ChatGPT prompts for marketers work best when they are built around a job, not a personality. A prompt that says “act as a world class copywriter” returns generic text. A prompt that hands over your landing page, audience, character limits and banned claims returns a draft you can ship.
Below are 25 prompts grouped by the ten jobs marketers and creators do weekly, each with a note on when to use it and what a good answer looks like. For the thinking behind the patterns, read our guide to prompt engineering for marketers.
Key takeaways
- Context beats clever wording: what you paste in matters more than the role you assign.
- Every prompt names a task, supplies context and fixes an output format.
- Strip names, emails and account IDs before pasting any customer record.
- Keep 10 to 15 prompts used weekly, not a library of 200 nobody opens.
- Judge output by how much rewriting it needs, not by how confident it sounds.
What makes a good ChatGPT prompt for marketers?
A good marketing prompt does three things: it names one specific job, it supplies the real material that job needs, and it fixes the output format including limits. Persona lines have little effect. If you can delete a sentence and the answer does not change, delete it.
Treat a prompt like a brief for a freelancer. You would not tell a copywriter to be brilliant. You would send the page, the audience, the offer, the character counts and the banned words. OpenAI says models “benefit from precise instructions that explicitly provide the logic and data required to complete the task in the prompt” (OpenAI prompt engineering guide).
Why does context beat clever wording?
Because a model can only recombine what it has. Without your data it falls back on the average of everything it has read, which is why vague prompts produce copy that sounds like every other brand. Adding 300 words of real customer language changes the output more than any rewording.
Test it. Ask for five headlines for a project management tool and you get five versions of the same tired promise. Paste 20 verbatim review sentences first and tell the model to reuse that vocabulary. The headlines start using words your buyers say.
Three kinds of context carry the weight: source material, constraints such as character limits and banned claims, and judgement rules set before the answer is written. Save that context once. ChatGPT Projects attach files and custom instructions to a workspace, on every plan including Free, with file limits rising on paid tiers (OpenAI help centre).
The 25 ChatGPT prompts, grouped by job
These 25 prompts cover ten recurring jobs: audience research, positioning, ad copy, landing pages, SEO and GEO, email, analytics, reporting, competitor teardown and creative briefs. Replace the bracketed placeholders, and paste your source material below the instruction.
| Job | Prompts | Context to paste in |
|---|---|---|
| Audience research | 1 to 3 | Reviews, tickets, call notes |
| Positioning | 4 to 5 | Positioning doc, proof points |
| Ad copy | 6 to 8 | Landing page, limits, banned claims |
| Landing pages | 9 to 10 | Traffic source, offer, live page |
| SEO and GEO | 11 to 13 | Site search logs, real questions |
| 14 to 16 | Product flow, segment, body copy | |
| Analytics and reporting | 17 to 20 | Metric list, the actual data table |
| Competitor teardown | 21 to 22 | Competitor pages and ads |
| Creative briefs | 23 to 25 | Objective, budget, channels |
Audience research prompts
1. Customer language miner
Here are 40 reviews and support tickets for [product]. Group the language into five themes. For each, give the three most repeated phrases word for word and the emotion behind them. No paraphrasing.
Use when writing for a product you did not build. Good: Verbatim phrases, not marketing words.
2. Buying committee map
We sell [product] to [company type] at [price band]. List everyone who touches the decision: title, what each cares about, the question that stalls the deal, the proof that answers it.
Use when purchases with more than one sign off. Good: A role you forgot to market to.
3. Objection ladder
List the ten objections a buyer raises about [product] at [price], earliest to latest. For each: their exact words, the shortest honest answer, and where it belongs (ad, page, email, call).
Use when conversion drops between two steps. Good: Objections mapped onto real page sections.
Positioning and messaging prompts
4. Positioning stress test
Here is our positioning statement: [paste]. Which claims could a competitor copy today? Which are true but unprovable? Which would a sceptical buyer ignore? Rewrite so every claim is provable or dropped.
Use when a rebrand or a pitch deck. Good: A shorter statement, proof behind each claim.
5. Message hierarchy
Build a message hierarchy for [product] aimed at [audience]. One promise under 12 words, three supporting messages, two proof points each. Flag any message that needs existing trust to work.
Use when five people describe the product five ways. Good: A promise that survives a follow up question.
Ad copy prompts
6. Search ad variant set
Write 12 search ad headlines (max 30 characters) and 4 descriptions (max 90 characters) for [product], keyword [keyword], page below. Half problem led, half outcome led. No superlatives. Table with character counts.
Use when building or refreshing a search campaign. Good: Every line inside the limit.
