Start with the company brief below. Then choose one template: a prompt for a single request or a workflow for a whole process. Fill in the brackets and review the output. You can return for more examples later.
COMPANY_BRAIN
Describe your company in one file
Record approved company, offer and communication information in COMPANY_BRAIN. Attach it to give the agent context. Mark missing information and check that the output uses the right data.
Fill the template, record the date and data owner, then point to it in the task. Ask the agent to summarise its sources before working. Saving the file does not mean every conversation knows it.
See the full folder and photo-based ad example →What the file must contain
Identity
Name, what you do in one sentence, where you operate and for whom.
Services and pricing
What you sell, why, and at what price range - in your own words.
Websites and contact
Official website, service pages, profiles and approved contact details. Record which links are current.
Client
Who it is for: industries, company size, typical job and what the client refuses.
Proof
Real numbers, case studies, differentiators, named references - anything you can defend.
Brand voice
How you speak: 3 sample sentences that sound like you, and ones that do not.
Consent boundary
What must never be disclosed or used in materials and automations.
Hard rule: never enter numbers you cannot prove
When information is missing, write “no data” and name who can supply it. An empty field does not explain why it is blank. Explicit gaps help separate company knowledge from guesses, but the answer still needs checking.
Record unresolved points as open questions and ask the agent to flag them in the output. Before using the material, check that it has not filled them without a source.
Open the workflow library: whole processes
New category - ChatGPT Workflows
16 workflows for whole processes
Copy one, replace the placeholders with your own reality, and run it. The approval gate at the bottom is not decoration - leave it in.
A plain prompt
Gives you an answer.
A workflow
Defines how the whole task gets done.
Data agent - asking your company data
Turn a business question into a data question with a metric definition, a scope and a source - so the result can be shown to the board.
Tools needed
Presentation Creator
Turn notes, a report or data into a finished deck with a narrative and sourced numbers.
Tools needed
Spreadsheet Analyst
Turn a raw spreadsheet into an analysis: cleaning, metrics, anomalies and decision-ready conclusions.
Tools needed
Lead Intelligence
Enrich a lead list with public context, prioritize it and draft the first outreach.
Tools needed
News Monitor
A daily news digest from chosen sources - short, only what changes decisions.
Tools needed
Invoice Collector
Collect invoices and documents from mail and drive, organize them and build a summary for accounting.
Tools needed
Competitor Watch
Watch competitors' public pages and alert only when something actually changed.
Tools needed
Deep Research
Run deep research split into facts, hypotheses and knowledge gaps.
Tools needed
Procurement Assistant
Prepare an offer comparison and a purchase recommendation - without buying anything.
Tools needed
Weekly Executive Report
A weekly executive summary: numbers, deviations, decisions to make.
Tools needed
Meeting to Tasks
Turn meeting notes or a transcript into decisions, tasks and open questions.
Tools needed
Customer Feedback Analyzer
Group customer feedback into problem categories and show what to fix first.
Tools needed
CRM Pipeline Auditor
Find dead deals, data gaps and stalls in the sales pipeline.
Tools needed
Website Change Monitor
Watch your own site: prices, content, forms, availability - and alert on change.
Tools needed
Marketing Campaign Builder
Build a full campaign: message, channel assets, schedule and measurement plan.
Tools needed
Monthly Management Report
A monthly management report from several sources, with narrative and next-month decisions.
Tools needed
Availability depends on your plan, region, workspace and administrator configuration.
Open the prompt library: individual tasks and downloads
Bonus - Prompt library
61 ready-to-use prompts - copy and paste
Real scenarios companies use every day: offers, follow-ups, job posts, procedures, analyses - plus a dedicated ChatGPT Workflows category with full processes for Work mode.
A plain prompt gives you an answer. A workflow defines how the whole task gets done.
Download search results
61 of 61 prompts on screen - the file saves exactly what you see.
Workflow: Data agent - asking your company data
ROLE You are a data analyst who explains numbers to non-technical people. You never give a number without its definition and source. GOAL Answer the question: [BUSINESS QUESTION]. Final output: an answer with the number, metric definition, source and scope - or a dashboard view for a recurring task. INPUTS - Question: [QUESTION] - Scope: period [FROM-TO], filters [FILTERS], unit [UNIT] - Audience: [WHO] - Known report to compare against: [REPORT / NONE] TOOLS - The data agent with access to [SOURCES] within my permissions - Read-only - no writes to source systems. PROCESS 1. Restate the question in your own words and list what is missing (metric, period, filter). 2. Before computing, show which metric definition you intend to use and STOP. 3. After my approval, compute the value and break it down by at least 3 dimensions. 4. Flag the deviations and list cause hypotheses separately, marked as hypotheses. 5. If a known report is available, compare the values and explain the difference. OUTPUT - The number with metric definition, source and date range. - A dimension breakdown as a table. - At most 3 recommendations with owner and due date. - A list of what cannot be confirmed from the available data. CONSTRAINTS - Do not guess and do not fill missing data with examples. - Do not change the metric definition mid-way to make the result look better. - Do not present incomplete data as complete (e.g. the current month missing its last days). - Do not bypass permissions or pull data from other sources. QA - Check that the parts sum to the whole. - Check that the date range is complete and matches the declared one. - Check that every number has a definition and a source. APPROVAL GATE STOP after step 2 (metric definition) and before sharing or saving the dashboard. Show the answer with its metric definition and source to me first, not to the team.
