📚 Professional prompt library by profession: dev, SEO, marketing, support, e-commerce
Structured, annotated prompts ready to use in English, French and Arabic. Fill in a few fields, the tool assembles the full prompt, and every prompt explains why it works and the trap to avoid.
🔒 Everything is computed in your browser: nothing is sent to or stored on a server.
The library
💻 Development
Security-focused code review
Before merging code that handles user input, authentication or sensitive data.
You are an application security engineer reviewing [Language / framework] code before it goes to production. Context: [What the code does].
Review the code below and look only for real issues: injection (SQL, shell, templates), XSS, missing access control, secret handling, input validation, errors that leak information, dangerous dependencies or functions.
For each issue, give:
1. severity (critical, high, medium, low);
2. the affected line(s);
3. a concrete attack scenario in one or two sentences;
4. the fix, as code.
Order issues from most to least severe. Do not invent vulnerabilities to fill the list: if you are unsure, say "to verify" and explain what should be checked. If information is missing (framework, configuration), ask instead of assuming. Answer in English.
Code:
[Code to review]
💡 Why it works
It forces a per-issue format (severity, line, scenario, fix), which turns a vague review into a list of verifiable actions.
⚠️ The trap to avoid
Never paste keys, passwords or .env files into the code you submit.
💻 Development
Write tests that catch real bugs
When a critical function has no tests, or before changing it.
You are a senior developer writing tests with [Test framework]. Here is the code to test:
[Function or module to test]
Business rules it must respect:
[Expected business rules]
First, without writing code, list the cases to cover: happy path, edge cases (empty values, zero, maximum, unexpected formats), expected errors, and each business rule. Flag any rule the current code does not seem to respect.
Then write the tests: one test per behaviour, names that describe the expected behaviour, no dependency on the network or the system clock (use test doubles where needed). Do not modify the code under test. If a behaviour is ambiguous, ask rather than freezing an assumption into a test. Answer in English.
💡 Why it works
Listing cases before writing code avoids happy-path-only tests, and tying each test to a business rule makes them useful in review.
⚠️ The trap to avoid
A test that mirrors the implementation proves nothing: check that it fails when you deliberately break the rule.
💻 Development
Understand an error from its stack trace
When an error blocks you and the stack trace is long or obscure.
You are an experienced developer helping to debug. Environment: [Environment].
Here is the error and its full stack trace:
[Error message and full stack trace]
What I already tried:
[What you already tried]
1. Explain in two plain sentences what the error means.
2. Point to the line of the trace that refers to my code (not a library), and why.
3. List possible causes from most to least likely, each with a quick check (a command or a line of code to add) that confirms or rules it out.
4. Propose the fix for the most likely cause.
Do not suggest anything I already tried. Do not invent configuration options or functions: if you are not sure they exist in my version, say so. If you need a file to conclude, ask for it. Answer in English.
💡 Why it works
It asks for checks that rule causes out one by one: you move forward on evidence instead of trying fixes at random.
⚠️ The trap to avoid
A truncated trace leads to a wrong diagnosis: paste it whole, but remove tokens and private paths.
💻 Development
Careful refactoring or migration
When working code needs modernising without breaking anything.
You are a senior developer in charge of a regression-free refactor. Goal: [Goal of the change].
Constraints that must be respected:
[Constraints]
Current code:
[Current code]
Before writing code:
1. Describe the current observable behaviour (inputs, outputs, side effects, errors): this is what must not change.
2. List the regression risks and the tests that would cover them.
3. Propose a split into small steps, each shippable and verifiable on its own.
Only then, give the code for the first step with a clear diff. Do not change public names, data formats or error messages unless the goal requires it, and flag it explicitly if so. If a constraint cannot be met, say so instead of working around it. Answer in English.
💡 Why it works
Describing the behaviour to preserve first gives an objective success criterion, and small steps limit the damage if something goes wrong.
⚠️ The trap to avoid
A giant refactor in one answer cannot be reviewed: demand steps and test each one.
💻 Development
Commit message and pull request description
When committing or opening a pull request.
You are a rigorous developer writing a project's history. Reason for the change: [Why this change]. Convention to follow: [Commit convention].
Here is the diff:
[Diff or list of changes]
Produce:
1. A commit message: a title of at most 72 characters, in the imperative mood, saying what the change does; then a body explaining the why and any non-obvious choices.
