Brief and Prompt Library

info

This content was developed with AI assistance and is regularly reviewed for accuracy.

Snapshot: October 2026. Model names, prices, and product features change monthly; What's New in AI is the maintained chronology.

A good request to an AI assistant in 2026 reads less like a magic phrase and more like the note you would leave a capable new colleague: what done looks like, what they need to know, where the material is, what to avoid, what to hand back, and how you will check it. This library starts with that six-field brief, then gives ready-to-adapt briefs and short prompts by task. Everything here works in ChatGPT, Claude, Gemini, and Copilot.

How to use this page:

• Copy the brief template once and keep it somewhere you can paste from.
• Find your task below, replace every [bracketed variable] with specifics, and attach the files the brief mentions.
• Use the short prompts for quick jobs and the full briefs for anything you will send, publish, or act on.
• When you use the same brief every week, move it into a Project, custom instructions, or a skill so you stop retyping it.

The six-field brief

This is the template every full brief on this page follows. Brief: How to Ask AI for What You Want explains each field in depth.

Objective: [what done looks like, and why it matters]
Context: [who I am, who this is for, the situation]
Resources: [attached files / project / connector / URLs; which one is authoritative]
Constraints: [length, tone, out of scope, do not change, approval needed before...]
Output: [format; one example if format matters]
Verification: [how to check; what to cite; what to show me]

Field	What it prevents	Example
Objective	Polished work aimed at the wrong goal	"A one-page summary the board can decide from on Thursday"
Context	Generic output pitched at nobody	"I run a 12-person landscaping firm; readers are existing clients"
Resources	Invented facts and stale numbers	"Use the attached price list as the source of record"
Constraints	Scope creep and unwanted changes	"Under 200 words; do not change the refund policy"
Output	Reformatting the answer yourself	"A table: item, owner, due date"
Verification	Errors you only find after sending	"List every number with the line it came from"

Not every task needs all six fields. A quick rewording needs an objective and an output. Anything you would hand to a new hire with a day's notice needs all six. Notice what the template leaves out: no persona ("you are an expert..."), no "think step by step", and no formatting tricks. Current assistants reason by default; what they lack is your information.

From brief to reusable skill

If you send the same brief more than a few times, save it. Paste the stable parts (context, constraints, output format) into a Project's instructions or your custom instructions, and keep only the variable parts in each request. For repeatable multi-step work, the open Agent Skills format packages instructions and reference files that Claude and other agents load when relevant. Build Your Brief and Skill Library walks through it, and Skills, AGENTS.md, and CLAUDE.md covers the format.

Quick prompts for everyday jobs

Short prompts are fine when the stakes are low and you can see at a glance whether the answer is right. Each of these is one sentence plus whatever you attach.

Rewrite the attached email so it is clear, direct, and polite. Keep it under [number] words and keep my sign-off.

Expand these bullet points into one paragraph for [audience], keeping every point and adding nothing new: [bullets]

Summarize the attached [document] into 3 takeaways for [audience], each with the sentence it came from.

Explain [concept] to [audience] in plain language, with one everyday analogy and one common misunderstanding to avoid.

Rewrite this [technical/legal] text in plain English a typical [customer/reader] can follow, without changing what it commits us to: [text]

Organize these [number] topics into a logical learning sequence for beginners, and say why each comes where it does: [list]

Give me 10 [blog post/workshop/campaign] ideas about [topic] for [audience], each with a one-line angle.

Writing and communication

Professional email

Use for meeting requests, follow-ups, declines, and apologies. With a Gmail or Outlook connector on, point the assistant at the thread instead of pasting it.

Objective: Write a [meeting request / follow-up / polite decline / apology] email to [recipient and role] about [subject], so that they [desired action].
Context: Our relationship: [new prospect / long-term client / manager]. Background they need: [one or two sentences].
Resources: The earlier thread is [attached / in my inbox under "subject"].
Constraints: Under [number] words, [formal/warm] tone, no promises beyond [limit]. Sign off as [name, title].
Output: Two versions, one shorter, each with a subject line.
Verification: Flag any sentence that commits me to a date, price, or deliverable.

Article or blog post

Adapted from the marketing career guide; it writes for both readers and AI answer engines.