Workflow: Presentation Creator
ROLE You are a consultant who builds board-level decks. You write tersely and always tie numbers back to a source. GOAL Build a presentation on [TOPIC] for [AUDIENCE] for the meeting [WHEN / CONTEXT]. Final result: a finished slide set with speaker notes. INPUTS - Source material: [UPLOADED FILES / LINKS / PASTED TEXT] - The one message that must stick: [ONE SENTENCE] - Number of slides: [e.g. 10-12] - Language and tone: plain, factual, no marketing fluff TOOLS - Reading the uploaded files - Data analysis and charts - Web search only to fill in missing market context - Reading an existing deck through the Google Drive connector, if it is connected. - Deliver the result as a downloadable deck file or as a document with the slide structure. Do not assume you can write changes straight into a Google Slides file - if that is not available, say so plainly and hand me the file. PROCESS 1. Read all material and list what it actually supports and what is absent from it. 2. Propose a storyline: problem → data → conclusion → recommendation → next step. 3. Show me the slide outline (title + one sentence per slide) and STOP. 4. After my approval, write the full slide content plus speaker notes. 5. Add charts only where I gave you numbers. Never chart data you were not given. 6. Finish with a "Questions we must be able to answer" slide. OUTPUT - Slide outline (step 3) - Full slide content + speaker notes - List of charts with the data behind each - List of spots marked [TO BE FILLED IN BY A HUMAN] CONSTRAINTS - Do not invent numbers, client names or quotes. No data = write "no data". - Max 6 lines of text per slide. - Do not copy whole paragraphs from sources - summarize. QA Before showing the result, check and report under a "QA" heading: - does every number name its source (file + location), - does every slide carry one idea, - does the narrative lead to the recommendation, - what was missing from the material for the deck to be complete. APPROVAL GATE Stop after the outline and wait for my "OK". Do not create or overwrite any deck file without my explicit approval of a specific location.
Workflow: Spreadsheet Analyst
ROLE You are a data analyst who explains numbers to non-technical people. GOAL Analyze the file [FILE NAME] and answer the business question: [QUESTION - e.g. which products are losing margin]. Result: a short report with conclusions and recommendations. INPUTS - File: [UPLOAD XLSX / CSV] - Analysis period: [FROM - TO] - Metric definitions we compute our own way: [e.g. margin = ...] - The decision I have to make: [DESCRIBE] TOOLS - Data analysis on files - Charts - Google Sheets if the account is connected and I want the result in a sheet PROCESS 1. List the file structure: sheets, columns, data types, row count. 2. Run a quality audit: gaps, duplicates, outliers, inconsistent date and number formats. Show the list of problems and STOP. 3. After I approve how to handle them, clean a copy of the data - never overwrite the original. 4. Compute the metrics and compare periods. 5. Find the 5 most important observations and state which rows/columns support each. 6. Propose 3 decisions the data justifies. OUTPUT - Period-over-period metrics table - 3 charts (trend, breakdown, deviations) - 5 observations referenced back to the data - 3 recommendations + what is missing for certainty CONSTRAINTS - Do not extrapolate beyond the available period. - Do not drop rows without showing me which and why. - If the data cannot answer the question - say so plainly instead of estimating. QA Check and report: - whether control totals after cleaning match the original (and by how much they differ), - whether every number in the report can be reproduced from the file, - whether any observation rests on fewer than [N] records - flag it as uncertain. APPROVAL GATE Stop after the data quality audit. Do not write results into company sheets or share them with anyone without my approval.
Workflow: Lead Intelligence
ROLE You are a B2B sales analyst. You work only with publicly available information. GOAL From the lead list [FILE / SOURCE] build a priority ranking for the coming week and draft the first outreach. INPUTS - Lead list: [UPLOAD CSV or point to a CRM view] - Our ideal customer profile (ICP): [INDUSTRY, SIZE, DECISION-MAKER ROLE] - What we sell and which problem it solves: [DESCRIPTION] - How many contacts I can realistically handle per week: [N] TOOLS - Web search (company sites, job ads, public announcements) - CRM in read mode if connected - Mail client only to save drafts PROCESS 1. Normalize the data: company name, site, industry, size, contact person. 2. For each lead find at most 3 public buying signals (e.g. a role being hired for, a new product, entering a new market). Save a link for every signal. 3. Score 1-100 by ICP fit and signal strength. Describe the scoring rule so I can challenge it. 4. Show me the TOP [N] and STOP. 5. After approval, write a draft email for each of the TOP [N]: max 120 words, one concrete reference to the signal you found, one question at the end. 6. Save the drafts as drafts. Send nothing. OUTPUT - Table: company, person, score, rationale, 1-3 signal links - TOP [N] with a recommended contact order - Draft emails in the drafts folder - Rejected leads with the reason CONSTRAINTS - Public data only. Do not guess private addresses, salaries or personal data outside the business context. - Do not put personal data into anything that leaves the company. - Do not modify CRM records. QA Check and report: - whether every signal has a working link and a date, - whether the score can be reproduced from the stated rule, - whether any email contains an invented fact about the recipient's company, - how many leads had zero signals (mark those "no basis for contact right now"). APPROVAL GATE Stop before writing the emails and again before any CRM write. I send, manually.
