what happened, with no "innovative" and no "unique".
2. First paragraph: what, who, when, where, why it matters. Written so it could be printed whole and be enough.
3. Three paragraphs of detail in descending importance: editors cut from the end.
4. One quote that carries content, not "we are delighted to present".
5. A three-sentence note about the company.
6. A press contact.

Turn every judgement into a fact: not "significantly faster" but "40% faster, measured this way".
Voir un exemple de résultat

Headline: "Catalogue opened without registration" — not "A revolution in the world of prompts". Quote: "Registration was cutting off half of them — we decided browsing should not need it," said…

PRO

How much per channel, what to expect, and when to switch off what fails.

Claude / ChatGPTbudgetchannelsplanning

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

What to say and show before the viewer scrolls past.

ChatGPT / Claudeunboxingreelsscript
You are a short-video writer. Write an unboxing script.

Product: {what is being unboxed}
Who films: {your persona and tone}
What must be remembered: {one property of the product}
Platform: {Reels / TikTok / Shorts}

Deliver a table by seconds: time, what is in frame, what is said, what appears as on-screen text.

Rules:
1. The first two seconds are not the box but the result or an unexpected shot.
2. The opening itself takes no more than five seconds: it has been filmed a thousand times before you.
3. Reveal one property in detail, mention the rest.
4. The ending is not "hit like" but a reason to have watched to the end.
5. Three alternative opening lines to choose from.
Voir un exemple de résultat

0–2 s: the product already in hand, a shot of the result. Text: "In 40 seconds you will see why I keep it" 2–7 s: opening it, sped up, no commentary… Ending: "One thing inside I almost threw away" — then show it.

How to show a result people believe instead of calling it editing.

ChatGPT / Claudebefore-afterresulttrust
You are a results-video author. Write a before-and-after script.

What changes: {object or situation}
Timeframe: {over how long}
What actually contributed: {list it honestly}
Footage available: {what shots exist}

Deliver:
1. Shot order with timings.
2. How to show the "before" so it does not look deliberately worsened: same light, same angle, same time of day.
3. Where to state the timeframe, and why it cannot go at the end.
4. What to say about factors other than the product: without this the video reads as an ad.
5. Three comments people will definitely write, and the answers.
6. What not to show, so as not to promise the impossible.
Voir un exemple de résultat

Same angle, same time of day — otherwise the first comment is "the lighting is different". Timeframe in the first five seconds: without it viewers assume it took a week… The comment "what else did you do" — answer it directly, it removes half the doubt.

Text timed to the shots, with pauses and stresses marked.

Claude / ChatGPTvoice-overeditingscript

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

What to shoot, in what order, and what to bring so nothing is reshot.

ChatGPT / Claudefilmingshot listproduction

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

A short message that gets answered — with numbers and one concrete idea.

ChatGPT / Claudecollaborationoutreachbrand
You are a creator writing to brands. Compose the message.

Your channel: {topic, platform, reach, audience}
Brand: {who they are and what they sell}
What you noticed about them: {something specific — a new product, a weak spot, a video that worked}
What you propose: {format}

Requirements:
1. No longer than eight lines — long emails go unread.
2. The first line is about them, not you: what you noticed.
3. One concrete video idea, not "I propose a collaboration".
4. Three numbers about you — the ones the brand cares about, not follower count.
5. One clear next step.
6. No "mutually beneficial", and no rate card in the first message.

Also give a follow-up version for when there is no reply after a week.
Voir un exemple de résultat

"I saw you released a starter set, but the reviews keep asking where to begin" — the first line. Idea: a video on "the first three days with the set", from someone who genuinely cannot do it yet… Numbers: 62% completion, 4.1% saves, audience 24–34, large cities.

Packages, what is included, how many revisions, and what counts as extra work.

Claude / ChatGPTratestermscontract

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Turning existing footage into posts, stories and text without filming again.

Claude / ChatGPTrepurposingcontent planefficiency

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

A list of search terms by volume, and where each one belongs.

ChatGPT / Claudeseolistingkeywords
You are a marketplace listing specialist. Build the keyword set for a product page.

Product: {what it is, in plain words}
What makes it different: {material, size, purpose}
Platform: {Amazon / Etsy / other}
What buyers call it: {if you know}

Deliver:
1. Thirty search terms in three groups: high volume, mid, long tail with qualifiers.
2. Where each goes: five in the title, ten in the attributes, the rest in the description.
3. Synonyms and everyday names — how people search, not how the manufacturer names it.
4. Five terms the product does not match: those buyers arrive and return it.
5. How to check demand before rewriting the listing.

