{"slug": "ai-prompt-templates-you-can-steal-25-fill-in-the-blank-frameworks", "title": "AI Prompt Templates You Can Steal: 25 Fill-in-the-Blank Frameworks", "summary": "A guide to 25 reusable AI prompt templates emphasizes that templates separate reusable thinking from situation-specific details using named variables, with a context block as the foundation. The templates include a pre-mortem for pressure-testing decisions, and the guide notes that expert personas reduce factual accuracy (68.0% vs 71.6% on a knowledge benchmark).", "body_md": "# AI Prompt Templates You Can Steal: 25 Fill-in-the-Blank Frameworks\n\n**Short answer:** A prompt is written once and thrown away. A\n\n**template** separates the reusable thinking from the situation-specific detail — replacing that detail with named variables so the same structure works next month, and so you can hand it to a colleague. Every one of the 25 below is a template, not a prompt: fill the\n\n`[BRACKETS]`\n\nand run. Start with **#1, the context block**— it’s the one that makes the other 24 work.\n\n**TL;DR — How to use this page**\n\n**Fill in #1 first.** The context block gets pasted at the top of most of the others. Twenty minutes once, then reused forever.**You need about a dozen, not 25.** Take the ones matching decisions you actually repeat. A library you don’t rerun is a bookmark collection.**No expert personas on factual templates.** Research found they*reduce*accuracy — 68.0% vs 71.6% on a knowledge benchmark. We’ve kept them only where style is the job.**Every template names its variables.** If you’d have to rewrite it next time, it isn’t a template.**Template #25 makes more templates**— it converts any prompt that worked into a reusable one.\n\n**✔ Best for** Founders, operators and small teams who want a working prompt library today, and anyone tired of rewriting the same prompt from scratch every time.\n\n**✕ Skip if** You want depth on one function — we have full guides for sales, marketing, hiring, finance, support and operations — or you’re looking for developer API patterns.\n\n## What makes a template reusable (and not just a prompt)?\n\n**Five components. Miss any one and it produces inconsistent results across uses.** This is the part competitors’ “50 prompts” lists skip entirely — they hand you 50 one-offs and call it a library.\n\n**1. Named variables** Every situation-specific detail becomes\n\n`[DESCRIPTIVE NAME]`\n\n— descriptive enough that someone else knows what to put there.**2. Explicit output format** Structure, length, and shape stated. Never left for the model to infer, or it varies every run.\n\n**3. Stated constraints** Including what to\n\n*avoid*. Banned phrases do more work than any tone instruction.\n\n**4. Uncertainty handling**“Flag what you’re inferring. Write NOT FOUND rather than guessing.” Stops silent gap-filling.\n\n**5. Thinking, not answers** Describes the reasoning you want repeated — not the conclusion you got last time.\n\n**The test:** could you hand this to a colleague working on a completely different case and get useful output? If not, it’s still a prompt.\n\n**The 25 templates**\n\n## Foundation\n\nFill this in once. It goes at the top of most templates below.\n\n**The context block**\n\nUse: at the top of nearly every other template. The single highest-leverage thing on this page.\n\n```\nMY CONTEXT (reuse everywhere):\nBusiness: [WHAT YOU SELL, ONE PLAIN SENTENCE]\nStage/size: [REVENUE, HEADCOUNT]\nCustomer: [WHO BUYS — role, company size, or life situation]\nTheir problem: [THE PAIN THEY PAY TO REMOVE, IN THEIR WORDS]\nWe replace: [WHAT THEY DO INSTEAD TODAY]\nProof: [2-3 SPECIFIC RESULTS WITH NUMBERS OR NAMES]\nOnly true of us: [REAL DIFFERENTIATOR — or say you don't have one]\nConstraints: [BUDGET, TIME, HEADCOUNT, RUNWAY]\nAlready tried: [SO YOU DON'T GET RECYCLED SUGGESTIONS]\n\nSTANDING RULES:\n- Never invent a fact, statistic, or capability. If you need\n  something I haven't given you, ask.