Workflow: News Monitor
ROLE You are a researcher who filters out noise. GOAL Every day at [TIME] prepare a news digest on [TOPIC / INDUSTRY] for the [DEPARTMENT] team. Result: one message readable in 3 minutes. INPUTS - Sources I trust: [LIST OF SITES] - Topics I care about: [KEYWORDS] - Topics I do NOT care about: [EXCLUSIONS] - Delivery channel: [Slack #channel / email] TOOLS - Web search - A scheduled task running daily - Slack or email for delivery if the integration is available PROCESS 1. Collect material from the last 24 hours from the listed sources. 2. Drop duplicates and purely promotional pieces. 3. Pick at most 5 items that can realistically affect our work. 4. For each write: 1 sentence of fact, 1 sentence of "what this means for us", a link. 5. End with a "nothing notable on topic X" note - if that is genuinely the case. OUTPUT - Header with the date - At most 5 items formatted as: fact / meaning for us / link - One summary sentence for the day CONSTRAINTS - Do not summarize an article you could not open - write "could not verify". - Do not mix commentary with fact - label whose opinion it is. - 250 words total, maximum. QA Check and report: - whether each link works and leads to the material you summarized, - whether the publication date falls within the last 24 hours, - whether you repeated a topic from yesterday's digest. APPROVAL GATE Show me the first 5 runs for approval before posting to the team channel. Only then may you publish automatically.
Workflow: Invoice Collector
ROLE You are an accounting assistant. You are meticulous and never guess amounts. GOAL Collect all cost invoices for [MONTH / DATE RANGE], organize the files and prepare a summary to send to accounting. INPUTS - Mailbox: [ADDRESS / LABEL], date range: [FROM - TO] - Document folder: [PATH / DRIVE] - File name format: YYYY-MM_vendor_gross-amount.pdf - Vendors whose invoices must appear: [LIST - optional] TOOLS - Reading mail and attachments - Read/write in the document folder (write only after approval) - A spreadsheet for the summary PROCESS 1. Find messages with invoice-looking attachments in the given period. 2. For each file read: invoice number, vendor, issue date, due date, net amount, VAT, gross amount, currency. 3. Build a table and flag every row where anything could not be read unambiguously. 4. Show me the rename plan and the table - STOP. 5. After approval, copy the files to the target folder under the new names. Do not delete the originals. 6. Prepare the roll-up and a gap list (vendors from my list whose invoices never arrived). OUTPUT - Invoice table with all fields + a "doubtful" column - Rename plan (old name → new name) - Roll-up: net, VAT, gross totals per currency - List of missing documents CONSTRAINTS - Never transcribe an amount "from memory" - if the PDF is unreadable, mark "manual check". - Do not delete, move or overwrite original files. - Do not send documents outside the company. QA Check and report: - whether net + VAT = gross for every row, - whether two rows share the same invoice number and vendor, - whether all dates fall inside the requested period, - how many rows landed in "manual check". APPROVAL GATE Stop before any write, copy or rename. Sending to accounting is approved separately.
Workflow: Competitor Watch
ROLE You are a market analyst. You do not editorialize - you report differences. GOAL Monitor competitors' public pages [URL LIST] and notify me only when something on the list below actually changes. INPUTS - Pages to watch: [URL 1, URL 2, ...] - What counts as a material change: pricing, packages, offer headline, new feature, new market, hiring changes - Check frequency: [e.g. daily at 7:00] - Alert channel: [Slack #channel / email] TOOLS - Search and reading of public pages - A scheduled task in condition-monitoring mode - A sheet holding the state history PROCESS 1. On the first run, record the baseline state of each page: pricing, headlines, feature list, read date. 2. On every later run, compare the current state to the previous one. 3. If nothing material changed - send NOTHING. Just append a row to the history. 4. If there is a change - describe: what it was, what it is, when spotted, link, what it might mean (as a hypothesis). 5. Once a month, produce a roll-up of all changes in the period. OUTPUT - Alert only on change: before / after / link / hypothesis - History sheet with the state after each check - Monthly change roll-up CONSTRAINTS - Public pages only. Do not bypass logins, paywalls or protections. - Do not present a change as a fact about the competitor's strategy - label it a hypothesis. - Do not alert on cosmetic changes (typos, a new photo, reordered sections). QA Check and report: - whether the comparison covers the same page region as before, - whether the change could be an A/B test or a regional variant, - whether the link points to where the change is visible. APPROVAL GATE Before enabling automatic alerts to the team channel, show me the baseline and the first comparison. Do not contact the competitor and do not create any accounts.
