21 AI Prompts for Marketing Teams, Ready to Paste

Plenty of marketing prompt lists open with "act as a world-class marketer" and stop there. A role can set the tone. It can't tell the model what your customers said on Tuesday's call.

So every prompt here has a slot for something only you have. Skip it and you'll get copy that could describe any product in your category.

How to use these prompts

Paste one prompt, fill the brackets, and read the output before you run the next one. The triple quotes mark where your pasted material starts and ends. OpenAI's prompting advice recommends that kind of separator, and it's why these prompts spell out length and format: its help center tells you to be specific about context, outcome, length, format and style.

Retyping your brand details gets old by the third prompt, so store them once.

• ChatGPT custom instructions apply to all your chats. OpenAI lists them on all plans, with a limit of 1,500 characters on Free and Go and 5,000 on Plus, Pro, Business, Enterprise and Education.
• ChatGPT projects keep related chats, files and instructions together. Project instructions apply only inside that project and override your global custom instructions.
• Claude projects give you a workspace with its own chat history, knowledge base and project instructions. Anthropic says projects are open to all users, with a maximum of five on a free account. Chats in a project don't share context unless you add the information to its knowledge base.

A good first project holds your positioning statement, three real email samples and your list of provable claims. If Claude is your main assistant, our collection of the best Claude prompts covers its documented techniques, and these 12 prompt patterns help when you want to write your own.

Which prompt for which job

Positioning prompts

Positioning comes first because every later prompt inherits it.

1. Find your positioning in customer quotes

Start here. It keeps the positioning tied to what buyers actually said.

I am positioning [product], a [one-line description] sold to [buyer role] at [type of company].
Below are [number] word-for-word quotes from sales calls, reviews and support tickets.

Quotes:
"""
[paste quotes, one per line, with the source of each]
"""

Using only these quotes:
1. List the 5 problems mentioned most, with a count for each and one quote as evidence.
2. List the alternatives customers compare us with, including spreadsheets or doing nothing.
3. Write 3 positioning statements in this form: For (buyer) who (problem), (product) is the (category) that (main benefit), unlike (alternative).
If the quotes do not support a claim, say so instead of filling the gap.

2. Compare competitor messaging

Shows which claims everyone in your category makes, so you stop leading with them.

Below is the homepage copy for [our product] and [2 to 4 competitors], each under its own label.
"""
[paste copy]
"""

Make a table with one row per company: headline promise, audience named, proof used, words repeated.
Then list the claims every company makes, and the claims nobody makes that we could support with these capabilities: [list what your product really does].
Finish with 3 positioning angles.
Use only the text above. Do not add facts about any competitor.

3. Build a message hierarchy

Turns one statement into something a writer can brief from.

Turn this positioning statement into a message hierarchy for [audience]: [statement].

Return:
- One core message under 12 words.
- 3 supporting pillars, each with proof from this list only: [features, real metrics, customers you may name].
- The 3 objections [audience] will raise, each with a one-line reply.
- 5 words to avoid because [audience] would find them vague.
Mark any pillar with no proof in my list as NEEDS PROOF.

4. Pressure-test copy as a skeptical buyer

A fast read before you show copy to real customers. It does not replace them.

You are a skeptical [buyer role] at a [type of company] that already uses [current tool].
Read this copy:
"""
[paste copy]
"""

Answer as that buyer:
1. After the first 2 lines only, what do you think this product does?
2. What are your first 3 questions?
3. What is your best reason to ignore it?
4. Rewrite the headline so you would keep reading.
Be blunt. Base every answer on the copy, and say where you are guessing.

Segmentation prompts

A chat assistant can suggest segments, but it won't keep them current. Once a rule proves useful, build it in a tool: our comparison of customer segmentation software covers the options, and this guide to the market segment description helps with the write-up.

5. Propose segments from a customer export

Gets you from a raw export to rules you can test. Strip personal details first.

Here are the columns and [number] sample rows from our customer export, with names and emails removed:
"""
[paste columns and rows]
"""

We sell [product] and want to [goal, e.g. reduce churn in the first 90 days].
Propose 4 to 6 segments. For each one give: a name, the exact rule using only these column names, why this group behaves differently, and one action marketing should take.
Do not infer gender, ethnicity, health or other sensitive traits.

6. Turn a segment into filter logic

Moves a plain-English idea into rules someone can build in your tool.

Write this segment as filter rules for [your CRM or email tool]: [plain-English description].

List each condition, the AND/OR logic and the time window.
Name each field you assume exists, and tell me how to check that it does. Do not invent menu or field names; if you are unsure, say so.
Then give 2 edge cases that would wrongly enter this segment and how to exclude them.