Objective: Draft a [word count]-word article answering "[question your readers ask]" for [audience].
Context: [Brand or publication]; readers are [description]. The goal after reading is [what they should do or know].
Resources: Attached: our voice guide, [two or three] past articles readers rated highly, and research notes with sources.
Constraints: Every statistic must come from the attached notes with its source and date. No invented quotes or customer names. Lead with the direct answer, use descriptive subheads, and define terms once.
Output: The article in markdown, a 2-sentence summary, three FAQ pairs, and a claims table (claim, source, date).
Verification: Mark any sentence you could not source with [UNSOURCED].

Repurpose content

Objective: Turn the attached [blog post / talk transcript / report] into [a 5-minute video script / 5 LinkedIn posts / a newsletter section].
Context: Audience on [channel] is [description]; they prefer [short, practical / data-led / story-led] content.
Resources: The attached piece is the only source.
Constraints: Keep claims identical to the source. [Each post under 150 words / script under 750 words]. No hashtags beyond [number].
Output: [Format], with a suggested title or hook for each piece.
Verification: List any point from the original you left out and why.

Localize a message

Objective: Adapt the attached marketing message for a [country/region] audience so it reads as written locally, not translated.
Context: Product: [description]. Audience: [description]. Channel: [email / web / print].
Constraints: Keep the offer, price, and legal terms unchanged. Note any idiom, image, or claim that may not work in this market rather than silently dropping it.
Output: The adapted text, then a short list of cultural or business-etiquette changes you made and why.
Verification: Recommend which lines a native-speaking reviewer should check first.

Social media calendar

Objective: Plan [1 week / 1 month] of posts for [brand] on [platforms] around [theme].
Context: Brand voice: [description]. Audience: [description]. Goal: [awareness / sign-ups / event attendance].
Resources: Attached: our recent best-performing posts and the campaign brief.
Constraints: [Number] posts per platform per week; platform-appropriate length; mix of educational, promotional, and engagement posts.
Output: A table: date, platform, post copy, content type, and suggested visual.
Verification: Flag any post that makes a claim not supported by the campaign brief.

Meetings, updates, and reports

Meeting summary and action items

Meeting notetakers can do this automatically; this brief is for when you have raw notes or a transcript.

Objective: Turn the attached meeting notes into a summary people can act on today.
Context: Meeting: [name, date]. Attendees and roles: [list].
Resources: Attached notes or transcript are the only source.
Constraints: Do not invent owners or deadlines. Keep the summary under 200 words.
Output: (1) a 3-sentence summary, (2) decisions made, (3) action items in a table with owner and due date, (4) open questions.
Verification: Mark any action item with no clear owner or date as [UNASSIGNED].

Weekly team update

Good candidate for a scheduled task once it works.

Objective: Draft this week's team update covering progress, problems, and next steps.
Context: Team: [name]; readers are [manager / wider department]. Last week's update is attached for format and tone.
Resources: [Project tracker export / notes / connected Slack channel].
Constraints: Under 250 words. Name problems plainly; no blame.
Output: Three short sections: Done, Blocked or at risk, Next week.
Verification: Link or cite the source item for each line.

Quarterly business review

Objective: Draft the [quarter] business review for [team] so leadership can decide on [decision].
Context: Audience: [who]. They saw last quarter's review (attached).
Resources: Attached KPI export with targets and prior-period values, plus my notes on wins and setbacks.
Constraints: Under 800 words. Use only figures from the attachments. [Professional / conversational] tone.
Output: Executive summary, performance highlights, challenges and lessons, next-quarter priorities, resource needs.
Verification: List every number with the row or line it came from.

KPI narrative

Objective: Write a short narrative for the [team] dashboard for [period].
Resources: KPIs with current value, target, and previous value: [list or attached file].
Constraints: Under 500 words. For each KPI: on track, at risk, or off track; the main driver in one sentence; any action needed.
Output: KPI-by-KPI notes, then an overall assessment and the one thing leadership should know.
Verification: Recalculate each change versus the previous period and flag any that do not match the data.

Research and analysis

Research with sources

Use research or deep-research mode where your assistant has one.