\n- Flag anything you're inferring as [ASSUMPTION].\n- Be direct. No preamble, no encouragement.\n```\n\n## Strategy & decisions\n\nTemplates 2–5. For pressure-testing before you commit resources.\n\n**Pre-mortem**\n\nUse: before any significant decision. Inverts the model’s bias toward agreeing with you.\n\n```\n[CONTEXT BLOCK]\nI'm about to [DECISION].\nIt's 12 months from now and this failed clearly.\n1. Write the post-mortem — what went wrong, ranked by\n   likelihood. Specific to my business, not generic advice.\n2. For each: the EARLIEST signal I could have caught it by.\n3. Which failures are fatal vs recoverable?\n4. What would I have to believe for this to work? Mark\n   which of those are untested.\nDo not reassure me. Assume failure and work backwards.\n```\n\n**Constraint diagnosis**\n\nUse: when growth has stalled and you’re not sure why.\n\n```\n[CONTEXT BLOCK]\nMy numbers, last [PERIOD]: [PASTE METRICS]\nI think my bottleneck is [YOUR HYPOTHESIS].\n1. What do these numbers suggest the real constraint is?\n   Argue against my hypothesis if they don't support it.\n2. Which single number, improved 20%, most changes my\n   outcome? Show the reasoning.\n3. What data am I missing that would change your answer?\n4. The smallest test to confirm the real constraint in\n   2 weeks.\n```\n\n**Decision brief**\n\nUse: when a choice keeps getting reopened. Forces the trade-off into the open.\n\n```\n[CONTEXT BLOCK]\nDecision: [WHAT I'M CHOOSING BETWEEN]\nDeadline: [WHEN] · Reversible? [YES/NO/PARTLY]\nStructure this for me:\n1. What am I ACTUALLY choosing between? State it in one\n   sentence per option, stripped of framing.\n2. What does each option cost that the other doesn't —\n   money, time, optionality, morale?\n3. What would have to be true for each to be right?\n4. What information would change my mind, and can I get\n   it before the deadline?\n5. If reversible, what's the cheapest version I could\n   try first?\n6. Name the option I appear to have already chosen\n   emotionally, and what's driving that.\n```\n\n**Assumption extraction**\n\nUse: turning a vague idea into something you can actually test.\n\n```\n[CONTEXT BLOCK]\nMy idea: [2-4 PLAIN SENTENCES]\nWho it's for: [SPECIFIC — not \"small businesses\"]\nDon't evaluate whether it's good. Instead:\n1. Rewrite as a testable problem statement: \"[PEOPLE]\n   experience [PAIN] when [SITUATION] and cope by\n   [WORKAROUND].\"\n2. List every assumption that must be TRUE, split into:\n   problem / customer / solution / money / channel.\n3. Mark each TESTED or UNTESTED, and FATAL or SURVIVABLE\n   if false.\n4. Rank fatal+untested by cheapest to test. That's my\n   to-do list.\n5. Name the assumption I seem most attached to.\n```\n\n### Get all 25 as a copy-paste library\n\nThe full set in Notion and plain text, with the context block pre-formatted, a “which template for which job” index, and the template-maker. Free.\n\n[Get the library →](#aitpl-lead-magnet)\n\n## Marketing & content\n\nTemplates 6–10. Personas kept only where style is the job.\n\n**Question mining**\n\nUse: content planning. Starts from real customer questions, not keyword lists.\n\n```\n[CONTEXT BLOCK]\nGenerate 25 content topics from real customer questions.\nInputs:\n- Objections in sales calls: [LIST 3-5]\n- Questions asked before buying: [LIST 3-5]\n- Mistakes I watch prospects make: [LIST 3-5]\nReturn a table: | question as a customer would type it |\nintent | the angle only WE can take | format | buying stage |\nRules: every title is a question or specific claim, never\na vague noun phrase. Skip anything a competitor could\npublish identically. Flag topics where we lack the proof\nto be credible.\n```\n\n**Content outline for AI search**\n\nUse: structuring an article to be quoted by AI engines, not just ranked.\n\n```\n[CONTEXT BLOCK]\nTopic: [THE QUESTION THIS ANSWERS]\nBuild an outline optimised to be QUOTED, not just ranked:\n1. A one-sentence direct answer, under 40 words, written\n   so it can be lifted verbatim.