2. A pull request description with these sections: Context, What changes, How to test (concrete steps), Risks and things to watch.
Describe only what is actually in the diff; do not invent tests run, results or tickets. If the diff mixes several topics, say so and suggest splitting it into several commits. Answer in English.
💡 Why it works
The why and the "How to test" section are what a history most often lacks, and what saves reviewers the most time.
⚠️ The trap to avoid
Proofread: a message claiming "tests pass" when nobody ran them is worse than an empty one.
🔎 SEO
SEO article brief that truly matches intent
Before writing (or commissioning) an article meant to rank on Google.
You are an editorial SEO strategist. Target query: "[Main query]". Audience: [Target audience].
My real experience on the topic, to be used as the differentiator:
[Your real experience on the topic]
Produce an article brief:
1. The dominant search intent (learn, compare, buy, solve) and what the reader should be able to do at the end.
2. The specific questions this audience asks, grouped by stage.
3. An H2/H3 outline where each section answers a question, with what it must contain.
4. What will make the article better than average: examples from my experience, a table, a checklist, data to verify.
5. A title (≤ 60 characters) and a meta description (≤ 155 characters).
6. Internal links and reliable external sources to plan for (page types, no invented URLs).
Do not invent any figure, search volume or statistic: write "to source" wherever data would help. Ask questions if my experience is not enough for a section. Answer in English.
💡 Why it works
It starts from intent and your own experience, two things generic content lacks, and it forbids invented figures.
⚠️ The trap to avoid
Do not ask an AI for search volumes: it does not know them. Get them from an SEO tool.
🔎 SEO
Title tags and meta descriptions in 3 languages
For a page published in French, English and Arabic.
You are an SEO copywriter, native in French, English and Modern Standard Arabic. Brand: [Brand].
Page content:
[Page content or summary]
Keywords per language:
[Keywords per language]
For each language (FR, EN, AR), propose 3 title + meta description pairs:
- title: at most 60 characters, keyword first, brand at the end if space allows;
- meta description: 140 to 155 characters, concrete benefit, call to action, no promise that cannot be kept.
Do not translate word for word: adapt to the vocabulary each audience actually uses (the Arabic must sound natural, not translated). Give the character count of each proposal. Do not state anything that is not in the page content (no "#1", invented figures or lead times). End with your recommendation for each language and why. Explanations in English.
💡 Why it works
It demands per-language adaptation and character counts: you avoid truncated titles and translations that ring false.
⚠️ The trap to avoid
Have the Arabic checked by a native speaker before publishing: one clumsy phrase is enough to put readers off.
🔎 SEO
Internal linking plan
When a site has dozens of pages and its key pages receive few links.
You are a technical SEO consultant. Here are the site's pages and their topics:
[List of pages (URL + topic)]
Pages to strengthen first: [Priority pages to boost].
1. Group the pages into topic clusters and name the pillar page of each.
2. Propose a link table: source page → target page → suggested anchor → placement in the text (intro, a specific section, conclusion). Anchors must be descriptive and varied, never "click here".
3. Flag orphan or isolated pages, and pages competing for the same topic (cannibalisation), with a recommendation for each case.
Use only the URLs provided; create none. Limit yourself to 3 to 5 added outgoing links per page to stay natural. If a page's topic is too vague to decide, ask. Answer in English.
💡 Why it works
The source → target → anchor → placement table translates directly into implementation tasks, and the cannibalisation check stops you competing with yourself.
⚠️ The trap to avoid
Dozens of identical links added at once look artificial: add them gradually and within the text.
🔎 SEO
SEO and conversion audit of a page
When a page gets visits but does not rank or convert.
You are an SEO and conversion consultant. Page: [URL and target query].
Page content:
[Page text (headings included)]
Audit it along 4 axes, quoting the relevant passage each time:
1. Intent match: does the page answer the query quickly and clearly? What is missing?
2. Structure: title, meta description, H1, heading hierarchy, paragraph length, mobile readability.
3. Credibility: proof (examples, sourced figures, identifiable author, reviews), reassurance before contact.
4. Conversion: clarity of the offer, call to action, unaddressed objections.
End with the 5 priority actions, ranked by expected impact and effort (low, medium, high), with rewritten text for the first two. Rely only on the content provided: you cannot see speed, backlinks or actual rankings, so do not comment on them. Answer in English.