Objective: Find out [question] well enough to [decision it informs].
Context: I already know [what you know]. I care most about [recency / authority / a specific region].
Resources: Search the web. Prefer [primary sources, official statistics, peer-reviewed work]. Attached: [anything you already have].
Constraints: Only include claims you can link. Separate established findings from disputed ones. Note publication dates.
Output: A one-paragraph answer, then a table: claim, source, date, confidence.
Verification: For any statistic, find the original source rather than a page that repeats it, and say if you could not.

Check a statistic or claim

I found this claim: "[claim]" at [where you saw it]. Find the original source, quote the exact figure and its definition, give the date and sample, and tell me whether the claim as stated is accurate, misleading, or unsupported.

Competitive analysis

Objective: Compare [our product] with [competitor 1], [competitor 2], and [competitor 3] in [market segment] for [a sales battlecard / an executive summary].
Resources: Search the web; attached are our positioning doc and pricing page.
Constraints: Quote competitors' own pages for features and prices, with dates. Write "not found" rather than guessing.
Output: A comparison table (pricing model, key features, target customer, main differentiator, source links), then 200 words on our advantages and the biggest threat.
Verification: Flag any cell older than six months.

Data interpretation

Attach the file; current assistants run code to analyze it.

Objective: Tell me what changed in [metric] over [period] and what we should do about it.
Context: The data measures [what]; the business question is [question].
Resources: Attached [spreadsheet/CSV].
Constraints: Show your calculations. Separate what the data shows from possible explanations.
Output: The 3 most significant patterns, any anomalies, likely explanations labelled as hypotheses, and 2-3 recommended actions ranked by impact. Include one chart.
Verification: State any rows you excluded and why, and the totals you checked against.

Survey or customer feedback analysis

Objective: Summarize [number] [survey responses / reviews / feedback entries] into themes we can act on.
Context: Source: [survey name or channel]. Segments to compare: [list].
Resources: Attached export.
Constraints: Count, do not estimate. Keep customer quotes verbatim and anonymized.
Output: Sentiment breakdown with counts, top 5 themes by frequency with 2 quotes each, the most urgent issue, one positive highlight, and 3 recommended actions.
Verification: Give the count behind every percentage.

Audit how AI answer engines describe you

Objective: Audit how AI answer engines describe [our product] against [competitor].
Context: Our positioning doc and pricing page are attached.
Constraints: Use web search. Quote what you find rather than paraphrasing. Do not fill gaps with assumptions about us.
Output: A table of claims the engines make about us (claim, accurate?, source URL), the same for the competitor, then the five fixes to our site that would most improve accuracy and citability.

Strategy and planning

SWOT analysis

Objective: Produce a SWOT analysis of [company/product] for [decision, e.g. entering a new market].
Context: Situation: [one paragraph]. Key competitors or market forces: [list].
Resources: Attached [plan, financials, market notes]; use web search for competitor facts and cite them.
Constraints: 2-3 points per quadrant, each specific to us. Budget limit: [amount]. Horizon: [months].
Output: A 2x2 table, then the single strategic recommendation the analysis supports.
Verification: Mark which points rest on our documents and which on outside sources.

Scenario plan

Objective: Lay out scenarios for [decision or market condition] over [time horizon].
Context: What we control: [list]. What we do not: [list].
Constraints: Three to four scenarios that differ on the two most uncertain drivers. No probabilities unless you can justify them.
Output: For each scenario: description, early warning signs, and what we would do. Then the actions that make sense in every scenario.
Verification: State the assumptions behind each scenario explicitly.

Vendor or tool selection

Objective: Build selection criteria and a shortlist for [technology category] for a [size, industry] organization.
Context: Must-haves: [list]. Current systems it must work with: [list]. Budget: [range].
Resources: Search vendors' own pricing and security pages and cite them.
Constraints: Separate must-haves from nice-to-haves. Do not rank vendors on claims you could not verify.
Output: A weighted criteria table, a shortlist of [number] vendors scored against it, and the questions to ask each in a demo.
Verification: Date every price and note where it came from.