7. Define at-risk customers

Forces a definition you can check against people who already left.

Help me define an at-risk customer for [product].
Normal usage rhythm: [daily, weekly or monthly].
The action that shows a customer is getting value: [action].
Data: [paste usage for churned and retained accounts, or summary numbers].

Propose 3 definitions, from loose to strict. For each: the rule, who it would wrongly flag, and how I can test it against the accounts that churned last [quarter].
Recommend one.

8. Write a segment brief for creative

Gives writers and designers a brief built on evidence, not a made-up persona.

Write a one-page brief on [segment] from these interview notes and support tickets:
"""
[paste notes]
"""

Cover: what they are trying to get done, what makes them start looking, the exact words they use, their objections, what we should not say, and the one action we want next.
Label each line QUOTE or INFERENCE. Do not invent a name, age or backstory for this person.

Email prompts

Paste a real email as the voice sample. Describing your tone as "friendly but professional" tells the model almost nothing. Still picking a sender? We compare two common options in HubSpot vs Mailchimp.

9. Map an onboarding sequence

Ties each email to a behavior, so people who finish the action stop getting nudged.

Plan a [3 to 5] email onboarding sequence for [product]. New signups come from [source].
The action we want them to take: [activation action].
Our voice, from a real email:
"""
[paste a sample]
"""

For each email give: the send trigger (a behavior, not only a day number), its single goal, 3 subject lines under 45 characters, a body under 80 words and one call to action.
Add an exit rule: anyone who completes the action leaves the sequence.
Only mention features from this list: [features].

10. Write a win-back email with a real reason to return

Skips the guilt trip. Use it when something has actually changed.

Write a win-back email for customers who stopped [action] more than [number] days ago.
The most common reason they gave for leaving: [reason, from cancel surveys or tickets].
What changed since: [real product change or offer]. If nothing changed, say so and suggest a better next step than this email.

Give me 2 versions under 100 words, each with one call to action. No false urgency, and no discount unless I listed one above.

11. Generate subject lines to test

Makes you pick a test that teaches something instead of just more options.

Here is the email:
"""
[paste email]
"""

Write 10 subject lines under 50 characters, grouped by angle: outcome, curiosity, a specific number (only one that appears in the email), question, plain statement.
Flag any that could read as misleading.
Then choose 2 for an A/B test that differ on one variable, and say what each result would teach us about [audience].

12. Cut an email without changing its facts

The editing pass that gets skipped when you are in a hurry.

Edit this email:
"""
[paste email]
"""

Rules: cut at least 30 percent. The first sentence must mention the reader's situation. Keep one call to action. Keep every fact, price and date exactly as written.
Match the tone of this sample: [paste sample].
After the edit, list what you removed, and separately list any claim that has no evidence in the email.

Ad prompts

Ads are where invented claims get expensive, so each prompt below limits the model to proof you supply.

13. Write search ads from your search terms report

Builds copy from what people typed. Check the platform's current character limits first.

Here is our search terms report for [campaign], with search term, clicks, conversions and cost:
"""
[paste rows]
"""

Group the terms into 3 to 5 themes.
For each theme write 5 headlines of up to [headline limit] characters and 2 descriptions of up to [description limit] characters. Use the words searchers used.
List terms to add as negatives, with the reason.
Only claim what this list supports: [offers, features, guarantees].

14. Draft paid social concepts

Five angles to choose from, each with its proof source named so legal can check it.

Write ad concepts for [audience] on [channel], promoting [offer].

Give 5 concepts, one per angle: pain, outcome, objection, old way versus new way, proof.
For each: a first line under 12 words, primary text under 40 words, a visual direction, and which item from my proof list it relies on.
Proof list: [real numbers, reviews you may quote, customers you may name].
If an angle has no support in the list, write a version with no claim and say so.

15. Read an ad performance export

Works out the metrics first, and refuses to crown a winner on thin data.

Here is our ad export, with ad name, spend, impressions, clicks, conversions and revenue:
"""
[paste rows]
"""

Make a table with click-through rate, cost per conversion and return on ad spend for each ad, and show one worked calculation.
Leave ads under [number] conversions out of your conclusions and say why.
Then tell me what the top ads share, write one hypothesis for the next test, and name what this data cannot tell us.

Landing page prompts

These three fix a page you already have. To generate a page from scratch with an AI builder, use our landing page prompts instead.

16. Check message match from ad to page

Run this before you blame the traffic for a weak page.