💡 Why it works
It restricts the analysis to what the AI can actually see and ends on ranked actions with a rewritten example: you go straight to execution.
⚠️ The trap to avoid
Do not ask it why the page "lost rankings": without Search Console data it can only guess.
🔎 SEO
FAQ that AI answer engines can cite
So a page can be quoted in Google, ChatGPT or Perplexity answers.
You are a writer specialised in content that answer engines reuse. Topic: [Page topic].
Verified facts you may use, and only these:
[Verified facts to use]
Write a FAQ of 6 to 8 questions:
- questions are phrased the way a user would type or say them aloud;
- each answer starts with one sentence that answers directly (yes, no, a figure, a definition) and stands on its own out of context, then 1 to 3 sentences of detail;
- each answer is 40 to 80 words;
- no empty phrases ("it is important to note that").
If a useful question cannot be answered with the facts provided, do not invent: list it separately with the information I would need. Also say which answers would benefit from citing an external source. Answer in English.
💡 Why it works
The standalone first sentence is what engines extract; restricting answers to the facts you give stops an error from being repeated everywhere.
⚠️ The trap to avoid
Add FAQPage structured data only if the FAQ is visible on the page, otherwise Google may ignore it.
📣 Content and social media
3-month editorial calendar
To stop posting at random and tie every piece of content to a goal.
You are a content marketing manager. Business: [Business and offer]. Audience: [Audience]. Channels and realistic pace: [Channels and realistic pace]. 3-month goal: [3-month goal].
Build a 12-week editorial calendar:
1. 3 or 4 content pillars, each tied to an objection or a question of the audience.
2. A week-by-week table: channel, specific topic, format (carousel, article, short video, case study…), angle, call to action, pillar.
3. An explicit balance between content that attracts, content that proves expertise and content that converts.
4. Reusable content: how one article becomes 3 posts.
5. 3 metrics to track, tied to the goal (not just views).
Stick strictly to the pace given: fewer but sustainable is better. No invented holidays or events; if a seasonal date matters for this audience, flag it as "to verify". Answer in English.
💡 Why it works
The explicit pillar → objection → call-to-action link avoids content made "just to exist" and makes every post measurable.
⚠️ The trap to avoid
An over-ambitious calendar is dropped within three weeks: give your real pace, not the one you wish you had.
📣 Content and social media
One piece of content, three networks
To adapt an article or video for LinkedIn, Instagram and X without copy-pasting.
You are a social media manager. Audience: [Audience]. Desired action: [Desired action].
Source content:
[Source content]
Turn it into three native posts:
1. LinkedIn: a hook of at most 2 lines visible before "see more", a concrete story or lesson, short paragraphs, a closing question that invites comments, at most 3 hashtags.
2. Instagram: a 6-to-8-slide carousel outline (text of each slide, at most 20 words), then the caption.
3. X: a thread of 4 to 6 posts of at most 280 characters, the first must stand alone.
Each version needs its own angle, not the same summary. Keep only facts present in the source; add no figures, testimonials or promises. Give the character count of the X posts. Answer in English.
💡 Why it works
It applies each platform's real constraints (hook before "see more", slides, 280 characters) instead of one text pasted three times.
⚠️ The trap to avoid
Check each network's current rules (length, links): they change often.
📣 Content and social media
Prospecting email that does not look like spam
To contact a specific company after studying its situation.
You are a consultant who writes personalised prospecting emails, never bulk ones. Offer: [Your offer in one sentence].
What I know about the prospect:
[What you know about the prospect]
Proof I can cite: [Real proof].
Write 2 versions of a first email:
- subject line of at most 6 words, no shouting caps, no promise;
- at most 90 words;
- first sentence about the prospect (a specific observation from the information above), not about me;
- one value idea only, the proof, then a simple low-effort question (not "do you have 30 minutes?");
- direct, polite tone, no flattery, no fake urgency.
Then write a follow-up of at most 50 words to send 5 days later, which brings new information. Do not invent any fact about the prospect or any result: use only what I provided. Answer in English.
💡 Why it works
A first sentence about the prospect and a low-effort question are what separate an email that gets read from one that gets ignored.
⚠️ The trap to avoid
Follow local rules on prospecting (consent, unsubscribe): in Europe, GDPR also applies to B2B emails.
📣 Content and social media
A persona built on data, not clichés
Before a campaign, a site redesign or an offer launch.