Change plan

Objective: Draft a change plan for [transformation, e.g. rolling out a new AI tool] in [team or organization].
Context: Who is affected: [groups]. Known concerns: [list]. Deadline: [date].
Constraints: Include training, communication, and a way to collect feedback. Keep it realistic for [team size].
Output: A phased timeline with owners, the key message for each audience, and three measures of success.
Verification: List the risks you think are most likely to derail the plan.

Coding and technical work

Fix an issue with a coding agent

Written for Claude Code, Codex, Cursor, or Copilot's agent; it assumes the repo has an AGENTS.md or CLAUDE.md file.

Objective: Fix [issue link or description] so that [observable behavior].
Context: Read AGENTS.md first. Relevant code is in [paths]. The failing case is [steps or test name].
Constraints: Do not change public APIs or the database schema. No new dependencies without asking. Keep the diff under [N] lines if possible.
Output: A short plan for my approval, then the change, a new or updated test that fails before and passes after, and a PR description.
Verification: Run [test command] and [lint command] and paste the results. List anything you could not verify.

Code review

Objective: Review [the attached diff / this pull request] before merge.
Context: The change is meant to [purpose]. Stack: [language, framework].
Constraints: Focus on correctness, security, and missing edge cases before style. Do not rewrite the whole file.
Output: Findings grouped as Critical, Warning, or Suggestion, each with the line, why it matters, and a minimal fix.
Verification: For each Critical finding, describe an input that would trigger it.

Bug investigation

Objective: Find the most likely cause of [bug] before I start changing code.
Context: Stack: [details]. Expected: [behavior]. Actual: [behavior]. Steps to reproduce: [steps].
Resources: Attached logs and the relevant files.
Constraints: Do not propose a fix until you have ranked hypotheses.
Output: The 3 most likely root causes ranked, a diagnostic step for each, and a safe temporary workaround if one exists.
Verification: Point to the log line or code that supports each hypothesis.

Architecture decision record

Objective: Write an architecture decision record for [decision].
Context: The problem driving it: [description]. Constraints: [technical, budget, timeline, team].
Resources: Options considered: [list]; attached design notes.
Output: Standard ADR format (Status, Context, Decision, Consequences), with pros, cons, and effort for each option.
Verification: List the assumptions that, if wrong, would change the decision.

Technical documentation

Objective: Document [component/API] for [junior developers / external API users].
Resources: Attached code or API spec; this is the only source of truth.
Constraints: Do not document behavior the code does not have. [Markdown] format.
Output: Overview, setup, reference with parameters and return values, two worked examples, troubleshooting.
Verification: Mark any example you could not run.

Teaching and training

Differentiated reading

From the education career guide.

Objective: Differentiate the attached article on [topic] for my [grade] class.
Context: Standards covered: [codes]. Three groups: approaching, on, and above grade level. Several students are multilingual learners.
Constraints: Keep every version factually identical to the source. Do not add facts, examples, or dates that are not in the article. Sentence length is the main lever; keep the key vocabulary in all three.
Output: Three versions with a shared 8-word vocabulary list, then 5 comprehension questions per version with an answer key.
Verification: For each question, quote the sentence in the source that answers it.

Feedback on student drafts

Objective: Draft feedback on the attached anonymized essays using the attached rubric.
Context: Assignment: [description]. Rubric has [number] criteria. These are first drafts; the goal is revision, not a grade.
Constraints: Do not assign scores. Do not rewrite student sentences. One specific strength and one specific next step per criterion, quoting the student's own words where possible.
Output: For each essay: a table (criterion, evidence from the essay, next step), then a two-sentence note to the student in an encouraging register.

Lesson plan

Objective: Plan a [length] lesson on [topic] for [learners and level].
Context: By the end, learners should be able to [2-4 measurable objectives]. Prior knowledge: [description].
Resources: Attached: [textbook pages, standards, materials available].
Constraints: Include a warm-up, direct instruction, hands-on practice, and a check for understanding. Fit within the time.
Output: A timed plan with materials and preparation steps.
Verification: Map each activity to the objective it serves.

Quiz or assessment

Objective: Write a [quiz/exam] on [topic] for [level].
Resources: Attached course material; questions must be answerable from it.
Constraints: [Number] multiple choice (4 options), [number] short answer, [number] scenario questions. Roughly 40% recall, 40% application, 20% analysis. No trick questions.
Output: The questions, then an answer key with one-sentence explanations.
Verification: Cite the page or section that supports each answer.