Ad copy:
"""
[paste ad]
"""
Landing page copy:
"""
[paste page]
"""

Compare them. Does the page repeat the promise the ad made in its first 2 lines? Is the audience the same? Is the offer the same?
List each mismatch, then rewrite only the hero headline, subheadline and button label to fix it.
Do not add claims that are in neither text.

17. Find the objections a page leaves open

Sales already knows what stops people. This puts it on the page.

Here is our landing page copy and a list of objections from sales calls:
"""
[paste page]
"""
"""
[paste objections, most common first]
"""

For each objection, say whether the page answers it, where, and how well.
For each gap, write a short FAQ answer or a line of body copy, and say where it belongs.
If answering honestly would need a fact I have not given you, write NEEDS FACT.

18. Rewrite a hero for one segment

For a segment big enough to deserve its own page or variant.

Rewrite this hero section for [segment]. They arrive from [source] and care most about [priority].
Current hero:
"""
[paste hero]
"""

Give 3 options, each with a headline under 10 words, a subheadline under 25 words and a button label that starts with a verb.
Use words from these customer quotes where you can: [paste quotes].
For each option, say which objection it handles.

Reporting prompts

Models can make arithmetic mistakes and offer tidy explanations for noise. That's why these prompts ask for the calculation and label guesses. Recheck the math in a spreadsheet, and pull the source numbers from one of your marketing analytics tools rather than from memory.

19. Turn raw numbers into a weekly update

Keeps the update short and separates what happened from what you think caused it.

Here are this week's numbers by channel, last week's and the 4-week average:
"""
[paste table]
"""

Write an update for [reader, e.g. the CEO] in under 150 words: what changed, likely reasons, and what we do next.
Label every reason as a hypothesis unless the data proves it.
Below the update, add a table with each percentage change and the calculation behind it.
Do not call anything a trend if it rests on fewer than [number] data points.

20. Investigate a drop in one metric

Stops the team from guessing out loud in the meeting.

[Metric] fell from [number] to [number] between [date] and [date].
Here is the breakdown by channel, device, country and landing page:
"""
[paste data]
"""

Calculate which rows account for most of the change.
Rank 5 possible causes, including tracking problems. For each one, say what evidence would confirm it or rule it out, and where I should look.
Keep what the data shows separate from what you are guessing.

21. Keep an experiment log

Makes last quarter's tests useful to whoever runs the next one.

Here are the tests we ran this [month or quarter], with notes and results:
"""
[paste notes]
"""

Turn them into a log with one row per test: hypothesis, audience, sample size, result, decision and what to test next.
Flag any test that ended early, changed more than one variable or had too little data to trust.
Write 3 lessons we can act on, using no flagged test as the only evidence.

Common mistakes with marketing prompts

• No source material. With nothing to work from, the model writes from general patterns, and your copy ends up sounding like everyone else's.
• Trusting confident numbers. A statistic in the output needs a source you can open. If you can't find one, delete it.
• Asking for a whole campaign at once. Run positioning, then the message, then the assets, and read each result.
• Pasting personal data. Remove names and emails first, and check your plan's settings. OpenAI's help center explains how to opt out of having your content used to improve its models.
• Saying only what to avoid. OpenAI advises stating what to do instead. "Use a 3 to 5 sentence paragraph" beats "fairly short."

Prompts also need someone to run them every week. If you're a SaaS founder with no marketer, that may be the real bottleneck. Questera, which publishes this blog, describes its product as an AI marketing team for SaaS that writes, posts, makes videos, runs ads and reports back each morning. As of November 2026, Starter is free, Growth is $49 a month and Scale is $199 a month.

Frequently Asked Questions

What are the best AI prompts for marketing?
The ones that hand the model your own material and ask for a fixed output. A good marketing prompt names the customer, includes real inputs such as call quotes or exports, sets the format and length, and says what to do when the data does not support a claim.

Do these prompts work as ChatGPT prompts for marketing?
Yes. They are plain text and use no feature that belongs to one assistant, so you can paste them into ChatGPT, Claude or another chat assistant. Replace everything in square brackets first.

How do I give AI my brand voice?
Paste real samples rather than adjectives, as prompts 9 and 12 do. For repeat use, add those samples to a Claude or ChatGPT project.

Is it safe to paste customer data into an AI chat?
Remove names, email addresses and anything else that identifies a person before you paste, and follow your company's data policy. Check your plan's data settings too: OpenAI's help center says content can be used to improve its models and explains how to opt out.

Can AI prompts replace a marketer?
No. They shorten the drafting and first-pass analysis. Someone still has to own the customer research, choose the angle, check every number and claim, and decide what ships.