Guide
Prompt Packs by Role: Copy-Paste Prompts for Marketing, Ops, Finance and Support
A prompt for “anyone in marketing” and one for a hotel marketing manager chasing shoulder-season occupancy produce different output. Role, context, task, format.
A prompt for “anyone in marketing” and one for a hotel marketing manager chasing shoulder-season occupancy produce different output. Not because the second is cleverer, but because it supplies what a generic prompt can’t: audience, constraint, and the format the output needs to land in. Role, context, task, format — twenty complete prompts below, five each for marketing, operations, finance and support.
Most “AI prompt library” content is a wall of one-liners: “Write a marketing plan.” “Summarise this report.” “Draft a customer email.” They work fine as a demo and fall over in real use, because they’re missing what makes a prompt useful to a specific role — context the model can’t guess.
“Write me a report” is a task. “You’re a management accountant explaining a margin drop to a department head who’s never read a P&L” is a role, with context built in. Every prompt here follows the same shape: a role, context only you can supply, one clear task, and an output format that saves you the second draft. If you still want the five-ingredient craft behind that shape, start with Prompting Basics.
What makes a prompt “role-specific” (not just role-labelled)
Slapping “You are a marketer” in front of a generic task doesn’t make a prompt role-specific — it makes it role-labelled. Anthropic’s own prompting documentation is blunt about this: treat the model like a brilliant new employee with zero context on your norms — the golden rule is to show the prompt to a colleague with no background on the task; if they’d be confused, the model will be too.
Slapping “You are a marketer” in front of a generic task doesn’t make a prompt role-specific — it makes it role-labelled.
A genuinely role-specific prompt has four parts doing distinct work:
- Role — who the model is being for this task, and what it must never do. A support agent doesn’t promise refunds outside policy; an accountant flags issues, not approves them.
- Context — the facts only you know: the audience, the numbers, the policy limits, the transcript.
- Task — one clear instruction, not three vague ones bundled together.
- Output format — the structure you’ll actually use: a table, a word count, a tone.
Every prompt below is built in that order — copy the shape as much as the wording, since that’s what makes it portable across Claude, ChatGPT, Copilot or Gemini. See LOKAL’s AI Tools & Models guide for what each one costs.
Marketing prompt pack
Marketing prompts break down first when the audience and offer are generic — fill in a real campaign below, not a hypothetical one. Want a live taste of this first? See LOKAL’s public Claude AI for Marketing workshop.
Prompt 01
Weekly content calendar
When to use: you need a week of posts that actually support the campaign you’re running, not generic content ideas.
You are a social media strategist who writes in [BRAND NAME]'s established voice. Context: We are a [BUSINESS TYPE] targeting [AUDIENCE DESCRIPTION]. We post on [PLATFORMS AND FREQUENCY]. Our current focus is [CAMPAIGN OR SEASONAL FOCUS, e.g. shoulder-season occupancy]. Task: Draft 7 post ideas for the coming week, one per day, that support this focus. Output format: A table with columns for Day, Platform, Format (carousel/reel/single image/text), Hook (first line, written in full), and Goal (awareness/engagement/conversion).
Prompt 02
Repurpose one long asset into a week of posts
When to use: you’ve got a blog, video or case study and need a week of posts without starting from scratch.
You are a content repurposing specialist for [BRAND NAME]. Context: Our pillar content: [PASTE BLOG POST, VIDEO TRANSCRIPT OR CASE STUDY]. Audience: [AUDIENCE DESCRIPTION]. We publish on [PLATFORMS]. Task: Break this single piece into 6 standalone posts spread across 2 weeks, each pulling a different insight or section. Output format: For each post — platform, format, the specific section it draws from, a full caption (not a summary of one), and suggested posting day.
Prompt 03
Ad copy variant tester
When to use: you’re about to spend on ads and want five angles before you pick one.
You are a direct-response copywriter who has run paid campaigns for [INDUSTRY] businesses. Context: We're promoting [OFFER/PRODUCT] to [AUDIENCE] on [PLATFORM]. Budget and timeline: [BUDGET/DATES]. Current ad copy: [PASTE EXISTING AD OR BRIEF]. Task: Write 5 variations of the primary text and headline, each testing a different angle (price, urgency, social proof, problem/solution, curiosity). Output format: Numbered list — angle name, primary text (under 125 words), headline (under 40 characters), and which metric you'd expect it to move.
Prompt 04
Campaign brief from a rough idea
When to use: you’ve got a one-line idea and need a brief a designer and copywriter can work from.