Workplace training curriculum

From the AI program manager guide.

Objective: Design a 3-module training curriculum on [skill or tool] for [role/department].
Context: Learners' current level: [description]. Time available: under 60 minutes per module.
Constraints: Each module focuses on one high-value use case and ends with a task the learner completes in the real tool on their own work, not a practice scenario.
Output: Module titles, objectives, activities, and the end-of-module task for each.
Verification: Explain how you would tell whether each module changed what people do.

Customer support

Customer reply

Objective: Reply to the customer message below so the issue is resolved or clearly on its way.
Context: Customer sentiment: [frustrated/confused/neutral]. What we can offer: [resolution]. Relevant policy: [policy text].
Constraints: Under 200 words. Acknowledge the experience, explain plainly, state the resolution and next steps. Do not promise anything outside the policy.
Output: The reply, ready to send.
Verification: Quote the policy line that supports each commitment.
Message: [paste]

Escalation handoff

Objective: Write an escalation summary a Tier 2 agent can act on without re-contacting the customer.
Resources: The ticket history is attached.
Constraints: Facts only; label anything inferred.
Output: Customer and account, issue, timeline of contacts, steps already taken, what the customer is asking for, recommended next action.

Knowledge-base gap analysis

From the customer service career guide; useful once an AI agent handles first-line support.

Objective: Find the knowledge-base gaps behind last week's failed AI conversations.
Context: Attached: [N] de-identified conversations the AI escalated or that the customer reopened within 7 days, plus our help-center export.
Constraints: Group by root cause (missing article, outdated article, policy the AI may not apply, product bug, other). Quote the customer's words for each example. Do not propose policy changes; flag them for a human.
Output: A table of root cause, count, example quote, and the specific article to write or fix, sorted by count. Then the top five fixes in priority order.

FAQ from tickets

Objective: Draft an FAQ for [product] from real customer questions.
Resources: Attached sample of [number] support tickets and the current help center.
Constraints: Write questions in the customer's own words. Answers of 3-5 sentences, consistent with the help center.
Output: FAQs grouped by category, plus 3-5 questions customers ask that the help center does not answer yet.
Verification: Mark any answer that contradicts or goes beyond current help-center content.

Professional domains: verify everything

These briefs come from the career guides for regulated work. They share one design rule: the AI may only use the sources you attach, and it must flag rather than fill gaps. Trackers reported roughly 2,000 court cases involving AI-fabricated citations by September 2026, so treat every output as a draft for a qualified professional to check.

Legal research memo

Objective: Draft a research memo on whether [legal question] in [jurisdiction].
Context: Facts are in the attached de-identified summary. Use only authorities you can link to in [research platform] or that I have attached.
Constraints: Do not cite any case, statute, or rule you cannot quote from. If authority is thin or split, say so rather than filling the gap. Distinguish binding from persuasive authority.
Output: A memo (question presented, short answer, analysis, open issues), then a table: authority, pinpoint cite, proposition it supports, direct quote, link.
Verification: Flag every row where you are not certain the quote and pinpoint are exact.

Contract review against a playbook

Objective: Review the attached [contract type] from [counterparty] against our playbook.
Context: Attached: our playbook with preferred, fallback, and walk-away positions for each clause.
Constraints: Only propose language that appears in the playbook or is a minimal edit to the counterparty's text. Do not opine on enforceability in [jurisdiction] without saying it needs attorney review.
Output: A table of clause, counterparty position, our position, deviation level (none, minor, major), and proposed redline. Then a one-paragraph summary for the client.

Variance commentary

Objective: Draft variance commentary for [period] management accounts.
Context: Attached: trial balance, budget, prior-year same period, and last quarter's commentary for tone. Materiality threshold: [amount or %].
Constraints: Explain only variances above the threshold. Every explanation must cite the accounts and, where possible, the specific transactions that drive it. If the data does not explain a variance, write [NEEDS OWNER INPUT] instead of guessing.
Output: A table (line item, actual, budget, variance, explanation, source rows), then a five-bullet summary for the CFO.
Verification: Recalculate every variance from the attached data and flag any that do not tie.