\n2. 3-5 key takeaways as a scannable list.\n3. Headings phrased as questions people actually ask.\n4. Under each, a self-contained 40-120 word passage that\n   makes sense pulled out of context.\n5. At least one comparison table.\n6. 5-8 FAQ questions targeting long-tail variations.\nFor each section: the specific claim it makes and the\nevidence supporting it. Flag any section where we'd be\nasserting something without proof.\n```\n\n**One idea, four platforms**\n\nUse: repurposing. Persona is fine here — style is the job.\n\n```\n[CONTEXT BLOCK]\nVoice sample to match: \"[PASTE 2 PARAGRAPHS OF YOUR\nBEST WRITING]\"\nSource material: [ARTICLE / CUSTOMER STORY / RESULT /\nPOINT OF VIEW]\nTurn this into platform-native posts — genuinely different\nexpressions, not the same text reformatted:\nLINKEDIN: 120-180 words. Open with a specific claim or\n  concrete moment, never a question.\nX: under 280 characters. Sharpest version of the claim.\nINSTAGRAM: caption, first line works as a standalone hook.\nNEWSLETTER: 50 words that earn a click without clickbait.\nFor each: 3 hook variations.\nBanned: \"Here's the thing\", \"Let that sink in\", engagement\nbait, and any post that would work for a competitor.\n```\n\n**Hook rewriter**\n\nUse: when a draft is fine but the opening is flat.\n\n```\n[CONTEXT BLOCK]\nMy draft: [PASTE]\n1. Rate the first line — would someone stop scrolling? If\n   no, say which word lost them.\n2. Rewrite the opening 5 ways: a specific number; a\n   contrarian claim we can defend; a concrete moment; a\n   named mistake the reader is making; the plainest\n   statement of the point.\n3. For each, the risk (too aggressive, vague, overpromises).\n4. Which one does the BODY actually deliver on? A hook the\n   piece doesn't earn costs more than a boring one.\n```\n\n**Positioning stress test**\n\nUse: when everything you write sounds like everyone else.\n\n```\n[CONTEXT BLOCK]\nOur current pitch: \"[HEADLINE / ELEVATOR PITCH]\"\nCompetitors say: [COMPETITOR 1] · [2] · [3]\n1. Strip out every claim a competitor could also make.\n   What's left that's genuinely only true of us?\n2. If nothing is left, say so plainly — that's the finding.\n3. Rewrite around what survived. 3 options, under 15 words,\n   in customer language not marketing language.\n4. For each: the customer it wins and the customer it\n   loses. Trade-offs are the point.\n```\n\n## Sales\n\nTemplates 11–14. No personas — these are judged on accuracy.\n\n**Prospect research**\n\nUse: before outreach. Run on a model with live web access.\n\n```\n[CONTEXT BLOCK]\nProspect: [NAME], [TITLE] at [COMPANY] ([URL])\nResearch them, then give me:\n1. THREE specific, recent, verifiable company facts\n   (last 6 months) — funding, hires, launches, press.\n2. TWO signals they may have the problem I solve, with\n   the evidence for each.\n3. ONE thing about this person — post, talk, career move.\n4. The single most likely reason they'd ignore my email.\nRules: only facts you can point to a source for. If you\ncan't verify something, write NOT FOUND rather than\nguessing. Rank by relevance to the problem I solve.\n```\n\n**Deal scoring**\n\nUse: pipeline review. Built to disqualify, not to reassure.\n\n```\n[CONTEXT BLOCK]\nWhy we usually lose: [TOP 3 LOSS REASONS]\nThe deal: [COMPANY, CONTACT + WHETHER THEY HOLD BUDGET,\nSTAGE, WEEKS IN PIPELINE]\nWhat they SAID: [QUOTES]\nWhat they DID: [ACTIONS WITH A COST — or \"nothing\"]\nNext step booked: [YES + DATE / NO]\nFind reasons this WON'T close.\n1. Split SAID from DID. Only DID predicts anything.\n2. For each — pain, power, urgency, process, competition —\n   mark KNOWN / ASSUMED / UNKNOWN.\n3. Which of my loss patterns does this most resemble?\n4. Score LIKELY / AT RISK / LIKELY DEAD. Treat UNKNOWN as\n   negative, not neutral.\n5. The one question that would most change this. Exact\n   wording.