ChatGPT Prompt Templates: 25 Free Prompt Engineering Templates

Look at enough well-built prompts and the same anatomy keeps showing up. The labels differ from guide to guide, but the parts don't.

OpenAI's own prompt engineering best practices say much the same thing. Put instructions first and separate them from the context, then be specific about outcome, length, format and style. Of the four parts, format is the one people skip most and the one that changes results the most. End with "return it as a table with columns for channel, hook and effort." It beats "give me some ideas" every time.

You'll sometimes see vendors claim structured, role-based prompts produce "3x better output." Treat that number as marketing. What holds up in practice is simpler: the more of the four parts you specify, the less the model has to guess.

If you remember nothing else, remember this prompt engineering template. It generates a usable prompt for almost any task:

You are a [ROLE]. [TASK] about [CONTEXT]. Return it as [FORMAT], keeping it [CONSTRAINT].

For deeper, framework-based structures such as RTF, CO-STAR and RISEN, see our guide to prompt frameworks.

The everyday workhorses: turning raw material into structure, drafting, editing and headlines.

Use this when you already have raw material and need a clean structure before writing. Call notes, transcripts, research and messy drafts all work. It's the template people search for most on this site, so it's first.

You are an experienced content editor. Below is my source material for an article about [TOPIC] for [AUDIENCE].

Turn it into a detailed outline:
- An H1 title under 60 characters
- 6-8 H2 sections, each with 2-4 H3 points
- Under each heading, note which part of the source material supports it
- A "Gaps" list of anything the outline needs that the source material does not cover
- A FAQ section with 5 questions the source material can answer

Use only the source material. Do not add facts, statistics or examples that are not in it. If a section has no support in the material, flag it instead of filling it.

Source material:
"""
[PASTE NOTES, TRANSCRIPT OR DRAFT]
"""

Get this right: paste the material inside the triple quotes. Separating instructions from context is one of the most reliable ways to stop the model blending the two.

You are an experienced content writer. Create a detailed blog post outline for an article about [TOPIC] targeting [AUDIENCE]. The article angle is: [WHAT MAKES THIS DIFFERENT FROM GENERIC COVERAGE]. Include: H1 title, 6-8 H2 sections with a brief note on each, a FAQ section with 5 questions, and a recommended word count range. Audience knowledge level: [BEGINNER / INTERMEDIATE / ADVANCED]. Tone: [DESCRIBE TONE].

Get this right: the angle. Without it you get the same outline as every other page on the topic.

Edit the text below for clarity and conciseness. Cut it by [PERCENTAGE, e.g. 30%], remove filler words, and keep every fact and the original meaning. Use plain, professional English. Return the edited text first, then a short list of the main changes you made.

Text:
"""
[PASTE TEXT]
"""

Get this right: the target length. "Shorter" is vague; "30% shorter" isn't.

Give me 10 headline options for [CONTENT TYPE] about [TOPIC] for [AUDIENCE]. Mix styles: curiosity, direct benefit and number-led. Keep each under 60 characters. Then pick your best 3 and explain each choice in one line.

You are an email newsletter writer. Write one issue about [TOPIC] for [AUDIENCE].
- Subject line under 50 characters, plus preview text
- Open with a short story or insight, not a product push
- No more than 3 sections, short paragraphs, conversational first-person tone
- One primary call to action: [CTA]
Do not open with "I hope this email finds you well."

Need subject line options to test? Run the topic through our email subject line generator.

Briefs, pages and ads: give the model intent and audience and it stops writing for everyone.

Act as an SEO content strategist. Create a content brief for an article targeting the keyword "[KEYWORD]".
Search intent: [INFORMATIONAL / COMMERCIAL / TRANSACTIONAL]
Audience: [DESCRIBE]
Include: an SEO title under 60 characters, a meta description of 150-160 characters, 5-8 secondary keywords, 8-10 H2 headings with H3 subheadings, the key points under each, internal linking opportunities, a FAQ section with 5-8 questions, a recommended word count, and one angle that differentiates the piece from the current top 3 results.

Get this right: search intent. An informational brief and a commercial brief for the same keyword look completely different.

You are a conversion copywriter. Draft above-the-fold landing page copy for [PRODUCT] aimed at [AUDIENCE]. Primary benefit: [MAIN OUTCOME].
Return: a headline that leads with the outcome (not the feature), a subhead, 3 benefit bullets, one line that handles the biggest objection ([OBJECTION]), and a CTA.
Use the words this audience actually uses. Avoid buzzwords like "revolutionary," "game-changing" and "best-in-class."

Write 5 ad variations for [PRODUCT] aimed at [AUDIENCE] on [PLATFORM]. Give each a different angle: pain, benefit, social proof, curiosity, urgency. Label the angle. Respect platform limits: [e.g. Google RSA headlines 30 characters, descriptions 90 characters; Facebook hook inside the first 125 characters, headline under 40].