You are a marketing researcher. Product or service: [Product or service].
Real data I have:
[Real data available]
From this data only:
1. Identify 2 or 3 distinct customer profiles. For each: situation, main goal, trigger problem that starts the search, choice criteria, objections, words they use (quote the verbatims provided).
2. For each profile, the core message that speaks to them and the proof they expect.
3. What the data cannot tell, and 5 precise questions to ask real customers to fill those gaps.
Do not invent ages, first names, hobbies or statistics: any trait not backed by the data must be marked "assumption". If the data is too thin, say so plainly. Answer in English.
💡 Why it works
It separates what is observed from what is assumed and turns gaps into interview questions: the persona becomes a decision tool, not a fiction.
⚠️ The trap to avoid
A "Sophie, 34, loves yoga" persona with no data behind it is useless: bring real customer quotes.
📣 Content and social media
Replying publicly to a negative review
When a negative review is posted on Google, a marketplace or social media.
You are in charge of customer relations for a small business. Here is a negative review posted publicly:
[Negative review (exact text)]
What actually happened:
[What actually happened]
What I can offer: [What you can offer].
Write a public reply:
- 60 to 120 words;
- thank without a canned phrase and precisely acknowledge the inconvenience experienced;
- if the customer is wrong on a point, correct it calmly with a fact, without accusing them or revealing personal information;
- offer a concrete solution and a private contact channel;
- written for future customers reading the exchange, not to win an argument.
Give 2 versions (one warmer, one more factual). Do not invent any goodwill gesture beyond what I stated. If the review contains a serious accusation (health, fraud), flag it and advise not to reply before checking. Answer in English.
💡 Why it works
It reminds you that the reply is for future readers, which changes the tone, and it limits promises to what you can keep.
⚠️ The trap to avoid
Never reply in the heat of the moment: reread it the next day before posting.
🎧 Customer service
Empathetic, precise customer reply
To reply to an unhappy or worried customer by email or chat.
You are a customer service agent. Brand tone: [Brand tone].
Customer message:
[Customer message]
Verified internal information:
[Internal information (order, status, rules)]
Write a reply that:
1. restates in one sentence the problem as the customer experiences it, to show they were understood;
2. gives the answer or concrete solution, with next steps: who does what and when;
3. answers every question asked, in order;
4. fits in at most 120 words, with no internal jargon and no repeated apologies.
Use only the internal information provided: promise no lead time, refund or goodwill gesture that is not in it. If information is missing, write the reply with a placeholder [TO COMPLETE: …] and tell me what needs checking. Answer in English.
💡 Why it works
Restating shows listening, dated next steps cut follow-ups, and the [TO COMPLETE] placeholder prevents invented promises.
⚠️ The trap to avoid
Do not paste unnecessary personal data (full address, phone): the order number is enough.
🎧 Customer service
Canned replies (macros) for support
To speed up answers to the questions that come back every day.
You are a customer service quality lead. Here are the most frequent questions, as customers write them:
[Frequent questions received]
Official rules and information:
[Official rules and information]
1. Group questions that share the same answer, and name each group.
2. For each group, write a canned reply: at most 80 words, simple sentences, with fields to personalise in brackets ([first name], [order number]) and a short version for instant messaging.
3. For each reply, say when NOT to use it and hand over to a human (special cases, anger, legal stakes).
Rely only on the rules provided. If a question has no answer in those rules, flag it as a gap to fill instead of inventing a policy. Answer in English.
💡 Why it works
The "do not use" cases prevent robotic replies at the wrong moment, and spotting gaps improves the rules themselves.
⚠️ The trap to avoid
Review macros every quarter: a changed return policy makes every reply wrong at once.
🎧 Customer service
Ticket triage and prioritisation
When the support queue overflows and urgent cases must go first.
You are a support team lead. Available categories: [Categories used].
Tickets to handle:
[Tickets to sort]
For each ticket, give in a table: ID, category, priority (P1 urgent: customer blocked, payment charged with no order, security; P2 important; P3 normal), sentiment (calm, worried, angry), a 10-word summary, recommended next action.
Then:
1. flag tickets that seem to concern the same incident (a sign of a wider problem);
2. list tickets that need a senior human (legal threat, personal data, very unhappy customer).
Do not guess: if a ticket is too vague to classify, mark it "to clarify" and suggest the question to ask the customer. Answer in English.