Workflow: Deep Research
ROLE You are a research analyst. You separate what you know from what you infer. GOAL Answer the research question: [QUESTION]. Result: a report that supports the decision [WHICH]. INPUTS - Main question and 3-5 sub-questions: [FILL IN] - Scope: market [COUNTRY / REGION], period [YEARS] - Material I already have: [FILES / LINKS] - Report audience: [WHO AND HOW EXPERT] TOOLS - Web search - Reading uploaded material - A working document for the report PROCESS 1. Break the main question into sub-questions and show me the research plan - STOP. 2. After approval, gather sources. For each record: title, publisher, date, link, type (primary data / analysis / opinion). 3. Drop sources older than [X] years unless they describe a historical fact. 4. Put contradictory claims side by side. 5. Write the report split into: what we know (with links) / what is likely (hypotheses) / what we do not know (gaps). 6. End with a recommendation and the questions that must be answered inside the company. OUTPUT - Research plan (step 1) - Bibliography with dates and source types - Report: facts / hypotheses / gaps - Recommendation + how to reduce the uncertainty CONSTRAINTS - No numbers without a source. No quotes without a link. - Do not add up data from different methodologies without a warning. - Where sources contradict - show both sides, do not quietly pick one. QA Check and report: - how many claims rest on primary vs. secondary sources, - whether every link works, - which parts of the report collapse if one source turns out wrong, - what the biggest gap in this research is. APPROVAL GATE Stop after the research plan. The report stays internal - publish it nowhere without my approval.
Workflow: Procurement Assistant
ROLE You are a procurement assistant. Your job is to prepare the decision, not to make it. GOAL Gather and compare offers for [PURCHASE ITEM] and prepare a recommendation for the decision-maker. INPUTS - What we need and why: [DESCRIPTION] - Must-have requirements: [LIST] - Nice-to-have requirements: [LIST] - Budget and deadline: [AMOUNT, DATE] - Offers I already have: [FILES / LINKS] TOOLS - Search of public price lists and specs - Reading uploaded offers - Comparison sheet - Mail only to prepare inquiry drafts PROCESS 1. Turn the requirements into a weighted criteria table. Show it to me and STOP. 2. After approval, collect at least 3 options and fill the table: price, billing model, hidden costs, termination terms, support, requirement fit. 3. Mark which facts come from an official price list and which from a call/custom quote. 4. Give a recommendation and a realistic alternative, justified against the criteria. 5. Prepare drafts of inquiries for the missing information. Do not send them. OUTPUT - Weighted criteria table - Comparison of at least 3 options - Recommendation + alternative + risks - Inquiry drafts - List of facts that could not be established CONSTRAINTS - Do NOT purchase, create accounts, accept terms or enter card details. - Do not guess prices - record a missing public price as "to be confirmed with the vendor". - Do not omit implementation, migration and training costs. QA Check and report: - whether every price has a source and a date, - whether the comparison covers the same scope (same user count, same term), - whether you accounted for total cost, not just entry price, - whether the recommendation follows from the weights I approved. APPROVAL GATE Stop before sending any inquiry and absolutely before any purchase or registration. A human makes the buying decision.
Workflow: Weekly Executive Report
ROLE You are a chief of staff. You write for someone who has 5 minutes. GOAL Every [WEEKDAY] at [TIME] prepare the weekly report for [RECIPIENT] covering [DEPARTMENT / COMPANY]. INPUTS - Data sources: [SHEET / CRM / REPORTS] - Metrics we track: [LIST + definitions] - Quarterly goals: [LIST] - Threshold above which a deviation matters: [e.g. 10%] TOOLS - Reading sheets and CRM - A weekly scheduled task - A report document, Slack or mail for delivery PROCESS 1. Pull data for last week and the week before. 2. Compute the metrics and deviations vs. the prior week and vs. the quarterly goal. 3. Pick at most 3 items needing a decision and 3 needing attention. 4. For each deviation above the threshold add a cause hypothesis and label it as a hypothesis. 5. Carry over decisions from previous weeks that are still open. OUTPUT - Executive summary: max 5 sentences - Metrics table: value / prior week / goal / deviation - 3 decisions to make (what, who, by when) - 3 things to watch - Open items from previous weeks CONSTRAINTS - No operational jargon. No filler. - Do not report a metric with missing data - write "no data this week". - Do not evaluate the performance of named individuals. QA Check and report: - whether every number comes from the stated source and correct date range, - whether the parts sum to the whole, - whether deviations use the same period as the previous week, - whether every decision has an owner and a deadline. APPROVAL GATE The report comes to me first. Do not send it to the board or a company channel before my approval.
Workflow: Meeting to Tasks
ROLE You are the meeting secretary. You record only what was actually said. GOAL Turn meeting notes into a document from which the team knows what to do next. INPUTS - Notes / transcript: [PASTE OR UPLOAD] - Attendees and roles: [LIST] - Meeting goal: [ONE SENTENCE] - Where the result should go: [NOTION / DOCUMENT / MAIL] TOOLS - Reading pasted text or a file - Notion or a document if the integration is available - Calendar only to propose times - do not create events PROCESS 1. List the decisions (max 7) - only those that were actually agreed. 2. List tasks as: who / what / by when. If the owner or date is missing, write "TBD". 3. List open questions and who should answer each. 4. List topics deferred. 5. Propose a next meeting time if the notes imply one is needed. OUTPUT - Decisions - Tasks (who / what / by when) - Open questions - Deferred topics - Proposed next meeting CONSTRAINTS - Do not fill in blanks. Missing information = "no data". - Do not assign a task to someone who did not agree to it. - Do not record loose discussion that reached no conclusion as a decision. QA Check and report: - how many tasks have a complete who/what/when and how many are incomplete, - whether every decision can be pointed to in the notes, - whether any decisions contradict each other. APPROVAL GATE Show me the result before saving it to Notion or sending it to attendees. Do not create calendar events - only propose them.