You are a marketing researcher. Build a realistic customer persona for [PRODUCT OR SERVICE] in [TARGET MARKET]. Business goal: [GOAL].
Give the persona a name, role and short backstory, then cover: goals, pain points, buying motivations, where they spend time online, how they discover products, the top 5 objections they'd raise before buying, and the exact phrases they'd use to describe their problem.
Skip demographic details that wouldn't change how we market to them.

These are the ai prompt templates for business owners and operators use to pressure-test an idea, not admire it. Unlike most business prompt templates, they're built to find what's wrong, not to flatter the plan.

Act as a skeptical investor. Here is my business idea: [IDEA].
Assume we launched it today and it failed completely 12 months from now. List the 3 most likely market or structural reasons it died. For each, give one cheap test I could run this week and one pivot if the test fails. Finish with the single assumption the whole idea depends on most.

Act as a pricing strategist. Product: [PRODUCT]. Cost per unit: [COST]. Target margin: [MARGIN]. Competitor price range: [RANGE]. Customer: [DESCRIBE].
Compare cost-plus and value-based pricing, then propose 3 tiers (or a free vs premium split with clear feature gates). Give a one-line rationale for each tier and say which one you'd test first.

Write a one-page decision brief on [DECISION] for [EXECUTIVE / TEAM / STAKEHOLDER] readers.
Structure: Background, Options Considered (2-4), Recommendation, Risks, Next Steps. Keep each section under 80 words. Flag any assumption you had to make.

Built with B2B marketers and SDRs in mind, but they work for anyone writing to a stranger who owes them nothing.

Write a cold email to a [PROSPECT ROLE] at a [INDUSTRY] company about [VALUE WE OFFER]. Personalise the opening line to this trigger: [TRIGGER, e.g. a new funding round or job post]. Focus the subject line on [PAIN POINT]. Body: under 90 words, no more than 3 short paragraphs. One low-friction CTA: [SPECIFIC ASK]. Tone: direct, helpful, not salesy.

Get this right: the trigger. A real reason for writing now is the difference between a reply and a delete. For a full sequence, add: "Then write 2 follow-ups: day 3 adds value, day 7 is a polite breakup email."

A prospect said: "[OBJECTION TEXT]".
Give me 3 ways to respond, each under 50 words. Each should acknowledge the concern, reframe it, and pivot to [KEY DIFFERENTIATOR]. Consultative, not pushy. End each with a question that moves the conversation forward.

Replies that acknowledge the real issue and set a clear timeline, without promising fixes you can't deliver.

You are a customer support agent for [COMPANY]. A customer submitted this ticket:
"""
[TICKET TEXT]
"""
Write a reply that acknowledges the specific issue, explains what happens next, and sets a clear timeline. Tone: [FORMAL / FRIENDLY / CONCISE]. Under 120 words. Do not promise refunds or fixes that aren't listed here: [WHAT WE CAN OFFER].

Variation (escalation handoff): "Summarise this ticket thread in 3 bullets for a tier-2 agent: the core issue, what has been tried, and the customer's current sentiment."

Turn messy notes into decisions, owners and updates, using only what was actually said.

Turn the meeting notes below into:
1. A 3-sentence summary
2. Key decisions
3. An action-item table with columns: task, owner, due date ("not stated" if missing)
4. Open questions
Do not add anything that isn't in the notes. Leave out small talk.

Meeting type: [STANDUP / CLIENT CALL / PLANNING]
Notes:
"""
[PASTE NOTES OR TRANSCRIPT]
"""

Write a weekly status update for [PROJECT] for [AUDIENCE]. Three sections: Completed, In progress, Blockers (with what's needed to unblock each). Under 200 words. Use only these notes:
"""
[PASTE NOTES]
"""

Grounded summaries and stress-tests that cite what they use and flag what they can't support.

I'm pasting [NUMBER] sources on [TOPIC]. Using only these sources:
1. Extract the most relevant direct quotes, numbered.
2. Summarise where the sources agree and where they conflict, citing quote numbers.
3. List gaps the sources don't cover.
4. Give a balanced one-paragraph takeaway.
If the sources don't support a claim, say "not supported by the sources" rather than guessing.

Sources:
"""
[PASTE SOURCES]
"""

This one is built to cut hallucinations. More on why it works in the hallucinations section below.

Here is my plan / thesis: [PASTE]. My stance: [STANCE].
Argue against it as a smart, fair skeptic would. Give the 3 strongest objections, the weakest assumption in my logic, and what would have to be true for each objection to be wrong. Suggest one source or data type I should check.

Find root causes and scaffold agents, with strict output shapes so the result drops into your workflow.