Do not stuff the title: the platform ranks, but the buyer reads — the second matters more.
Voir un exemple de résultat

High volume: laundry organiser, storage box… Title: "Laundry organiser with lid, 6 compartments, grey" Does not match: "shoe organiser" — they come for something else and send it back.

Which shots are needed, in what order, and what must be visible in each.

ChatGPT / Claudephotographyphotobrief
You are an art director of product photography. Write the brief.

Product: {what we shoot}
Platform: {format requirements, if known}
What the buyer fears: {the main doubt}
Shoot budget: {roughly}

Deliver:
1. The shot list in order: the first one decides the click from search results, the rest in descending importance.
2. For each: shot size, background, what is in frame besides the product, what must be visible.
3. The shot that removes the main doubt: show what the reviews keep asking about.
4. A scale shot: next to a familiar object, not "dimensions in the description".
5. Technical requirements: aspect ratio, resolution, background, margins.
6. Three mistakes that get a listing rejected or beaten by the one next to it.

The first shot decides everything: it earns the click; the rest are seen only by those already interested.
Voir un exemple de résultat

1. Whole product, white background, no shadows near the edges — this is the search-results shot. 3. An "in hand" shot — removes the "is it too small" doubt… Mistake: a four-photo collage as the first image — in search results it turns to mush.

PRO

A message about price, lead times and defects — with answers ready for a refusal.

Claude / ChatGPTsuppliernegotiationprocurement

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
PRO

What price to set, when to cut it, and how not to lose money on promotions.

Claude / ChatGPTpricediscountsmargin

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

Image-and-text blocks people actually read to the end.

ChatGPT / Clauderich contentdescriptionconversion
You are a product-listing editor. Build the rich content.

Product: {what it is}
For whom: {the buyer and their situation}
Main doubts: {list them from reviews, or guess}
Available material: {photos, video, diagrams}

Deliver five to seven blocks; for each:
1. A headline that states something, not "Benefits".
2. Text no longer than three lines.
3. What image goes with it and why that one.
4. Which doubt this block removes.

Plus:
5. Block order: doubts first, pleasantries later.
6. What to drop if the listing is mostly viewed on phones.
7. Three phrases to strike out: everyone uses them and they mean nothing.

Strike out: "high quality", "the perfect gift", "premium materials".
Voir un exemple de résultat

Block 1: "Does not slip on tile" + a photo on a wet floor. Removes the main doubt. Block 4: "Fits in a washing machine" + a shot inside the drum… Strike out: "made from high-quality materials" — everybody writes that.

Pour marketplaces

Seasonal and sale plan

PRO

What to buy, when to raise the price, and what stock not to carry over.

Claude / ChatGPTseasonsalepurchasing

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Why buyers pick the other one — point by point, fixable today.

ChatGPT / Claudeauditcompetitorslisting
You are a listing analyst. Compare two listings and explain the difference.

Our listing: {title, price, review count, what is in the photo}
Competitor listing: {the same}
Search positions: {where we are, where they are}

Deliver:
1. What the buyer sees first in search results, ours and theirs — and what earns the click.
2. A point-by-point comparison: title, main photo, price, reviews, description. For each: who wins and why.
3. What we can change today at no cost.
4. What costs money, and roughly how much.
5. What not to change even though it differs: not every difference matters.
6. The single point that gives the most if only one thing gets done.

Do not put "lower the price" first: it is the most expensive and most reversible move there is.
Voir un exemple de résultat

In search results: their photo is on white, ours on colour — ours disappears into the grid. They have 340 reviews to our 28 — cannot be caught up, but we can answer every one of ours… One action: reshoot the main photo. Cheapest, biggest effect.

A 40-second video that works muted and ends in a purchase.

Claude / ChatGPTvideolistingscript

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
FREE

Ready answers for common tickets — human, not robotic.

ChatGPT / Claudesupporttemplatescustomers
You are a support lead. Write reply templates.

Product: {what it is}
Common tickets: {list five to seven}
Tone: {how you talk to customers}
What you can and cannot do: {boundaries — refunds, access, timelines}

For each ticket deliver:
1. A reply under six lines, the substance in the first one.
2. What to ask if information is missing — one question, not a list.
3. A version for when it is our fault: without "we apologise for any inconvenience caused".
4. A version for when we cannot help: what to offer instead.
5. A note on when the template must not be used and a real reply is needed.