Workflow: Customer Feedback Analyzer
ROLE You are a customer experience analyst. You do not soften negative feedback. GOAL Analyze customer feedback from [SOURCE, PERIOD] and show which problems to fix first. INPUTS - Feedback / ticket file: [UPLOAD] - Columns: date, channel, text, rating, customer segment [ADJUST] - Our current product priorities: [LIST] - What counts as a critical problem: [DEFINITION] TOOLS - Data analysis on the file - A results sheet PROCESS 1. Propose problem categories that emerge from the data (not imposed up front). Show the list and STOP. 2. After approval, assign each item to a category. Leave ambiguous items in "unclassified". 3. Compute: count, share, trend over time, average rating per category, split by segment. 4. For the 5 most frequent categories quote 3 representative comments each (no personal data). 5. Combine frequency with severity and propose a fix order. OUTPUT - Category list with definitions - Table: category / count / share / trend / average rating - 3 anonymized quotes per category - Recommended fix order + rationale - Share of unclassified feedback CONSTRAINTS - Do not draw conclusions from a category with fewer than [N] items - mark it a weak signal. - Strip names, emails and order numbers from quotes. - Do not evaluate employees named in the feedback. QA Check and report: - whether category counts sum to the file's row count, - what share landed in "unclassified" (above 15%, propose better categories), - whether any personal data survived in the quotes, - whether the trend uses comparable periods. APPROVAL GATE Stop after proposing the categories. Do not publish the results outside the team without my approval.
Workflow: CRM Pipeline Auditor
ROLE You are a sales process auditor. You look at data, not at stories. GOAL Audit the [TEAM / PERIOD] pipeline and produce a concrete clean-up action list. INPUTS - Source: [CRM view / CSV export] - Funnel stages and their definitions: [LIST] - Normal time in stage: [e.g. stage 2: max 14 days] - "Dead deal" criteria: [e.g. no activity for 30 days] TOOLS - Reading the CRM or an export file - Data analysis - An action list sheet PROCESS 1. Lay out the pipeline: deal count and value per stage, owner, last activity date. 2. Find deals: inactive beyond the threshold, with a close date in the past, with no amount, with no contact, duplicated. 3. Compute stage-to-stage conversion and average time in stage. 4. Name the stage where we stall most often and show the data behind it. 5. Produce an action list: what to close, update, escalate - assigned to owners. OUTPUT - Pipeline breakdown by stage and owner - Deals needing action, grouped by problem type - Conversions and stage durations - Clean-up action list with owners - CRM data quality assessment (what is incomplete and how often) CONSTRAINTS - Do NOT modify CRM records. Read-only mode. - Do not evaluate salespeople - describe the state of the data and process. - Do not forecast revenue from incomplete records. QA Check and report: - whether per-stage value sums match the overall total, - whether a deal appears in two problem lists at once, - whether time thresholds are in business or calendar days (state which), - how many records you excluded and why. APPROVAL GATE All CRM fixes are done by a human. Your role ends with the action list handed over for approval.
Workflow: Website Change Monitor
ROLE You are the caretaker of the company site. You report differences, not opinions. GOAL Monitor the listed pages and notify me when something changes or stops working. INPUTS - URLs to watch: [LIST] - What to watch on each: [e.g. price, headline, CTA button, form, phone number] - Frequency: [e.g. every 6 hours / daily] - Alert channel: [Slack #channel / mail] TOOLS - Reading public pages - A scheduled task in condition-monitoring mode - A state history sheet PROCESS 1. Record the baseline state of every watched section. 2. On each check, compare the current state with the last recorded one. 3. No change - send no alert, append a history row. 4. Change or error (page down, form gone) - send an alert: what, where, before/after, when spotted. 5. Once a week, summarize all changes. OUTPUT - Change alert: URL, element, before, after, time - Alert when a page or element is unavailable - History sheet - Weekly summary CONSTRAINTS - Change nothing on the site. Read only. - Do not test forms by submitting fake entries. - Do not alert on normal dynamic content (counters, rotating testimonials). QA Check and report: - whether you compare the same element (selector / text fragment) as before, - whether the change comes from a mobile or language variant, - whether the alert contains the exact before and after fragment. APPROVAL GATE Before enabling automatic alerts, show me the baseline and one sample alert. No edits to site content - a human does that.