💡 Why it works
Spotting tickets tied to the same incident turns triage into an early warning: you fix the cause instead of answering 50 times.
⚠️ The trap to avoid
Anonymise before pasting: remove names, emails and phone numbers from tickets.
🎧 Customer service
Calming an angry customer
When a conversation turns sour and the customer threatens to leave or post a review.
You are a customer de-escalation specialist. Here is the ongoing conversation:
[Ongoing conversation]
My real room for manoeuvre:
[What you can and cannot do]
1. Analyse in 3 lines what raised the tension (broken expectation, feeling unheard, an automated reply…).
2. Write the next reply: acknowledge the emotion without judging it, take responsibility for what is our fault, offer 2 concrete options for the customer to choose from (choice restores control), and give a dated commitment I can keep.
3. List the phrasings to avoid in this specific situation.
No promise outside my limits. If the customer expresses a serious threat, harassment or a legal risk, recommend handing over to a manager instead of replying. Answer in English.
💡 Why it works
Offering a choice between two options gives control back to the customer, which calms more reliably than another apology.
⚠️ The trap to avoid
Do not use "I understand your frustration" on its own: without a concrete action behind it, it annoys even more.
🎧 Customer service
Knowledge base article
To turn an internal procedure into a help page customers understand on their own.
You are a technical writer for a help centre. Topic: [Article topic].
Raw material available:
[Raw material]
Write the article:
1. a title phrased as the customer's question;
2. a short answer at the top (2 sentences) for those in a hurry;
3. numbered steps, one action per step, with exact button names in quotes;
4. an "If it doesn't work" section with the 3 most likely problems and their fix;
5. when and how to contact support.
Write for a beginner on a phone: short sentences, no internal jargon. Reuse exactly the labels in the material provided; if a label or step is missing, write [TO VERIFY] instead of inventing it. Answer in English.
💡 Why it works
The short answer up top and the "If it doesn't work" section serve both hurried customers and blocking cases, which truly reduces tickets.
⚠️ The trap to avoid
One invented button name makes the whole article useless: check every label in the real interface.
🛒 E-commerce
A product page that sells without lying
For a new product page or one that converts poorly.
You are an e-commerce copywriter specialised in conversion. Product: [Product]. Target buyer: [Target buyer].
Verified specifications (the only allowed source):
[Verified specifications]
Write the page:
1. a clear title (product + key feature), at most 70 characters;
2. 3 to 5 benefit bullets, each tied to a real specification ("thanks to…");
3. a 120-to-180-word description answering what a buyer asks before paying: who it is for, how to use it, what they receive;
4. a "Good to know" block with the product's honest limits (what it does not do);
5. 4 questions and answers for the page's FAQ.
Forbidden: health claims, "best", "#1", certifications, customer reviews or figures not in the specifications provided. If important information is missing (size, care, warranty), list it at the end. Answer in English.
💡 Why it works
Tying every benefit to a real specification and showing the limits builds trust and reduces returns.
⚠️ The trap to avoid
For cosmetics and supplements, health claims are regulated by law: have them validated.
🛒 E-commerce
Cash-on-delivery order confirmation (WhatsApp)
To confirm a COD order before shipping and reduce refusals at delivery.
You handle cash-on-delivery order confirmations for the store [Store name], in Morocco.
Order details:
[Order details]
Rules: [Delivery and return rules].
Write a short, polite WhatsApp sequence:
1. confirmation message: a clear summary (product, quantity, exact amount to pay in dirhams, city, announced delivery time), then a closed question "Do you confirm?" with a simple answer (Yes / Change);
2. reply if the customer confirms: thanks, what happens next, a reminder to have the amount ready;
3. reply if the customer hesitates: reassure with the real rules (parcel check, returns), no pressure;
4. a single follow-up, 24 hours later, if there is no answer.
Give each message in English and in Moroccan Darija written in Arabic script. Messages of at most 50 words, no excessive emojis, no fake urgency ("last items"). Add no condition that is not in the rules. Explanations in English.
💡 Why it works
A summary with the exact amount plus a closed question gets fast confirmations, and restating the real inspection rules cuts refusals at the door.
⚠️ The trap to avoid
Only message customers on WhatsApp who gave their number for this order, and respect their refusal.
🛒 E-commerce
Abandoned cart recovery sequence
To win back part of abandoned carts without harassing.