You are a campaign strategist producing a brief a designer and a copywriter can work from without asking follow-up questions. Context: The idea: [ONE-SENTENCE IDEA]. Objective: [OBJECTIVE]. Audience: [AUDIENCE]. Budget: [BUDGET]. Deadline: [DATE]. Brand constraints: [TONE/VISUAL RULES, IF ANY]. Task: Turn this into a one-page campaign brief. Output format: Sections for Objective, Audience & Insight, Key Message, Channels & Formats, Timeline, Success Metrics — each 2–4 sentences or a short list, no filler.
Prompt 05
Competitor content gap scan
When to use: you want to see what competitors cover that you don’t, without inventing the gaps.
You are a competitive content analyst for [INDUSTRY]. If you have live web access, use it; otherwise work only from the material I paste in. Context: Our competitors are [COMPETITOR NAMES/URLS]. We sell [PRODUCT/SERVICE] to [AUDIENCE]. Here are their published topics/titles I've collected: [PASTE LIST]. Here's what we've already covered: [PASTE OUR TITLES]. Task: Identify content topics or formats they cover that we don't, and flag anything you're inferring rather than confirming from the material provided. Output format: A prioritised list of 8 gaps, each with a one-line rationale, a suggested format (guide, comparison page, video, calculator), and a confidence note (confirmed from source / inferred).
Operations prompt pack
Ops prompts live or die on the format — a process nobody can audit isn’t a process, it’s a conversation. Each of these ends in something you can point to and check.
Prompt 06
Turn messy meeting notes into an action list
When to use: a meeting just ended and you need the action list, not a rewrite of the conversation.
You are an operations coordinator who turns rough meeting notes into a clean action list without adding new work that wasn't discussed. Context: Meeting notes/transcript: [PASTE NOTES]. Attendees: [NAMES/ROLES]. Project: [PROJECT NAME]. Task: Extract every action item actually agreed in the notes. Output format: A table — Action, Owner, Deadline (state "not specified" if none was given), and Source Quote (the line the action came from).
Prompt 07
SOP from a process walkthrough
When to use: someone can describe how they do the job, but nothing’s written down a new hire could follow.
You are a process documentation specialist writing SOPs that a new hire could follow with zero prior context. Context: Transcript or rough description of how we currently do [PROCESS NAME]: [PASTE TRANSCRIPT/NOTES]. Done by [ROLE], using [TOOLS/SYSTEMS], roughly [FREQUENCY]. Task: Turn this into a numbered SOP. Output format: Title, Purpose (1 sentence), Role Responsible, Tools/Access Required, Trigger (what starts this process), then numbered steps — each step ends with "Done when: [observable result]".
Prompt 08
SOP gap analysis
When to use: you suspect the documentation doesn’t match the work, and you need a shortlist of what to fix first.
You are an operations auditor reviewing process documentation for gaps. Context: Department: [DEPARTMENT]. Reason for review: [TRIGGER]. Processes this team runs: [LIST]. SOPs we currently have on file: [LIST OR "NONE"]. Recent issues or audit findings: [PASTE, IF ANY]. Task: Compare the process list against what's documented. Output format: A table — Process, SOP Exists (Y/N), Last Updated (or "unknown"), Risk If Undocumented (low/med/high), Priority to Write/Update. Close with a 3-item shortlist of what to fix first.
Prompt 09
Diagnose a recurring bottleneck
When to use: a process keeps slowing down and you want the constraint, not another symptom.
You are a process improvement analyst who looks for the actual constraint, not the symptom. Context: Process: [PROCESS NAME]. Where it slows down: [STAGE/HANDOFF]. Typical cycle time: [TIME]. Target: [TARGET TIME]. Relevant data: [PASTE TICKETS/TIMESTAMPS/VOLUME DATA]. Task: Identify the most likely root cause of the delay, distinguishing it from contributing factors. Output format: One paragraph naming the root cause and why, then a short list of 3 fixes ranked by effort-to-impact, each with what you'd measure to confirm it worked.
Prompt 10
Vendor/supplier comparison
When to use: you have quotes and need a comparison a manager can approve from.
You are a procurement analyst producing a comparison a manager can approve from, without needing a follow-up meeting. Context: We need [PRODUCT/SERVICE]. Options: [VENDOR NAMES]. Must-haves: [REQUIREMENTS]. Budget ceiling: [AMOUNT]. What I know about each: [PASTE QUOTES/NOTES/SPEC SHEETS]. Task: Compare the options against our must-haves, using only the information provided — flag data gaps rather than guessing. Output format: A comparison table (Vendor, Price, Meets Must-Haves Y/N per requirement, Gaps, Notes) plus a one-paragraph recommendation.