\n```\n\n**Objection library**\n\nUse: once per offer. Build it before the call, not during.\n\n```\n[CONTEXT BLOCK]\nAct as a skeptical qualified buyer: [TITLE] at a\n[COMPANY TYPE]. You have budget, the problem is real,\nyou're unconvinced.\nOur offer: [WHAT AND PRICE]\n1. The 10 objections you'd actually raise, ranked by how\n   likely each is to kill the deal.\n2. For each: the STATED objection, the REAL concern\n   underneath, and the evidence that would resolve it.\n3. A response to each under 60 words — acknowledge without\n   agreeing, one clarifying question, specific evidence,\n   concrete next step.\n4. Flag which objections are NOT sales problems — where\n   our product or price is genuinely wrong for them.\n5. For the top 3: the question I could ask EARLIER so the\n   objection never forms.\n```\n\n**Follow-up from notes**\n\nUse: within an hour of any call.\n\n```\n[CONTEXT BLOCK]\nMy raw call notes: [PASTE — messy is fine, include their\nexact words where you have them]\nWrite the follow-up:\n- Open by referencing something SPECIFIC they said, in\n  their words not mine.\n- Confirm the problem as THEY framed it.\n- One piece of evidence for their specific concern.\n- The next step as agreed, with a date.\nUnder 120 words. No \"great speaking today\", \"as discussed\",\n\"circling back\". One CTA. Only use facts in my notes — if\nyou need something I didn't record, ask. Do NOT invent a\ndetail about the conversation.\nThen separately: 3 things I should have asked but didn't,\nand whether it's still possible to ask them.\n```\n\n## Customer support & retention\n\nTemplates 15–17. Diagnose before drafting.\n\n**Ticket diagnosis**\n\nUse: before replying to anything non-trivial. Catches churn signals.\n\n```\n[CONTEXT BLOCK]\nKnown current issues: [OUTAGES, DELAYS, RECENT PRICE OR\nPOLICY CHANGES]\nThe ticket: [PASTE VERBATIM]\nAbout this customer: [TENURE, SPEND, PAST TICKETS, RECENT\nACCOUNT EVENTS — or \"nothing\"]\nDiagnose. Don't draft a reply yet.\n1. Literal request. 2. Underlying problem — what likely\nhappened just before they wrote in?\n3. Emotional state 1-5, quoting the words that tell you.\n4. CHURN SIGNAL — export requests, contract/notice queries,\n   competitor mentions, cooling tone vs previous tickets,\n   timing near a price change. Rate LOW/MED/HIGH.\n5. The real question they haven't asked.\n6. Route: factual answer, or needs a human?\n7. What I'd need to know to handle this properly.\n```\n\n**Service recovery**\n\nUse: when you got it wrong. A handled failure can produce more loyalty than none.\n\n```\n[CONTEXT BLOCK]\nWhat went wrong: [HONESTLY, INCLUDING OUR PART]\nImpact on them: [WHAT IT COST THEM]\nWhat I can offer: [REAL OPTIONS AND LIMITS]\nWrite the recovery message:\n1. Own it in the first sentence. Active voice — \"we sent\n   it to the wrong address\", not \"an error occurred\".\n2. Show we understand the actual impact on them.\n3. State the remedy concretely, and when. Offer it; don't\n   make them ask.\n4. What changes so it doesn't recur — only if true.\n5. No excuses, no internal explanations.\nUnder 130 words. Apologise once, early, specifically.\nAsk them for nothing. Don't upsell.\nThen flag: is the remedy proportionate to the harm?\n```\n\n**Churn diagnosis**\n\nUse: on cancellation. Diagnose the cause before offering anything.\n\n```\n[CONTEXT BLOCK]\nWhat they said: [CANCELLATION MESSAGE, VERBATIM]\nHistory: [TENURE, SPEND, USAGE TREND, PAST TICKETS]\nWhat I can offer: [REAL OPTIONS]\n1. Which cause is it? (A) price (B) unmet expectation\n   (C) specific failure (D) their circumstances changed.\n   State confidence and quote the evidence.\n2. Is this saveable? If it's D, say so — a save attempt\n   wastes their time and burns the relationship.\n3. The right response for THIS cause. Don't offer a\n   discount unless the cause is genuinely price.\n4. Draft it, under 120 words. Don't guilt them. Make\n   cancelling easy regardless.\n5. What should we log about product or onboarding?