Workflow: Marketing Campaign Builder
ROLE You are a marketing strategist who starts from the goal, not from a post idea. GOAL Design the [NAME] campaign for [PRODUCT / SERVICE] aiming at [MEASURABLE GOAL] during [DATES]. INPUTS - Target audience and its problem: [DESCRIPTION] - Our advantage and the evidence for it: [LIST] - Channels available: [LIST] - Budget and people: [DESCRIPTION] - What we must not promise (compliance): [LIST] TOOLS - Reading product material - Web search for market context - A document and sheet for the schedule PROCESS 1. Formulate one core message and 3 test variants. Show them to me and STOP. 2. After approval, write the assets per channel, fitted to each channel's format. 3. Build the schedule: date, channel, format, asset, owner. 4. Define measurement: leading metric, supporting metrics, a mid-campaign checkpoint. 5. List risks and a plan B if the leading metric does not move after [X] days. OUTPUT - Core message + 3 test variants - Assets per channel - Schedule (date / channel / format / owner) - Measurement plan with a checkpoint - Risks and plan B CONSTRAINTS - Do not promise results we cannot prove. No invented statistics or testimonials. - Do not use client names without consent. - Respect the forbidden-claims list from INPUTS. QA Check and report: - whether every asset carries the core message, - whether every number in the copy has a source, - whether the leading metric is actually measurable with our tools, - whether the schedule is feasible with the stated resources. APPROVAL GATE Stop after the message and again before anything is published. Publish and send nothing on your own.
Workflow: Monthly Management Report
ROLE You are a management controller. Your report should lead to decisions, not to arguments about the data. GOAL Prepare the management report for [MONTH] for [RECIPIENT], combining data from [SOURCES]. INPUTS - Data sources and scope: [LIST + owner of each] - Last month's report structure: [UPLOAD if we have it] - Plan / budget for the month: [DATA] - One-off events I know about: [LIST] TOOLS - Reading sheets, CRM and the finance system - Data analysis and charts - A report document - A monthly scheduled task PROCESS 1. Gather data from all sources and reconcile the period and metric definitions. Show discrepancies between sources and STOP. 2. Once the data version is agreed, compute: plan vs. actual, month-over-month, year-over-year. 3. Explain the largest deviations. Separate one-off from structural factors. 4. Build the narrative: what happened → why → what follows → what we do. 5. Propose 3-5 decisions for next month with estimated impact and cost. 6. Keep last month's structure so reports stay comparable. OUTPUT - Executive summary (max 8 sentences) - Tables: plan vs. actual, MoM, YoY - Charts of key trends - Deviation explanations split one-off / structural - 3-5 decisions with impact and cost - Appendix: discrepancies between data sources CONSTRAINTS - If sources disagree, do not quietly average - show both numbers and state which you used. - Do not forecast future months unless I ask. - Do not put employee personal data in the report. QA Check and report: - whether each table sums correctly and matches the source total, - whether metrics are computed as last month (if not - describe the change), - whether every deviation above the threshold has an explanation, - which numbers could not be reconciled across sources. APPROVAL GATE Stop after the data reconciliation. Hand the report to me first - do not send it to the board, shareholders or outside the company.
A question to company data (data agent)
I have a question about company data: [QUESTION]. Before computing, establish and show: (1) the metric definition you will use, (2) the source and table, (3) the date range and filters. Then give the value, break it down by at least 3 dimensions and list cause hypotheses (clearly marked as hypotheses). Every number needs a definition and a source. If something is not in the data, say so - do not guess. STOP after showing the metric definition and wait for my approval.
Auditing a number before the board
Before using the number [VALUE] for [METRIC] over [PERIOD] run an audit: metric definition, source and table, applied filters and exclusions, period completeness, and whether the same source has a report with a different value. List the differences and state clearly what cannot be confirmed from the available data. Do not silently fix the number - show both versions with a note on which one you recommend and why.
Sales offer for a new client
You are an experienced B2B salesperson. Write a short (max 200 words) sales offer for the company [CLIENT NAME] from the [INDUSTRY] industry. We offer [PRODUCT/SERVICE]. Main benefit: [BENEFIT]. Price: [PRICE]. Tone: professional, warm, no slogans. End with a specific CTA and a proposal for a 15-minute call.
Follow-up after a sales meeting
Write a follow-up email after a client meeting. Meeting notes: [PASTE NOTES]. Goal: recap decisions, remind about the next step ([NEXT STEP]) and propose a time. Short, max 120 words, tone: friendly but concrete.
Handling the "too expensive" objection
The client says: "[CLIENT OBJECTION]". Give me 3 different replies: (1) empathic, (2) value-shifting (moves the conversation to ROI), (3) closing with a pilot proposal. Max 3 sentences each. No fluff.
Product description for a webstore
Write a product description for an online store. Product: [NAME + CATEGORY]. Features: [FEATURES]. For whom: [TARGET AUDIENCE]. Format: headline (max 60 chars), 3-sentence intro, 5 bullet benefits (not features!), FAQ with 3 questions, CTA. SEO keywords: [KEYWORDS].
Expert LinkedIn post
Write a first-person LinkedIn post. Topic: [TOPIC]. My take / story: [DESCRIBE]. Format: 1-line hook, 3-4 short paragraphs (max 2 sentences each), payoff line, community question. No emoji spam (max 2), no "As a CEO…". Length: 900-1200 characters.
Campaign email subject lines (10 A/B variants)
Generate 10 email subject lines for a campaign for: [SEGMENT]. Goal: [GOAL - e.g. webinar sign-up]. 5 curiosity-driven, 5 concrete with a number or benefit. Max 50 chars each. No clickbait, no exclamation marks.
Job posting - no corporate speak
Write a job posting for [POSITION] at [COMPANY NAME, INDUSTRY, SIZE]. Requirements: [REQUIREMENTS]. What we offer: [BENEFITS, SALARY RANGE]. Tone: human, concrete, no "dynamic team". Sections: about us (3 sentences), your day, what we expect, what we give, how to apply.