Finance prompt pack
Finance prompts need one more guardrail: every one below treats the model as a drafting and pattern-spotting tool, not the sign-off. Keep a human on every number before it moves.
Prompt 11
Variance commentary
When to use: leadership wants to know why the numbers moved, in English, not jargon.
You are a management accountant drafting commentary for a leadership audience that doesn't want jargon. Context: Period: [MONTH/QUARTER]. Accounts or cost centres in scope: [LIST]. Figures: [PASTE ACTUALS VS BUDGET/PRIOR PERIOD]. Known one-off items this period: [LIST, IF ANY]. Task: Explain the three largest variances (favourable or unfavourable) in plain language, and note anything that looks like a timing difference versus a real trend. Output format: A short paragraph per variance (driver, amount, one-off or trend), then a 1-line executive summary at the top.
Prompt 12
Reconciliation discrepancy summary
When to use: two sources don’t match and a controller needs the gaps flagged, not resolved.
You are a finance analyst summarising a reconciliation for a controller to review and sign off — you flag, you don't decide. Context: Source A: [SYSTEM/ACCOUNT, e.g. general ledger]. Source B: [SYSTEM/ACCOUNT, e.g. bank statement]. Period: [DATE RANGE]. Materiality threshold: [AMOUNT]. Data: [PASTE OR ATTACH TRANSACTION LIST]. Task: Identify discrepancies above the materiality threshold between the two sources. Output format: A table — Transaction/Item, Source A Amount, Source B Amount, Variance, Likely Cause (timing/duplicate/missing/unknown), then total match rate and total variance value. End with "for human review" — do not mark anything as resolved.
Prompt 13
Cash flow scenario narrative
When to use: a founder needs to know when cash gets tight, not a model dump.
You are an FP&A analyst preparing scenarios for a founder who needs to make a decision, not read a model. Context: Current cash position: [AMOUNT]. Monthly burn/revenue trend: [FIGURES OR PASTE DATA]. Known upcoming commitments: [LIST]. Time horizon: [MONTHS]. Task: Build 3 scenarios — base, conservative (revenue -15%), and stretch (revenue +15%) — and state the month cash would run low in each, if any. Output format: A short table (Scenario, Ending Cash at [horizon], Month of Concern if any) followed by 2–3 sentences of plain-language commentary per scenario.
Prompt 14
Expense anomaly scan
When to use: you’re looking for outliers, not approving or rejecting spend.
You are an internal auditor scanning for outliers, not approving or rejecting spend. Context: Department: [DEPARTMENT]. Period: [MONTH]. Normal range for this category: [AMOUNT RANGE, IF KNOWN]. Data: [PASTE EXPENSE LIST/CSV]. Task: Flag transactions that are duplicates, above the normal range, missing a category, or missing a receipt reference. Output format: A table — Transaction, Amount, Flag Reason, Severity (low/med/high). Add a one-line note for anything you're uncertain about rather than guessing an explanation.
Prompt 15
Budget-vs-actual explainer for non-finance stakeholders
When to use: a department head with no finance background needs to understand where they overspent.
You are a finance business partner translating numbers for a department head with no finance background. Context: Department: [DEPARTMENT]. Period: [MONTH/QUARTER]. Budget: [AMOUNT]. Actual: [AMOUNT]. Line-item detail: [PASTE BREAKDOWN]. Task: Explain where they're over or under budget and why, in language a non-finance manager would understand on first read. Output format: 3 short paragraphs — headline number, biggest driver, and one recommended action — plus a simple table of the top 3 line items by variance.
Customer support prompt pack
Support prompts live or die on tone and policy limits — the fastest way to a bad experience is a reply that sounds right but promises something outside policy.
Prompt 16
Ticket triage and tagging
When to use: a ticket just landed and you need it classified, not answered.
You are a support triage assistant. You classify, you don't respond to the customer. Context: Our categories: [LIST OF CATEGORIES]. Priority levels: [PRIORITY LEVELS AND DEFINITIONS]. Ticket: [PASTE TICKET TEXT]. Task: Classify this ticket and assess urgency. Output format: Category, Priority, one-line summary for the queue, and a flag for "needs human judgement" if sentiment is angry or the issue is ambiguous.
Prompt 17
Empathetic response to a complaint
When to use: a customer is unhappy and the reply has to stay inside policy.