You are a senior [LANGUAGE] engineer. This code should [EXPECTED BEHAVIOUR] but instead [ACTUAL BEHAVIOUR]. Error message: [ERROR].
Find the root cause, not just the symptom. Explain it in plain English, give a minimal fix (do not rewrite the whole file), and write one regression test that would have caught it.

Code:
"""
[PASTE CODE]
"""

The most-asked-for ai agent prompt template is one that builds other prompts. This produces a single-purpose agent prompt with a strict output schema:

Build a system prompt for an AI agent whose only job is to [TASK]. The prompt must include:
- A system role and scope (what the agent must refuse or hand off)
- The input data format it will receive
- Step-by-step reasoning instructions it should follow before answering
- A strict JSON output schema with field names, types and an example
- What to output when the input is missing or ambiguous

For deeper agent and system-level patterns, a dedicated guide to system prompt templates is coming to this hub.

A review aid, not legal advice: it quotes clauses first so the model can't paraphrase them into something they don't say.

You are assisting a [LAWYER / PARALEGAL / FOUNDER] reviewing a [CONTRACT TYPE] from the perspective of [PARTY].
1. Extract the exact text of every clause about [TOPIC, e.g. termination, liability caps, IP ownership], numbered.
2. For each, explain in plain English what it obliges our side to do, and flag anything unusual or one-sided.
3. List questions to raise with counsel.
Base your analysis only on the quoted clauses. If a clause isn't in the document, say so. This is a review aid, not legal advice.

Contract:
"""
[PASTE CONTRACT]
"""

Get this right: never skip step 1. Making the model quote first keeps it from paraphrasing a clause into something it doesn't say.

Tutoring that tests you instead of lecturing, then hands you flashcards on what you covered.

Act as a patient tutor using the Feynman Technique. Teach me [TOPIC] at a [BEGINNER / INTERMEDIATE / ADVANCED] level.
Start with a 2-paragraph overview using one everyday analogy. Then ask me one question to test my understanding and wait for my answer before going deeper. After three rounds, give me 5 flashcards (question and a two-sentence answer) on what we covered.

Tailor a resume and cover letter to one job, truthfully, with a real hook instead of "I am writing to apply."

You are a career coach. Rewrite my resume bullets to match the job description below. Use strong action verbs and quantified results, and keep everything truthful: do not invent numbers. Then write a cover letter under 300 words that opens with a specific hook (not "I am writing to apply"), connects my experience to what [COMPANY] needs, and ends with a clear next step.

My bullets: [PASTE]
Job description: [PASTE]

These double as creative prompt templates for writers who are stuck on the blank page.

Act as a story developer. Genre: [GENRE]. Setting: [TIME AND PLACE]. Tone: [DARK / WHIMSICAL / DRAMATIC]. Premise: [ONE LINE].
Give me: 3 opening paragraphs in different tones (under 80 words each), a protagonist with a goal, a flaw and a secret, 3 rules of this world that create conflict, and 5 what-if twists.

Variation (social): "Turn the piece below into a LinkedIn post, a 5-post thread and a short caption. Open each with a hook in the first line."

I don't publish a template until it has survived real use. These 25 came out of a working library of around 40-50 content templates.

ChatGPT was my main test bed for marketing and blog work. I also ran the templates in Claude for long-form outlines and Gemini for research-heavy tasks. Then I spot-checked open-weight models such as Qwen and GLM, to make sure nothing only worked on one provider.

Each template ran on 5-10 different briefs, covering different topics, niches and difficulty levels. Each brief then got 2-3 rounds of tweaks to check the output stayed consistent. I also keep a small test set of five briefs, including one edge case and one "trap" brief with conflicting instructions. A change to a template only stays if it still passes all five.

Testing changed the templates. These fixes made the biggest difference:

• A "this is not" line. Adding "This is not a motivational post; don't use inspirational quotes" cut fluffy intros sharply.
• Locked section labels. Spelling out the exact order, like "H2: Rule #1 - [name]", reduced drift between runs.
• Stated assumptions. "If information is missing, state your assumption before writing" stopped models from inventing stats and case studies.
• Shorter roles. Long, flowery role paragraphs made output less consistent. One crisp line, with the detail moved into the context, worked better.

A real before and after. My old prompt was "Write a blog post about time management for remote workers." It gave me generic advice, a fluffy intro and no clear angle.

The template version gave the model a one-line role: a senior content writer for a SaaS blog. It named the reader's real problem, which was meeting overload and context switching. It set a specific title ("7 Time-Management Rules Remote Teams Actually Follow") and a specific audience: remote team leads at 10-200 person companies. It listed must-haves, phrases to avoid, and an exact structure of H1, a short intro, seven H2s and a CTA. The drafts came back 70-80% publish-ready, instead of needing a full rewrite.