Interview questions
Prepare an interview kit for [POSITION]. Split into: 3 behavioural questions (STAR), 3 technical, 2 motivation, 1 live case study. For each add a short "what I'm looking for in the answer".
1-page company AI usage policy
Write a simple AI usage policy (ChatGPT, Copilot etc.) for a company in [INDUSTRY, SIZE]. Max 1 A4 page. Sections: purpose, what's allowed, what must NOT be pasted (personal data, trade secrets, client data), how to label AI-generated content, who approves new tools, contact for the responsible person. Tone: readable for non-IT.
Meeting summary from raw notes
I have raw meeting notes: [PASTE]. Turn them into a professional summary. Format: date and attendees, key decisions (max 5), tasks (who / what / by when), open questions, next meeting. Don't invent anything - if it's missing, mark "no data".
Step-by-step procedure (SOP)
Create an SOP for: [TASK]. Owner: [ROLE]. Frequency: [HOW OFTEN]. Format: purpose, tools/access required, numbered steps (each starts with a verb), final checklist, what to do when things go wrong (3 common issues + fix).
Contract / terms review
Review the document [PASTE - ANONYMIZED]. Return: (1) 5 key obligations on my side, (2) risks and negotiation points, (3) deadlines and penalties, (4) unusual clauses. You are not replacing a lawyer - flag what absolutely needs legal review.
Complaint response (3 tones)
A customer wrote: "[COMPLAINT]". Prepare 3 responses: (1) short and apologetic, (2) explanatory with a compensation offer [COMPENSATION], (3) firm when the complaint is unfounded. Each max 150 words, professional, never patronizing.
Website FAQ from 10 real tickets
I have 10 recurring customer questions: [PASTE]. Turn them into a website FAQ. For each: title as the customer's question, 2-4 sentence answer, optional link to a full guide. Group the questions into 3 logical categories.
Win-back phone script
Write a phone script to win back a churned customer. Reason for churn: [REASON]. Structure: greeting, soft opener, question about the reason, 3 possible paths (depending on the reply), retention offer [OFFER], close. Concrete sentences, not generalities.
Invoice - explain a line item
A client asks about an invoice line item: "[QUESTION]". Line description: [PASTE DESCRIPTION, AMOUNT, DATE]. Write a short, professional reply (max 100 words) explaining exactly what this line covers and what work/delivery it relates to.
Cash flow - explanation for non-finance owner
I have financials: [PASTE - revenue, costs, monthly balances]. Explain in plain language: (1) is there a liquidity problem, (2) which month is worst and why, (3) 3 concrete recommendations. No jargon, for an owner without a finance background. If something can't be concluded from this data - say so.
Competitor analysis (4P)
Analyze the competitor [NAME, URL] for my company [NAME, INDUSTRY] using the 4P model: product, price, place, promotion. Finish with: 3 of our advantages, 3 of theirs, 3 market gaps to exploit. Mark everything as hypotheses to verify - don't invent numbers.
Brief for an agency / freelancer
Write a brief for [AGENCY / FREELANCER] for the project: [PROJECT]. Sections: company context (3 sentences), business goal, target audience, scope, what's out of scope, success criteria (measurable), budget, deadline, contact. No fluff.
Quarterly business review (QBR) agenda
Plan a 90-minute QBR agenda for [DEPARTMENT]. Attendees: [N]. Quarter goals: [GOALS]. Format: time block, owner, purpose, output (decision / info / discussion). Start with a short icebreaker (5 min), end with concrete action items.
Quarterly OKRs (3 objectives)
Help me draft 3 OKRs for next quarter for [DEPARTMENT / COMPANY, SIZE]. Context: [STRATEGY, BOTTLENECK]. Each objective: ambitious and qualitative (1 sentence), 3 key results (measurable, with numbers and deadlines). At the end flag any risk that OKRs conflict with each other.
30-day content ideas generator
Propose 30 post / video ideas for a company in [INDUSTRY] for a full month. Audience: [PERSONA]. Channels: [LinkedIn / IG / blog]. Table format: day, channel, format (post / carousel / reel), topic, hook (1 sentence), CTA.
Payment reminder email (3 versions)
A client is [DAYS] days overdue on invoice #[NUM] for [AMOUNT]. Write 3 reminder versions: (1) soft - first reminder, (2) firm - second, (3) final - with legal-action notice. Each professional, non-aggressive, ready to send.
New hire 30-day onboarding plan
Create a first-30-days onboarding plan for a new hire in [POSITION] at [INDUSTRY]. Split by week. Per week: goal, people to meet, tasks, what they should be able to do by the end of the week. Add 3 manager check-ins (after week 1, 2 and 4) with concrete questions.
Rewrite this text - simplify and shorten
Take this text: [PASTE]. Rewrite it so that: (1) it's 40% shorter, (2) uses simple words (B1 level), (3) has short sentences (max 15 words), (4) keeps all key information. At the end list what you cut and why.
EN→PL translation with localization
Translate from English into Polish: [PASTE]. Don't translate literally - localize. Keep tone and intent. Replace idioms with Polish equivalents. Convert units (miles, pounds, USD) to metric and PLN [RATE]. At the end list 3 places where you deliberately deviated from the original.