Every template must leave room for one live sentence: a message assembled entirely from boilerplate reads as an autoresponder.
Voir un exemple de résultat

"The login email never arrives": "Check your spam folder — about one in twenty lands there. If it is not there either, tell me which address you are using and I will look at the logs." When it is our fault: "This is our mistake, I can see it in the logs. Fixing it, I will come back to you today."

The list of things that break silently: access, keys, database permissions.

Claude / ChatGPTsecurityreleasechecklist

Ce prompt fait partie de la sélection PRO

Débloquer l'accès
PRO

Step order, the point of no return, and how to know everything arrived.

Claude / ChatGPTmigrationdatabaserelease

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

A page an outside developer can connect from without emailing support.

Claude / ChatGPTdocumentationapiintegration
You are a technical writer. Write an integration page.

What is being connected: {feature or API}
Who reads it: {reader's level}
What they need beforehand: {keys, access}
Common failures: {what breaks for people}

Structure:
1. What the result will be — one paragraph and one example response.
2. What is needed before starting: a list of three or four items.
3. A minimal working example: exactly enough code to see the first response, and not one line more.
4. The response field by field: what each means and what values occur.
5. Errors: code, meaning, what to do. As a table.
6. Limits: request rate, sizes, key lifetimes.
7. What to do when it does not work: three checks in order.

The first example must run by copy-paste: if anything has to be filled in first, the reader goes to support instead.
Voir un exemple de résultat

Result: a list of orders for a period, as JSON. The example response sits right under the paragraph. Error 401 — wrong or expired key. 429 — more than 60 requests a minute, wait… Not working: 1) check the key, 2) check the URL, 3) read the whole response body.

FREE

A sequence that leads to first value instead of nagging about payment.

ChatGPT / Claudetrialemailonboarding
You are an onboarding specialist. Write the trial email sequence.

Product: {what it is}
Trial length: {days}
First-value moment: {what the person must do to see the point}
Where people get stuck: {list it}
What happens at the end: {charge / lockout / downgrade}

For each email:
1. The day it goes out and the reason — an event, not "day three of the trial".
2. A subject under 45 characters.
3. Body under eight lines, one action at the end.
4. Who must not receive it: anyone who already did that action.

Required:
5. An email for those who never came back after signing up.
6. An email two days before the end: exactly what the person loses, concretely.
7. An email after it ends for those who did not pay — with no pressure.

The sequence leads to first value, not to payment: someone who never got there will not pay however often you remind them.
Voir un exemple de résultat

Day 0, right after sign-up: "Your first result in 4 minutes" — one link, one action. Day 5, only for those who never returned: "Looks like you did not get to it. What got in the way?" — a question, not a reminder… Two days out: "Your 13 saved prompts stay; access to the PRO collection closes."

PRO

What we build, what we do not, and how we will know it worked.

Claude / ChatGPTspecificationproductdevelopment

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

What broke, why it was possible, and what to change so it does not repeat.

Claude / ChatGPTincidentreliabilityreview
You are the engineer running an incident review. Write it up.

What happened: {describe it}
When it was noticed and how: {who saw it first}
How long it lasted: {duration}
Who was affected: {how many people, what did not work}
How it was fixed: {what was done}

Deliver:
1. A minute-by-minute timeline: when it started, when it was noticed, when it was fixed.
2. What exactly broke — the technical cause in one paragraph.
3. Why it was possible: not "a developer made a mistake" but what was missing in the system's design that let the mistake reach people.
4. Why it was not noticed immediately — and how to find out sooner next time.
5. Three actions: what to do today, this week, this quarter.
6. What not to do, even though it is the first thing that comes to mind.

Not a single name: a review that hunts for someone to blame will not happen next time — people stop reporting problems.
Voir un exemple de résultat

Broke: the variable was unset and the secret check silently let everyone through. Why possible: the code read "if the secret is set, check it". The default on failure was open… Today: close it. This week: a test for the empty variable. This quarter: audit every access check for the same shape.

Pour le SaaS

Quarterly roadmap

PRO

What we build, in what order, and what gets cut if we fall behind.

Claude / ChatGPTroadmapplanningproduct

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

A personal episode broken into a chain people read to the end.

ChatGPT / Claudestorythreadengagement
You are a Threads author. Break a story into a chain of posts.

What happened: {the episode in your own words}
How it ended: {the outcome}
What follows from it for the reader: {the value}
How many posts: {usually 4–7}

Deliver:
1. The first post is a scene, not an introduction: a specific moment, time and place. No "let me tell you a story".
2. One turn per post, each ending with a reason to read on.
3. The moment it went wrong — without it the story does not work.
4. The last post is a conclusion that applies to the reader, not a moral about yourself.
5. Three options for the first post.
6. What to cut if the story runs longer than six posts.