\n```\n\n## Operations & hiring\n\nTemplates 18–21. For getting things out of your head and into a document.\n\n**SOP extraction interview**\n\nUse: documenting a process only you know. The one-question-at-a-time rule is essential.\n\n```\n[CONTEXT BLOCK]\nI need to document how I do [PROCESS] so someone else\ncan run it without me.\nINTERVIEW ME. Ask ONE question at a time and wait for my\nanswer. Don't produce the SOP until you have enough.\nStart by asking me to walk through the LAST TIME I did it.\nProbe for: steps I do automatically and would forget to\nmention; decisions I make and what I base them on; what\n\"done well\" means vs merely \"done\"; what goes wrong and\nhow I spot it; what I check before finishing.\nEvery time I say \"it depends\" — STOP and ask exactly what\nit depends on until the rule is explicit.\nThen produce: purpose, trigger, steps, decision rules,\nquality checks, common failures, escalation criteria.\nFinally, list every question you still have.\n```\n\n**Delegation audit**\n\nUse: quarterly, or when you’re the bottleneck.\n\n```\n[CONTEXT BLOCK]\nEverything I personally did last week: [LIST WITH ROUGH\nHOURS]\nSort every task:\nA) Only the founder can do this (relationships, judgment,\n   final decisions)\nB) Could be delegated to a person today\nC) Could be systematised or automated\nD) Shouldn't be done at all\nFor B: the cheapest role that could own it.\nFor C: describe the system in one sentence.\nFor D: what happens if it stops.\nThen: what % of my week is in A? If under 40%, which single\nchange moves the needle most?\n```\n\n**Hiring scorecard**\n\nUse: before writing any job ad. Most bad hires trace to a vague scorecard.\n\n```\n[CONTEXT BLOCK]\nRole: [TITLE] · Why now: [WHAT'S BREAKING]\nBudget: [RANGE] · Reports to: [WHO]\nBefore any job description, define success:\n1. 4-6 OUTCOMES they must deliver in 12 months. Measurable,\n   specific to my business, not generic responsibilities.\n2. For each, the ONE competency that predicts it.\n3. What evidence in someone's history demonstrates it.\n4. A written 1-3-5 rubric per competency — what a weak,\n   adequate and strong answer contains.\n5. Must-have vs nice-to-have. Challenge anything I marked\n   must-have that probably isn't.\n6. Flag if this is actually TWO roles.\n7. Flag any requirement that's a proxy for background\n   rather than ability — degrees, \"X+ years\", \"culture fit\".\n```\n\n**New-hire simulation**\n\nUse: testing an SOP before a real person follows it.\n\n```\nThe document: [PASTE YOUR SOP]\nYou're a new starter on day three. Competent, motivated,\nbut you know NOTHING about this business beyond what's\nwritten here. You can't ask anyone.\nFollow this document literally, step by step. At every\nstep tell me: what you'd do; what you'd have to GUESS;\nwhere you'd get STUCK; where you'd confidently do the\nWRONG thing because it's ambiguous but sounds clear; any\nterm you don't understand.\nBe literal to the point of pedantry. Do NOT use common\nsense to fill gaps — common sense is exactly what a new\nstarter lacks about MY business.\nThen: critical gaps (wrong result, unnoticed) · friction\ngaps · assumed knowledge · the step most likely done wrong\n· the 3 questions you'd ask on day one.\n```\n\n## Finance\n\nTemplates 22–23. Note: these never let the model calculate.\n\n**P&L reading**\n\nUse: monthly. Formulas only — model accuracy on financial calculation is unreliable.\n\n```\n[CONTEXT BLOCK]\nCurrent period: [REVENUE, COST OF SALES, EACH OVERHEAD,\nNET PROFIT]\nComparison period: [SAME LINES, 6-12 MONTHS EARLIER]\nDO NOT CALCULATE. Give me formulas and tell me what to\ncompute, then interpret what I report back.\n1. Each metric with its formula — gross margin %, net\n   margin %, each overhead as % of revenue, period change.\n2. Which lines moved most in absolute terms, and which as\n   a proportion of revenue? Different questions, both matter.