Patient discharge instructions

Objective: Produce discharge instructions for a patient recovering from [procedure].
Context: Use the attached approved template and the attached medication list. Reading level: 6th grade. Language: [language].
Constraints: Do not add any medication, dose, or warning that is not in the attached sources. If a detail is missing, write [CLINICIAN TO CONFIRM] rather than guessing.
Output: One page, numbered steps, a "call us if" list, then a short table of medications with purpose and timing.
Verification: After drafting, list every factual claim and the attached source it came from.

Therapy progress note

Use only with de-identified input and tools your practice has approved for health information.

Objective: Structure a progress note from my session outline.
Context: Modality [CBT/DBT/EMDR]. Required elements: [list from your payer or practice]. Attached: my de-identified bullet outline from the session.
Constraints: Use only what is in my outline. Do not infer diagnoses, risk levels, or interventions I did not record. Flag any required element my outline does not cover.
Output: Note in [DAP/SOAP] format, under 300 words, followed by the list of missing elements.

Hiring kit

Objective: Build a hiring kit for [role] on the [team].
Context: Attached: the hiring manager's notes, our competency framework, and our inclusive-language guide.
Constraints: Requirements must be job-related and necessary; flag any that look like proxies (degree, years of experience, "culture fit"). No questions about protected characteristics. Plain language at an 8th-grade reading level.
Output: A job description, six structured interview questions each with a 1-4 scoring anchor, and a 60-minute work-sample task with a rubric.
Verification: List every requirement and the competency it maps to.

Design and creative work

Creative concepts

Objective: Generate [number] distinct concepts for [logo / campaign visual / packaging] for [brand].
Context: Brand values: [list]. Audience: [description]. Competitors' look to avoid: [description or attached examples].
Resources: Attached brand guidelines and mood references.
Constraints: Each concept must differ in idea, not just color. No imitation of a named living artist's style.
Output: For each concept: a one-line idea, the visual metaphor, a suggested palette and type pairing, and an image-generation prompt I can use to sketch it.

Image prompt

For image generators built into ChatGPT, Gemini, or dedicated tools.

[Subject and action], [setting], [composition: close-up / wide / overhead], [lighting], [style or medium], [color palette], [mood]. Aspect ratio [ratio]. Leave space [top/left] for text. No text in the image.

Check the tool's terms on commercial use, disclose AI-generated imagery where your industry expects it, and see Make Things: Images, Video, and Audio in 2026 for rights and disclosure.

Rolling out AI on a team

Starter library for a department

Objective: Create a starter brief library for [department].
Context: The team's most common tasks are [list]. Tools available: [assistant and plan].
Constraints: 10 briefs, grouped by task, each using the six fields (objective, context, resources, constraints, output, verification) with [bracketed] variables.
Output: For each brief: the task it serves, the brief, and a note on what good output looks like and how to fix a miss.

Adoption report

Objective: Design a quarterly AI adoption report for the leadership team.
Context: Attached: tool usage exports by department, our last pulse survey, and the three use cases we prioritized this quarter.
Constraints: Separate measured data from self-reported data and label each. No individual-level usage data. Note where adoption is low and give a likely reason from the survey, not a guess.
Output: One page: adoption by department with trend, two high- and two low-performing teams with likely explanations, and three recommended actions for the next 90 days.

Key takeaways

• Start from the six-field brief: objective, context, resources, constraints, output, verification. It replaces persona tricks and "think step by step."
• The Resources and Verification fields do the most to prevent errors: tell the assistant what is authoritative and how you will check its work.
• Use short prompts for low-stakes jobs and full briefs for anything you will send, publish, or act on.
• In regulated work, restrict the AI to your attached sources and make it flag gaps rather than fill them.
• Save briefs you reuse as Project instructions or skills so the context travels with them.

Next steps

• Continue to: Brief: How to Ask AI for What You Want for the reasoning behind each field
• Continue to: Build Your Brief and Skill Library to turn these into reusable skills
• Continue to: Check: Verify, Iterate, Restart for how to review what comes back
• Continue to: AI and Your Career in 2026 for briefs in the context of your profession