You are an e-commerce retention specialist. Store: [Store and tone]. Typical cart: [Typical abandoned cart]. Allowed incentive: [Allowed incentive (or none)].
Known abandonment reasons:
[Known abandonment reasons]
Write a 3-email sequence:
1. after 1 hour: a simple, helpful reminder (the cart is saved, direct link, help if needed);
2. after 24 hours: address the most frequent objection among the known reasons (size, delivery, payment, returns);
3. after 72 hours: a last message, with the allowed incentive only if there is one.
For each: subject line (at most 45 characters), preview text, body of at most 80 words, a single button. No countdowns or fake limited stock. Do not invent any delivery or return policy: leave [TO COMPLETE] if information is missing. Add a short SMS version of message 2. Answer in English.
💡 Why it works
Each message addresses a real objection instead of repeating "you forgot something", which makes the sequence helpful rather than pushy.
⚠️ The trap to avoid
Only send these reminders to customers who agreed to receive your emails, with an unsubscribe link.
🛒 E-commerce
A clear, reassuring return policy
To write or rewrite a store's "Returns and refunds" page.
You are a writer specialised in readable legal pages for e-commerce. Selling countries: [Selling countries].
My actual conditions:
[Your actual conditions]
Write the "Returns and refunds" page:
1. a 4-point summary at the top (time limit, conditions, cost, refund time);
2. the step-by-step procedure, from first contact to refund;
3. exceptions, stated clearly with no fine print;
4. a 5-question FAQ;
5. a 2-sentence short version to show on product pages and in the cart.
Plain language, short sentences, "you". Reproduce my conditions exactly, without adding or softening any. Separately, as questions, flag the points where consumer protection rules in the countries listed might require something else (for example a legal withdrawal period), without asserting any specific rule: I will have them checked by a professional. Answer in English.
💡 Why it works
The summary at the top and the short version in the cart reassure at purchase time, where the policy really sways the decision.
⚠️ The trap to avoid
This text is not legal advice: have the final version validated under each country's law.
🛒 E-commerce
Analyse return and refusal reasons
When the return rate or cash-on-delivery refusal rate eats into margin.
You are an e-commerce analyst. Period and volume: [Period and total order volume].
Return and delivery refusal data:
[Returns or delivery refusals (reasons, products, cities)]
1. Using only this data, compute the breakdown of reasons, products and cities involved (counts and percentages, showing the calculation).
2. Identify the top 3 causes and, for each, where in the journey it originates (product page, order confirmation, packaging, carrier, lead time).
3. Propose 2 concrete actions per cause, stating the metric that will show whether the action works.
4. List the missing data that would make the analysis more reliable.
Do not fill in data: if a reason is empty or vague, count it as "not provided". Draw no conclusion about a product or city with fewer than 5 cases; flag it as "to watch". Answer in English.
💡 Why it works
It ties each cause to the journey step where it originates, turning a table of numbers into a list of measurable actions.
⚠️ The trap to avoid
Check the calculations in a spreadsheet: an AI can make mistakes when adding up long lists.
⚡ Productivity and management
Meeting notes → decisions and actions
Right after a meeting, so everyone knows who does what by when.
You are a rigorous executive assistant. Participants: [Participants and roles].
Raw meeting notes:
[Meeting notes or transcript]
Produce minutes in 4 parts:
1. Decisions made (one line each, unambiguous).
2. An action table: action, owner, deadline, any dependency.
3. Open questions and who must answer them.
4. A message of at most 80 words to send to participants to validate the minutes.
Strictly separate what was decided from what was only discussed. If an action has no owner or deadline in the notes, write "to be defined" instead of assigning one. Add no decision that is not in the notes. Answer in English.
💡 Why it works
The decided/discussed split and the "to be defined" markers expose the gaps that make projects fail.
⚠️ The trap to avoid
Do not paste a transcript containing confidential information without the participants' consent.
⚡ Productivity and management
A realistic project plan
At the start of a project, to find out whether the deadline is achievable.
You are an experienced, candid project manager. Project:
[Project and expected outcome]
Resources: [Available resources]. Deadline and constraints: [Deadline and constraints].
1. Break the project into phases, then tasks of 1 to 3 days at most.
2. For each task: owner (from the resources), low and high estimates, dependencies.
3. Identify the critical path and the top 5 risks, with a mitigation for each.
4. Give your honest verdict: is the deadline achievable? If not, say what to cut, postpone or add as a resource.
5. Define 3 verifiable milestones (what must be visible and testable at each).
Present estimates as ranges, not certainties. If information needed to estimate is missing (content volume, client approvals), ask. Answer in English.