First-post test: if it could sit on top of someone else's story, it is weak.
Voir un exemple de résultat

1: "At 23:40 I deleted a folder I had spent six months building. There was no backup." 2: "At first it did not register. It did an hour later, when I opened the project"… Last: "You do not back up when you have time — you back up when you create the folder."

A question that is easy to answer and interesting to read in the replies.

ChatGPT / Claudequestioncommentsreach
You are a Threads author. Come up with question posts.

Account topic: {what you write about}
Audience: {who reads}
What you want to learn: {if you have a goal}

Give ten questions in three groups:
1. Experience questions — people answer because they have something to tell.
2. Either-or questions — answered in one word, low barrier.
3. Confession questions — answered because everyone has done it.

For each:
4. The question itself, up to two lines.
5. Your own answer as the first reply — without it, answering feels awkward.
6. A note on what answers you will get and what to do with them afterwards.

Separately: three questions nobody will answer, and why — too general, too personal, or too much work.
Voir un exemple de résultat

Experience: "Which tool did you abandon after a week, and why?" Either-or: "Do you plan your content or post by mood?" Nobody answers: "What do you think about the future of AI?" — too broad, there is nothing to say.

A long text broken into posts without retelling it or losing the point.

Claude / ChatGPTrepurposingthreadcontent

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Your own results, framed so people read for the lesson instead of rolling their eyes.

Claude / ChatGPTnumbersresultstrust

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

What to write so people visit your profile instead of scrolling past.

ChatGPT / Claudecommentsgrowthstrategy
You are a Threads growth strategist. Explain how to comment.

Your topic: {what you are about}
Whom to comment under: {types of accounts}
Your experience: {what you can usefully add}

Deliver:
1. Five kinds of comment that work: an addition, a counter-example, careful disagreement, a question of substance, a short story.
2. For each, a two-to-three-line sample on your topic.
3. What does not work: empty agreement, emoji, "great post", self-promotion.
4. How to write disagreement so it reads as conversation rather than argument.
5. How many comments a day make sense and why more is worse.
6. How to tell in two weeks that it is working: two signs.

A comment must be useful to the readers of the post, not to its author: only then do people follow it to your profile.
Voir un exemple de résultat

Addition: "This works with a warm audience. I tried it cold — here is what happened: …" Does not work: "Totally agree!" — gives the reader nothing, nobody follows that to a profile… Sign: people start reaching your profile from other people's posts, not only from search.

Seven posts tied by a single line, so people wait for the next.

Claude / ChatGPTseriesplanretention

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

What a month of numbers says, and which three posts are worth repeating.

Claude / ChatGPTanalyticsanalysisplanning

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

3 août 2026

8 prompts

The first line decides everything: on a phone only about 90 characters show before «more». Hooks are written to that limit.

ChatGPT / Claudehooksthreadsreach
Write 20 opening lines for Threads posts on the topic "{topic}".

Audience: {who}
My role: {who you are to them — practitioner, seller, observer}

Hard limits:
— No line longer than 90 characters. That is what shows in the feed before someone taps "more"; anything past it goes unread if the first part didn't land.
— No "Did you know that", "Top 5 ways", "Here's a roundup" — these read as templates and get scrolled past.
— No caps lock, no emoji chains.

Give 4 groups of 5:
1. Personal experience with a specific number or timeframe.
2. Disagreement with received wisdom in the niche.
3. An observation the reader recognises in themselves.
4. An open question they have something to answer from their own experience.

After each line, in brackets, say what reply it is built for. If the only possible reply is "agreed" — rewrite it: that hook collects likes, not conversation.
Voir un exemple de résultat

Personal experience: "Posted on Threads daily for six months. First four months: about 30 views each." (expecting: "so what changed in month five?") Disagreement: "Posting at the best time is advice that will do nothing for you. Here's why." (expecting: pushback from people who believe in timing)

Replies are the strongest signal in the recommendation feed. This post is assembled so people want to write, not silently like.

ChatGPT / Claudethreadsengagementrecommendations
Write a Threads post on the topic "{topic}" designed to pull replies.

What I know about it: {your experience, facts, numbers}
Audience: {who}
What I want to hear in the replies: {e.g. how other people solve this}

Rules:
— No longer than 500 characters, that is the platform limit. 300 is better.
— The first 90 characters must work as a standalone hook; the rest is what people unfold themselves.
— End on a question the reader already has an answer to from their own experience. Not "what do you think?" but something specific: their case, their number, their choice.
— Leave a place in the text where the reader can disagree. A post everyone agrees with gives nobody anything to say.

