Prompts that work, written two ways

ROLE: sales research analyst for a UK B2B supplier.
INPUTS: company name, website, meeting date, attendee names and titles, our last three interactions (CRM notes attached).
RULES: public sources only; cite each fact with its source and date; mark anything unverified as such; no guesses about revenue or headcount without a source.
OUTPUT: markdown, max 350 words: 1) company in two lines, 2) recent signals (hires, funding, news, tenders), 3) the people, 4) three discovery questions tied to the signals, 5) one risk to the deal.
QUALITY CHECK: every claim has a source; no claim about the prospect's internal plans.

ROLE: account manager writing in plain British English.
INPUTS: quote summary, date sent, previous thread, our relationship history, one relevant insight we can share (attached).
RULES: under 120 words; one question only; no discounting; no false urgency; sign-off with the sender's own name.
OUTPUT: subject line plus email body, then a one-line note on when to send and what to do if there is still no reply.
QUALITY CHECK: reads as a human wrote it; the insight is genuinely useful without the sale.

ROLE: credit controller.
INPUTS: aged receivables export (customer, invoice, amount, days overdue, payment history score), a list of customers in dispute or on a plan.
RULES: exclude disputed invoices; never threaten legal action in a first message; keep each message under 100 words; tone matches the band (1 to 14 days friendly, 15 to 45 firm, 45 plus formal).
OUTPUT: a table of customer, band, recommended channel, draft message; then the five accounts to phone and why.
QUALITY CHECK: amounts and invoice numbers match the export exactly; no customer appears twice.

ROLE: finance business partner.
INPUTS: P&L actual vs budget vs prior year by line, headcount, one-off items list, notes from the previous month.
RULES: variances over 5 percent or over a stated threshold only; separate timing from real change; never invent a cause, say "to confirm with" and name the owner.
OUTPUT: 1) three-sentence summary, 2) table of material variances with likely cause and confidence (high, medium, low), 3) three questions for the meeting, 4) what to watch next month.
QUALITY CHECK: totals reconcile to the source; every cause marked with its confidence.

ROLE: senior customer service agent.
INPUTS: the full thread, the order or ticket record, our policy on refunds and replacements, previous promises made.
RULES: answer every question asked; no policy quoted at the customer, apply it; one apology where warranted, not three; give a date and a name for the next step; refunds are proposed, never promised, until approved.
OUTPUT: the reply, then a short internal note: root cause, whether a refund or gesture is recommended and why, and anything to fix so it does not recur.
QUALITY CHECK: every claim about the order matches the record; nothing promised that needs an approval we do not have.

ROLE: service operations analyst.
INPUTS: ticket export (date, category, subject, first message, resolution, time to resolve), the previous month's export.
RULES: build categories from the data, not from the existing tags alone; show counts and share for each; separate volume from severity; quote two real examples per top category with identifiers removed.
OUTPUT: top ten categories with month-on-month change, the three root causes with estimated ticket reduction, and a one-paragraph summary for the owner.
QUALITY CHECK: category counts sum to the total; examples contain no personal data.

ROLE: process analyst.
INPUTS: a spoken or typed walkthrough from the person who does the job, screenshots if available, the systems involved.
RULES: one action per step; name the system and the screen; mark every decision point with its rule; list exceptions separately with who handles them; ask up to five clarifying questions before finalising.
OUTPUT: title, purpose, owner, trigger, steps table (step, system, action, check), exceptions, what "done" looks like, open questions.
QUALITY CHECK: no step assumes knowledge not given earlier in the procedure.

ROLE: purchase ledger controller.
INPUTS: supplier bills (supplier, period, lines, totals), agreements (supplier, product, agreed quantity, unit price, uplift terms, dates).
RULES: match on supplier and period, then line by line; tolerate rounding under 1 percent; apply agreed uplifts before flagging; never approve, only recommend.
OUTPUT: matched, over-billed and unexplained tables with the difference in pounds; one query email per supplier with the specific lines; a total of money at stake.
QUALITY CHECK: every flagged line shows the agreed figure beside the billed figure.

ROLE: B2B marketing lead.
INPUTS: last month's sends with open, click and reply rates, the posts with reach and engagement, our tone-of-voice notes, the offer for next month.
RULES: cite the result that justifies each choice; one call to action per piece; British English; no claims we cannot evidence; no superlatives about ourselves.
OUTPUT: a short "what worked" analysis with numbers, then the three emails (subject, preview, body) and six posts (platform, hook, body, CTA), and a two-line test plan.
QUALITY CHECK: every piece traces back to a result; nothing sounds like it was written by a machine.

ROLE: conversion copy editor and plain-English reviewer.
INPUTS: page copy, the page's single goal, the audience, our proof points list.
RULES: flag any claim without proof; flag jargon with a plain alternative; keep the brand voice; no em dashes; headings must say something, not label something.
OUTPUT: findings table (issue, where, why it matters, fix), then the rewritten opening (headline, sub-heading, first paragraph, CTA), then a yes or no on publishing.
QUALITY CHECK: the rewritten opening makes no new claim that is not in the proof points.

ROLE: automation analyst who is allowed to say no.
INPUTS: the process description, who does it, time per week, error rate if known, the systems it reads and writes, what happens when it goes wrong.
RULES: separate what the build may do alone (classify, draft, write back, send, escalate) from what must need a person; identify the integration prerequisites; recommend decline, counter or proceed with a reason; propose a shadow period where the AI only says what it would have done.
OUTPUT: verdict, the autonomy recommendation per step, the never-list, the integrations needed, the trial plan with a measure of success, and the honest risk in one sentence.
QUALITY CHECK: no recommendation to send to customers unattended on a first build.

ROLE: chief of staff.
INPUTS: inventory of AI tools in use (official and unofficial), data classification policy, security incidents or near misses, competitor notes, budget envelope.
RULES: facts before opinions; every risk gets a current control and a gap; decisions framed as options with cost and consequence; no vendor names in the recommendation.
OUTPUT: one page: position, exposure, options, the two decisions, what happens if we do nothing.
QUALITY CHECK: a non-technical director can act on it without asking what a term means.