7. Scroll stopping hooks
Write 15 opening lines for short form video ads for [product]. Five state a number, five name a myth, five open a loop. Under 9 words, under 3 seconds spoken. Rank them.
Use when the weekly creative refresh. Good: Hooks that work with the sound off.
8. Claim safety rewrite
Rewrite this copy so no claim implies a guaranteed result, a medical benefit or a comparison we cannot substantiate: [paste]. Keep length and rhythm. List every change and the risk it removes.
Use when finance, health, supplements, education outcomes. Good: A change log you can send to legal.
Landing page prompts
9. Landing page skeleton
Design a landing page for [offer], audience [audience], arriving from [traffic source]. Give the section order, the job of each section in one sentence, and what must be visible before a scroll.
Use when any new page, especially a mismatched source. Good: An order justified by buyer state.
10. Above the fold rewrite
Here is the top of our page: [paste headline, subhead, button]. Rewrite five ways: outcome led, problem led, number led, audience led, next step led. Which does a first time visitor grasp in two seconds?
Use when bounce rate is high and dwell time low. Good: Five angles, not five rewordings.
SEO and GEO prompts
Generative engine optimisation means writing pages an AI engine can quote accurately. Google says there are “no additional technical requirements” to appear in AI Overviews beyond being indexed and snippet eligible (Google Search Central). Score your pages with the free GEO readiness checker.
11. Intent cluster builder
For [topic], list 25 queries real people type, grouped by intent: definition, how to, comparison, cost, problem, buying. Per group, say whether it needs its own page or a section, and why.
Use when planning a content hub, not one post. Good: A plan showing which queries would compete.
12. Extractable answer writer
Write a 45 word answer to [question] that is true on its own with no surrounding context. Plain English, no brand name in the first sentence, one concrete number or rule.
Use when writing for AI answer engines. Good: A block you could lift out that still stands.
13. Real question harvest
Here are 60 questions from site search, sales calls and comments: [paste]. Deduplicate into 8, keeping people's phrasing. Answer each in 50 to 70 words. Mark any where we are not the right fit.
Use when building an FAQ that earns quotes. Good: One answer that sends the wrong buyer away.
Email prompts
14. Lifecycle sequence outline
Map a 6 email onboarding sequence for [product]. Per email: its one job, trigger, send delay, subject line, single call to action, and the behaviour that skips the reader ahead.
Use when a new trial flow or weak activation. Good: A sequence where deleting one email breaks something.
15. Subject line test set
Write 10 subject lines for this email: [paste body]. Two curiosity, two benefit, two number led, two literal, two referencing what the reader just did. Under 42 characters. Add preview text that adds information.
Use when any send big enough to split test. Good: Preview text that earns its space.
16. Win back email
Write a re-engagement email for people who bought [product] once and have not returned in 6 months. Give one reason to come back that is new information, not a discount. Under 120 words.
Use when quarterly list hygiene. Good: An email you would not mind receiving.
Analytics prompts
Assistants explain what a number could mean but cannot know which number is real. Define metrics first, as our guide to AI marketing analytics sets out.
17. Metric definition check
Here is our dashboard field list: [paste]. Per metric, give the exact formula, the window it uses, what it double counts, and one decision it should never inform.
Use when two tools disagree about the same number. Good: One definition that surprises the team.
18. Anomaly explanation
Conversion fell from 3.1 percent to 2.2 percent between these weeks. Channel, device, geography and page split: [paste]. List five likely causes by probability and the check that confirms or kills each.
Use when Monday morning, when a number moved. Good: Checks you can run in an hour.
Reporting prompts
19. Report narrative
Turn this month of campaign data into a one page narrative for a marketing director: [paste table]. What happened, why, what we changed, what is next, what we need from you. Every claim cites a number.
Use when reporting that reads like a data dump. Good: Every sentence points at a cell in the table.
20. Executive one pager
Compress the report below into 150 words for a chief executive reading on a phone: [paste]. Lead with the decision you need, then the result, the cost and the one risk.
Use when a board update or budget talk. Good: A first line that states the ask.
Competitor teardown prompts
21. Competitor messaging teardown
Here is a competitor home page, pricing page and three ads: [paste]. Identify their core promise, who they exclude, the objection they answer first, the proof they use, and the gap they leave open.
Use when the quarterly competitive review. Good: A gap tied to their product, not an adjective.
22. Share of model check
For the buying question [question], list the brands you would name in an answer, in order, and the evidence behind each. Then say what must exist on the open web for [our brand] to be named.
Use when checking how AI engines describe your category. Good: A list of missing evidence types.