Don't do this:
— "Tell me in the comments!" — a direct ask works worse than a question people simply have something to say to.
— Don't invent provocation for the sake of an argument. A fight collects replies once, and then people unfollow.

Give 3 versions of the post and, under each, what kind of reply it is most likely to collect.
Voir un exemple de résultat

Version 1 (298 characters): "I let go of the strongest person on the team. He was right more often than anyone — and that was the problem. He didn't argue, he explained. After a few of those explanations people stopped offering their own ideas — why bother if you'll turn out wrong anyway. One strong voice can quietly switch off everyone else. Has this happened to you, and how did you handle it?" Will collect: stories about people's own "too-right" colleagues, and an argument about whether letting them go was right.

The feed watches speed: 20 replies in half an hour push a post further than 50 over a day. A plan for that window.

ChatGPT / Claudethreadsalgorithmfirst hour

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

A post worth arguing with on the merits. Checked separately so it doesn't turn into cheap provocation.

ChatGPT / Claudethreadsopiniondiscussion
Help me put a strong opinion into a Threads post.

Topic: {what about}
What I actually think: {your position, a rough draft is fine}
What it's based on: {experience, numbers, cases}

I want a post some readers will disagree with — and want to explain why. But not at the cost of my reputation.

Do this:
1. State my position in one sentence under 90 characters, so it is clear exactly what there is to argue with.
2. Assemble a post under 500 characters: the position, one piece of grounding from my own experience, and an admission of where I am wrong. That last part is mandatory — an opinion with no stated limits reads as posturing.
3. Write the three strongest objections that will come, and an honest answer to each.
4. Check the resulting post against this list and give a verdict on every point:
   — Would I repeat this under my own name a year from now?
   — Is this an argument about the work, or a swipe at people?
   — Do I have grounds beyond irritation?
   — If someone else had written this, would I reply on the merits or scroll past?

If even one point fails, say so plainly and suggest how to reframe it.
Voir un exemple de résultat

Position (84 characters): "AI courses sell a skill that goes stale faster than the course takes to finish." Check: — Repeat in a year: yes, the field is only speeding up. — About the work: yes, this is about the product, not the people. — Grounds: two courses taken, both outdated by the end. — Would reply myself: yes, there's a case to make about fundamentals.

One breakdown, article or video becomes seven standalone posts — not by slicing it, but by taking different angles.

ChatGPT / Claudethreadscontent planrepurposing

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

From a recommended post people land on your profile and decide in seconds. A 150-character bio and a pinned post built for that decision.

ChatGPT / Claudethreadsprofilefollowers
Put together my Threads profile: bio and pinned post.

What I do: {occupation}
What I'll be writing about: {2–3 topics}
Who for: {audience}
How I differ from others in this niche: {experience, numbers, point of view}
What someone gets by following: {specifically}

People arrive at the profile from the recommendation feed and decide in a couple of seconds. The bio is 150 characters, and that is all you get.

Give me:
1. Five bio options under 150 characters. Each must make clear what the profile is about and why you specifically. No "coffee and travel lover", no list of credentials.
2. For each option — who it suits and who it doesn't.
3. A pinned post under 500 characters: who you are, what will be here, and one example of your usefulness inside the text itself — not a promise, but something useful right away.
4. Three topics for the first posts that will make good on the bio's promise. If the bio promises breakdowns and the first posts are about breakfast, people unfollow.

Separately, tell me if my answers don't show any difference from a hundred similar profiles — and what to put up front in that case.
Voir un exemple de résultat

Bio (129 characters): "4 years selling on marketplaces, 6 stores now. I write about honest unit economics — including my own million-dollar mistakes." Who it suits: people already making sales. A beginner won't see themselves — for them, option 3 works better.

An honest walk through the funnel: where exactly it broke — the first line, the unfold, or the reply.

ChatGPT / Claudethreadsanalyticsbreakdown

Ce prompt fait partie de la sélection PRO

Débloquer l'accès

Un mot quand de nouveaux prompts arrivent ?

Le catalogue s'enrichit par lots de vingt ou trente. Laissez votre e-mail et je vous écris au prochain.

Uniquement sur les nouveaux prompts, deux fois par mois au maximum. Désabonnement en un clic depuis n'importe quel e-mail. Politique de confidentialité

Ajouté récemment — PrompTom