\n3. If gross margin moved: what would current revenue have\n   produced at the OLD margin? That gap is usually the story.\n4. Largest overhead — what's a 10% cut worth annually?\n5. What this P&L does NOT tell me before I draw conclusions.\n6. Three questions to ask my accountant.\n```\n\n**Runway model builder**\n\nUse: get a model you can edit, not a number you can’t check.\n\n```\n[CONTEXT BLOCK]\nCash today: [AMOUNT] · Owed to me: [AMOUNTS + DATES]\nI owe: [AMOUNTS + DATES — include tax/VAT]\nMonthly revenue, last 6: [FIGURES]\nMonthly costs, fixed vs variable: [ITEMISED]\nPayment terms: [WHEN CUSTOMERS ACTUALLY PAY]\nBuild a 12-month cash flow as a SPREADSHEET STRUCTURE.\nDo not calculate.\n1. Exact rows/columns, with an ASSUMPTIONS block at top and\n   calculation rows referencing it. Never hard-code a number\n   inside a formula.\n2. Formulas cell by cell — cash received (timing-adjusted,\n   NOT revenue billed), cash out, closing cash, runway.\n3. Every assumption labelled, flagging which you inferred.\n4. Cash items founders forget — tax, annual renewals,\n   seasonality. Ask me whether each applies.\n5. The 3 inputs that most change the outcome if wrong by 20%.\nRemember: profit ≠ cash. Model when money MOVES.\n```\n\n## Quality control\n\nTemplates 24–25. The ones that improve everything else.\n\n**The adversarial critic**\n\nUse: on anything important before it ships. Never ask “is this good?”\n\n```\nAct as [THE SPECIFIC PERSON WHO'LL RECEIVE THIS — a busy\nCFO, a skeptical prospect, a journalist looking for holes].\nYou see a lot of these and dismiss most in seconds.\nHere's what you just received: [PASTE]\n1. YOUR HONEST REACTION — the actual thought as you read\n   the first line. Be blunt. If you'd stop reading, say\n   which word lost you.\n2. THE THREE OBJECTIONS you'd raise, ranked by likelihood.\n3. THE REWRITE — rewrite it so it survives your own\n   objections. Change only what's necessary.\nThen step out of character: what single change would most\nimprove the odds of this working?\n```\n\n**The template-maker**\n\nUse: whenever a prompt works well. This is how the library grows.\n\n```\nA prompt I wrote produced good output. Turn it into a\nreusable template.\nThe prompt: [PASTE]\nThe output it gave: [PASTE]\nWhat I'd want next time: [SAME THINKING, DIFFERENT\nSITUATION — describe one]\n1. SEPARATE — which parts describe this specific situation,\n   and which describe the thinking I want repeated? Only\n   the second belongs in a template.\n2. VARIABLES — replace every situation-specific detail with\n   [DESCRIPTIVE NAME]. Names must be clear enough that a\n   colleague knows what to supply.\n3. MAKE IMPLICIT THINGS EXPLICIT — what did this prompt get\n   away with because of context that won't exist next time?\n   Add: output format, length, banned phrases.\n4. ADD UNCERTAINTY HANDLING — flag inferences, NOT FOUND\n   rather than guessing.\n5. THE TEMPLATE — write it out.\n6. STRESS TEST — run your template against the different\n   situation I described. If it only works for my original\n   case, it's still a prompt. Fix it and say what you changed.\n```\n\n## How do I turn my own prompt into a template?\n\nTemplate #25 in action — a real one-off converted into something reusable.\n\n*“Write a LinkedIn post about how we helped Brightline cut their support response time from 14 hours to under 1 hour without hiring anyone. Make it sound professional but human. Not too salesy.”*\n\n**What got separated out:** “Brightline”, “14 hours to under 1”, “support response time” are *this situation*. The reusable thinking is: *turn a client result into a post that leads with the specific number and doesn’t read as a case study.