Creative brief prompts
23. One page creative brief
Write a creative brief for [campaign]: business problem, audience in one sentence, single minded proposition, reason to believe, mandatories, success with a number, and what we will not do. Ask three questions first if anything is missing.
Use when briefing an agency, freelancer or design team. Good: A brief that asks what you were avoiding.
24. Video script brief
Write a 30 second script brief for [product] on [platform]: hook, shot list, on screen text per beat, voice over, caption, end card. Then a 15 second cut down.
Use when planning a shoot or a batch of shorts. Good: A shot list a phone camera could produce.
25. Asset list and variants
For campaign [name] on [channels], list every asset needed: format, aspect ratio, duration, copy limits, and the variant axis it tests. Group by what one shoot can cover.
Use when production planning, so you shoot once. Good: A grouped list, one shoot covering the plan.
What are the most common ChatGPT prompt mistakes?
The common mistakes are asking for a persona instead of a job, giving no source material, asking for ten things at once, accepting the first answer, trusting numbers the model states, and never saving what worked. Every fix takes under a minute.
| Mistake | What you get | Fix |
|---|---|---|
| Persona instead of job | Confident generic copy | Name the deliverable and its limits |
| No source material | Category cliches | Paste real reviews, pages or data |
| Ten asks in one prompt | Everything done badly | One deliverable per prompt, then chain |
| Accepting draft one | The blandest safe option | Ask it to critique itself, then redo |
| Trusting stated numbers | Plausible fiction in a report | Allow only numbers you pasted in |
| Never saving winners | Starting from zero weekly | Keep 10 to 15 prompts in one workspace |
How much time do good prompts actually save?
Enough to notice, but only if the prompts stay saved with their context. The saving comes from skipping the blank page and the rebrief, not from the model writing faster. Measure it by timing the edit on one task for two weeks, before and after, rather than counting output.
Example, with figures to replace with your own. A two person team ships 8 ad variants, 2 landing page drafts and 1 report a month. Say an ad variant takes 25 minutes by hand against 10 with a good prompt, a landing page draft 3 hours against 1, the report 4 against 1.5.
Formula: Hours saved = (manual time minus assisted time) x volume
- Ads: (25 minus 10) x 8 = 120 minutes, or 2 hours.
- Landing pages: (3 minus 1) x 2 = 4 hours.
- Report: (4 minus 1.5) x 1 = 2.5 hours.
- Total: 8.5 hours a month, about one working day.
The figures illustrate the method, not a benchmark.
What this means for marketers and creators
What this means for marketers and creators
For marketers: pick the three prompts nearest your weekly bottleneck, usually ad copy, reporting and objection mapping. Build one shared workspace holding your positioning document, your last 50 reviews and your banned claims list. Set one reporting rule: no number appears unless it was pasted in. Measure editing time, not output volume, as our guide to AI personalisation in marketing also argues.
For creators: your advantage is raw material nobody else has: your comments, your DMs, your own recordings. Feed those in instead of asking for generic hooks. Run the hook prompt weekly, keep a file of lines that performed, and paste it back as examples. The model drafts, you decide.
Frequently asked questions
What are the best ChatGPT prompts for marketers?
The best prompts name one job and attach your own context. An ad copy prompt should include the landing page, the audience, the offer and the character limits. A reporting prompt should include the actual data table. Clever phrasing matters far less than your material.
Do these prompts work on Gemini and Claude?
Mostly yes. Each one names a task, supplies context and defines an output format, which every general assistant understands. Adjust length, since models differ in how much they pad. Run one prompt on two assistants and keep whichever needs less editing.
How long should a marketing prompt be?
As long as the context requires. A short task line, your pasted material, then output rules. Prompts under 20 words return generic text because the model has nothing specific to work with. Long instructions with no real data are worse, not better.
Should you paste customer data into ChatGPT?
Not personal data. Strip names, emails, phone numbers and account identifiers before pasting reviews, tickets or call notes. On consumer plans, check the data controls setting that decides whether your chats improve models. Business terms differ, so check your account.
How do you stop ChatGPT sounding like ChatGPT?
Give it real customer language to copy and a banned word list. Paste 20 verbatim review phrases and tell it to reuse that vocabulary. Add a rule such as no adjectives without a number. Generic output comes from prompts with no source material.
Next steps: pick three prompts, run them on a live task this week and time the edit, not the output. Then browse the free TechMachaw marketing tools, and if a campaign is next, our guide to building a digital media plan gives you the structure. Join the free TechMachaw newsletter for weekly prompt and tool tests.