Website audit - first impression
Act as a potential customer from [PERSONA] visiting [URL - describe what's there] for the first time. Answer: (1) in 5 seconds - what's on offer, (2) what irritates you, (3) what's missing to buy, (4) 5 concrete changes ordered by importance. No generic "improve UX".
Action plan from one sentence
I have a goal: "[DESCRIBE IN ONE SENTENCE]". Break it into: (1) 3 possible paths, (2) pick the best and justify, (3) concrete steps for the next 2 weeks (date, who, what), (4) how I'll know it's going well (3 indicators).
Explain this to a child (and my grandma)
Explain the concept / process: [PASTE SOMETHING DIFFICULT] twice: (1) to a 10-year-old - analogy, max 100 words, (2) to a retired accountant - no IT jargon, with a comparison to something they know from the office. End with one sentence for a hurried reader.
Pre-rollout risk assessment
We're rolling out [PROJECT / TOOL] at [INDUSTRY, SIZE]. Do a risk assessment: 8 top risks, each with probability (1-5) and impact (1-5), category (legal / technical / human / financial / reputational), a concrete mitigation. Sort by severity. Don't skip "team resistance".
ChatGPT Work: add tests for a function
ChatGPT Work: review the file [PATH] and function [NAME]. Add unit tests covering: happy path, invalid input, edge cases (empty/null/max). Use [JEST / VITEST]. First show me the plan (which tests you propose), wait for approval, only then create files.
ChatGPT Work: morning inbox triage
ChatGPT Work: go through my unread emails from the last 24h. Make a table: sender, subject, what it's about (1 sentence), priority (urgent / normal / ignore), suggested action. For urgent ones prepare reply drafts - DO NOT send anything, show me drafts for approval.
ChatGPT Work: weekly report (skill)
ChatGPT Work: create a skill "weekly-report". Instructions: 1) pull data from [SOURCE - CRM / sheet], 2) compute [KPIs - e.g. revenue, leads, conversion], 3) compare to previous week, 4) generate a PDF with 3 charts, 5) prepare a short Slack summary (max 10 sentences). Before saving the skill, show me its contents for approval.
ChatGPT Work: sort invoices in a folder
ChatGPT Work: the folder [PATH] contains PDF invoices with mixed names. For each read: invoice number, vendor, date and amount. Rename files to YYYY-MM_vendor_amount.pdf and build a summary in invoices.csv. First show me the rename plan - change nothing without approval.
ChatGPT Work: competitor price monitoring
ChatGPT Work: once a day check public competitor pricing pages: [URL LIST]. Log prices with dates to a table. If any price changes by more than [THRESHOLD %], prepare a short alert comparing old and new price. Show me the config before the first run.
ChatGPT Work: meeting notes → tasks
ChatGPT Work: here are raw meeting notes: [PASTE]. Produce: 1) decisions summary (max 5), 2) task list as who / what / by when, 3) open questions. Save the result in [FILENAME] in the project folder and show it to me before saving.
ChatGPT Work: project files audit
ChatGPT Work: audit the project folder [PATH]. Report: what file types are there, which are duplicates or old versions (by name and date), which haven't been touched for 6+ months. Propose a cleanup structure - but MOVE or DELETE nothing, only show the plan.
Company newsletter from 3 topics
Write a customer newsletter for a company in [INDUSTRY]. Topics this issue: [TOPIC 1, 2, 3]. Structure: subject (max 50 chars), short intro (2 sentences), 3 sections of 3-4 sentences with [URL], sign-off with CTA. Tone: human, no pushy selling.
Price increase announcement
Write a client email about a price increase from [DATE]. Reason: [REASON - e.g. rising costs]. Scale: [X%]. Rule: contract customers keep the current price until [DATE]. Structure: appreciation, specifics (what and by how much), justification (1-2 sentences, no over-apologizing), what the client gains, offer a call. Max 180 words.
20-minute SWOT analysis
Run a SWOT with me for [COMPANY, INDUSTRY, SIZE]. Ask me one question at a time (max 8), then build the SWOT with 4-5 points per quadrant. End with: 3 strategic conclusions linking strengths to opportunities, and 2 biggest risks to address this quarter.
Customer persona from real data
Based on descriptions of our 10 best clients: [PASTE - industry, size, how they bought, what they came with] build 2-3 buyer personas. For each: name, company profile, painful problem, what they read and where they hang out, buying objections, messaging that lands. Mark what's a data-driven finding vs. hypothesis.
Board report from department data
I have [DEPARTMENT] data for [PERIOD]: [PASTE / DESCRIBE]. Prepare a board report: 1) executive summary (max 5 sentences), 2) 3 key numbers vs. previous period, 3) what went well / poorly (3 points each), 4) 3 recommendations with cost and impact estimates. Language: executive, zero operational jargon.
Process description → ready diagram
I'll describe a process: [DESCRIBE IN YOUR OWN WORDS, e.g. how we handle a complaint]. Convert it into: 1) an ordered step list with roles (who does what), 2) a Mermaid flowchart, 3) a list of bottlenecks and where it most often breaks, 4) 2 automation candidates to start with.