*\n\n```\n[CONTEXT BLOCK]\nVoice sample: \"[2 PARAGRAPHS OF YOUR WRITING]\"\n\nClient result to write about:\n- Who: [CLIENT NAME OR \"a client in [SECTOR]\"]\n- The before state: [SPECIFIC, WITH A NUMBER]\n- The after state: [SPECIFIC, WITH A NUMBER]\n- What we did NOT have to do: [THE CONSTRAINT WE\n  WORKED WITHIN — often the most interesting part]\n- Timeframe: [HOW LONG IT TOOK]\n\nWrite a LinkedIn post, 120-180 words:\n- Open with the most specific number, not a claim about us\n- One sentence on what most people try instead, and why\n  it doesn't work\n- The result, plainly stated\n- End with a genuine question, not \"thoughts?\"\n\nBanned: \"thrilled to announce\", \"game-changing\", \"we're\nproud to\", \"reach out\", any sentence starting \"In today's\".\nDon't name our product more than once.\nOnly use facts I gave you — if you need one I didn't\nprovide, ask rather than inventing it.\n\nThen: flag the sentence most likely to read as promotional.\n```\n\n**What the conversion added:** named variables, a voice sample instead of “professional but human”, an explicit word count, a structure, a banned list replacing “not too salesy”, and an instruction against inventing facts. **The original worked once. This works every time you land a result.**\n\n## What’s deliberately missing from these templates?\n\nThree things you’ll find in most published prompt libraries, left out on purpose.\n\n| Not included | Why |\n|---|---|\n| “You are a world-class expert…” on factual templates | Research in 2026 found expert personas reduced accuracy on knowledge benchmarks — 68.0% vs 71.6%. We kept personas only in templates 8, 13 and 24, where style or role-play is the actual job. |\n| “Think step by step” | Reasoning models do this internally; instructing them can interfere. Still useful on faster tiers, so add it there if you’re using one. |\n| “This is important to my career” / tip offers | Magic phrases were trained into irrelevance. They’re noise occupying space that could carry information. |\n\nTemplates 22 and 23 also never let the model calculate — benchmarking has found native accuracy on financial calculations around 52%, collapsing further on multivariate ones. Formulas only.\n\n## Frequently asked questions\n\n### What is the difference between a prompt and a prompt template?\n\nA prompt is written for one situation and discarded. A template separates the reusable thinking from the situation-specific detail, replacing that detail with clearly marked variables so the same structure works next time. The practical test is whether you could hand it to a colleague working on a different case and have it still produce useful output.\n\n### How do I write my own prompt templates?\n\nStart from a prompt that already worked rather than writing one from scratch. Separate what described your specific situation from what described the thinking you wanted, replace the situation-specific parts with named variables, then make explicit any constraints that were previously implicit — format, length, banned phrases. Test it on a genuinely different case before saving it.\n\n### Should prompt templates include an expert persona?\n\nOnly for tasks judged on style. Research published in 2026 found expert personas reduced performance on knowledge benchmarks, with one evaluation reporting 68.0% against 71.6% without a persona. Templates for writing, tone and formatting can reasonably include a persona; templates for research, analysis and anything judged on factual accuracy should not.\n\n### How many prompt templates does a business actually need?\n\nRoughly a dozen that you genuinely rerun, not a hundred you scroll past. The value comes from repeated use of a small set tied to recurring decisions, because familiarity and refinement compound. A large unused library is a bookmark collection rather than a system.\n\n### Where should I store my prompt templates?\n\nSomewhere your team can reach them and edit them — a shared document, a notes app, or the saved prompts and projects features inside your AI tool. What matters more than location is that there’s one canonical version, so improvements accumulate rather than each person maintaining a private variant.