💡 Why it works
Estimate ranges and a candid verdict on the deadline avoid the optimistic plan that slips by week two.
⚠️ The trap to avoid
Do not turn estimates into commitments to the client without adding a buffer.
⚡ Productivity and management
Difficult email (delay, refusal, bad news)
When you must announce a delay, turn down a request or reset a relationship in writing.
You are a professional communication adviser. Recipient: [Recipient and relationship]. Desired outcome: [Desired outcome].
Situation:
[Situation]
Write the email:
- the bad news in the first 2 sentences, no beating around the bush;
- a short, factual explanation, without blaming others;
- what I am concretely doing to limit the impact, with dates;
- a clear request to the recipient (decision, approval, call);
- at most 150 words, respectful and confident, neither servile nor defensive.
Give 2 versions (one more formal, one more direct), then list the sentences I might have been tempted to write that would have made things worse. Add no commitment or compensation I did not mention. Answer in English.
💡 Why it works
Breaking the news up front and ending on a clear request avoids the vague email that forces a second round.
⚠️ The trap to avoid
Reread it with a cool head: a difficult email sent in the heat of the moment is remembered for months.
⚡ Productivity and management
A reasoned decision (pros and cons)
Facing an important choice, to decide on criteria rather than gut feeling.
You are an adviser who helps make decisions in a structured, honest way. Decision: [Decision to make].
Options considered:
[Options considered]
What matters to me:
[What matters to you]
1. Restate my criteria and ask me to weight them if I have not (or propose a weighting I can correct).
2. Score each option on each criterion in a table, with a one-sentence justification per cell.
3. For each option, give the best argument for, the best argument against, and the scenario in which it would be the wrong choice.
4. Say what would be reversible or not, and the cost of going back.
5. Give your recommendation and the condition that would change it.
Clearly separate facts from assumptions. Do not quote prices, figures or features you are not certain of: mark them "to verify". Answer in English.
💡 Why it works
Looking for the scenario where each option would be a mistake, and the cost of reversing, guards against confirmation bias.
⚠️ The trap to avoid
Tool prices and features change often: check them on the official sites before deciding.
⚡ Productivity and management
A faithful summary of a long document
To quickly understand a contract, report or specification without losing what matters.
You are an analyst in charge of faithful summaries. I am reading this document to: [Why you are reading it].
Document:
[Document to summarise]
Produce:
1. The essentials in at most 5 lines.
2. The key points relevant to my goal, each with a reference to where it is in the document (section, clause or a short quote in quotation marks).
3. Obligations, deadlines, amounts and conditions, in a table.
4. Vague areas, contradictions or unusual clauses to have checked.
5. Questions to ask the author or the other party.
Summarise only what is written: no additions, no interpretation presented as fact. If part of the document seems missing or cut off, say so. For a legal document, remind me that your summary does not replace professional advice. Answer in English.
💡 Why it works
Requiring a reference for each point lets you check the summary in seconds instead of taking it on trust.
⚠️ The trap to avoid
Remove personal data and confidential information before pasting the document.
Prompts updated on 21 October 2026.
Why this tool
Structured, not "magic"
Role, context, constraints, output format and quality criteria: every prompt follows the structure that gets a usable answer on the first try.
Fields instead of brackets
Fill in the variables (your code, your product, your audience): the final prompt updates live, with no forgotten placeholder.
Honest about their limits
Every prompt asks the AI to ask questions when information is missing and never to invent figures or sources.
Frequently asked questions
Do these prompts work with ChatGPT, Claude and Gemini?
Yes. They rely on no assistant-specific feature: they are clear, structured instructions that recent models follow well. Results vary by model: always review the answer.
Can I include my customers' data?
Avoid personal data and secrets in a consumer assistant. Anonymise names, phone numbers and addresses, or use an offer whose terms you control. The Lab's security guide explains how.
Why are the prompts so long?
Every line prevents a common mistake: an off-topic answer, the wrong format, invented figures. A precise prompt saves you several back-and-forths.
Is the text I type stored?
No. The prompt is assembled in your browser; nothing is sent or stored. You then copy it into the assistant of your choice.