You are a senior support agent for [BRAND NAME] who follows our tone guide: [TONE NOTES] and never promises anything outside policy: [POLICY LIMITS]. Context: Customer message: [PASTE MESSAGE]. Order/account details: [RELEVANT DETAILS]. What we can actually offer here: [OPTIONS]. Task: Draft a reply that acknowledges the specific issue (not a generic apology), explains next steps, and offers only what's within policy. Output format: Ready-to-send email, under 150 words, no placeholders left unfilled, sign-off as [AGENT NAME/BRAND].
Prompt 18
Macro library builder
When to use: agents keep rewriting the same replies and you want reusable macros with the variable bits marked.
You are a support operations lead building a macro library so agents stop rewriting the same replies. Context: 10 real tickets and the replies that resolved them well: [PASTE EXAMPLES]. Brand voice: [TONE NOTES]. Task: Turn the recurring patterns into reusable macros with blanks for the details that change. Output format: For each macro — Trigger (when to use it), Macro text with [BRACKETED FIELDS] for the variable parts, and a note on what must never be copy-pasted without editing (e.g. dates, amounts).
Prompt 19
Escalation script
When to use: a difficult case is going upstairs and the next person needs a clean handover, not a dump of the thread.
You are a support lead preparing a frontline agent to escalate a difficult case correctly. Context: Issue type: [ISSUE TYPE]. Customer history: [VIP/repeat/first-time]. What's already been tried: [LIST]. Escalation path: [WHO/WHERE THIS GOES NEXT]. Task: Write the internal handover note the next person will read before they open the ticket. Output format: Situation (2 sentences), What's Been Tried, What the Customer Wants, Recommended Next Step, Risk If Mishandled (e.g. churn, refund, PR) — under 120 words total.
Prompt 20
CSAT theme analysis
When to use: you have a pile of comments and need the themes, not individual complaint rewrites.
You are a CX analyst finding patterns across feedback, not individual complaints. Context: Period: [MONTH]. Data: [PASTE CSAT COMMENTS/SCORES, ANONYMISED]. Current known issues: [LIST, IF ANY]. Task: Group the feedback into themes and identify which themes are new versus already known. Output format: A ranked list of themes (Theme, Frequency, Sample Quote, New/Known), then a 3-sentence summary of what's driving the biggest score movement this period.
How to adapt this pack to your business
Three moves turn a prompt above from “close enough” into something your team reuses:
- 01
Replace every bracket before you save it
A prompt with [BRAND NAME] still in it is a template; one with your actual brand name, tone words and policy limits filled in is a working tool. Keep a master copy per prompt with defaults filled in, and only bracket what genuinely changes each time — the customer message, the period, the transcript.
- 02
Attach your own examples
Every prompt above works better with 2–3 real examples of good output pasted above the task — your last good campaign brief, your last well-handled complaint reply, your last clean SOP. It’s the single highest-leverage change you can make to any prompt here.
- 03
Fix one thing at a time
If the output isn’t right, change the context, the task or the format — not all three at once — and run it again. That’s the difference between giving up on a prompt after one bad draft and having 40 of these dialled in by quarter’s end.
If you’d rather your team build and dial in these prompts live, on your actual work, that’s most of what a LOKAL AI workshop covers — and how corporate AI training is scoped for teams of 10 to 4,000. Every session ends with participants keeping a copy-paste prompt library built from their own tasks, not a generic one.
Keep going: the prompting basics pack covers the five ingredients behind every prompt here, or browse the full AI guides hub.
FAQ
Common questions
What makes a prompt “role-specific” instead of just role-labelled?
Four parts working together: a role that includes what the model shouldn’t do, context only your business can supply, one clear task, and a specified output format. Writing “You are a marketer” above a vague task is role-labelling, not role-specific prompting.
How do I adapt these prompts to my own business?
Fill in every bracket with your actual details, then add 2–3 examples of output you already know is good. After that, change one part at a time — context, task or format — rather than rewriting the whole prompt.
Do these prompts work in ChatGPT and Gemini, or only Claude?
The structure works everywhere — Anthropic, OpenAI and Google publish the same underlying advice: a clear role, explicit context, and a defined format. LOKAL trains teams on Claude, ChatGPT, Copilot and Gemini for this reason; see the AI Tools & Models guide for what each costs.
How many of these should my team actually use day to day?
3–5, used weekly with your own examples built in. Running all 20 occasionally is worse. Start with whichever task currently eats the most time.
Keep reading
Related reading
Done-for-you
Talk to us about AI training
These 20 prompts are close enough to start using this week. The packs LOKAL builds inside training rooms come from your team’s real tickets, campaigns and numbers — which is why they get used past week one.