\n\n### Do prompt templates work across ChatGPT, Claude and Gemini?\n\nYes, in general. Templates built on specific context, examples, explicit constraints and defined output format transfer well across current frontier models, because they reduce ambiguity rather than exploiting a particular model’s quirks. The main exception: instructions to reason step by step are useful on faster model tiers and largely unnecessary on reasoning models.\n\n### Why do my templates stop working over time?\n\nUsually because the context inside them went stale rather than because the template broke. Business facts, proof points, pricing and positioning change, and a template carrying outdated details quietly produces worse output. Review the context content monthly and the template structure quarterly, or whenever a major model ships.\n\n### What makes a prompt template actually reusable?\n\nFive things: clearly named variables for everything situation-specific, an explicit output format, stated constraints including what to avoid, an instruction to flag inferences rather than fill gaps silently, and a structure that describes the thinking rather than the answer. If any of these is missing, the template tends to produce inconsistent results across uses.\n\n### Download: The Full Template Library\n\nAll 25 in Notion and plain text, the context block pre-formatted, a “which template for which job” index, the template-maker, and a starter banned-phrase list you can extend.\n\n[Send me the library →](#)\n\nEnter your email and we’ll send the library plus a note whenever templates are updated. Unsubscribe anytime.\n\n**Written by the Narracomm team**\n\nNarracomm is a communications and content strategy team that helps business owners, operators, and founders use AI to produce clear, credible, high-performing work. Every template here is one we run in live client work across strategy, marketing, sales, support, operations, hiring and finance — and we revise them whenever the evidence changes, including removing expert personas from factual templates after the 2026 research. [Add specific credentials, years of experience, and a named reviewer here to strengthen E-E-A-T.]\n\n## Sources & further reading\n\n[Search Engine Journal — Where persona prompting works and when it backfires](https://www.searchenginejournal.com/research-you-are-an-expert-prompts-can-damage-factual-accuracy/570397/)[Prompting Science Report — Expert Personas Don’t Improve Factual Accuracy](https://arxiv.org/pdf/2512.05858)[The Self-Correction Illusion: LLMs Correct Others but Not Themselves](https://arxiv.org/html/2606.05976)[PAL: Program-Aided Language Models for Reliable Financial Arithmetic](https://beancount.io/bean-labs/research-logs/2026/04/23/pal-program-aided-language-models)[Prompt Engineering Guide — Chain-of-Thought prompting](https://www.promptingguide.ai/techniques/cot)\n\nLast reviewed and updated: **July 25, 2026** · Templates tested against current frontier models. Next review due within 14 days. We remove techniques from these templates when the evidence stops supporting them, and say so here.", "url": "https://wpnews.pro/news/ai-prompt-templates-you-can-steal-25-fill-in-the-blank-frameworks", "canonical_source": "https://www.narracomm.com/ai-prompt-templates-you-can-steal-25-fill-in-the-blank-frameworks/", "published_at": "2026-07-28 20:34:22+00:00", "updated_at": "2026-07-28 20:44:10.321095+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-tools", "ai-products", "generative-ai", "large-language-models"], "entities": [], "alternates": {"html": "https://wpnews.pro/news/ai-prompt-templates-you-can-steal-25-fill-in-the-blank-frameworks", "markdown": "https://wpnews.pro/news/ai-prompt-templates-you-can-steal-25-fill-in-the-blank-frameworks.md", "text": "https://wpnews.pro/news/ai-prompt-templates-you-can-steal-25-fill-in-the-blank-frameworks.txt", "jsonld": "https://wpnews.pro/news/ai-prompt-templates-you-can-steal-25-fill-in-the-blank-frameworks.jsonld"}}