50 ChatGPT Prompts for Digital Marketing Ideas

Most marketing prompt lists are content generators. Write me ten headlines, write me a landing page, write me thirty social posts. That is the easiest part of the job, and the part that already has a dedicated article for every channel.

These fifty are organised by the digital marketing workflow instead: research and positioning, campaign planning, producing the assets, testing, and measurement. The research and measurement groups are the ones worth having, because they produce decisions rather than drafts, and almost nobody writes prompts for them.

Each prompt names a role or a task, at least one constraint, and the format to return. For prompting technique across every other kind of work, see our main ChatGPT prompts guide. Everything here works in ChatGPT and in any comparable model.

• Ask for the decision, not the copy. A prompt that returns eight ranked options with a stated trade-off is worth more than one that returns finished text you then have to judge.
• Give it your actual constraints. Budget, sales cycle, team size, the thing you cannot change. Generic input produces generic output, and marketing is almost entirely context.
• Specify the output format. A table with named columns, under 400 words, six variants labelled by angle. Anything unspecified comes back as prose you will reformat by hand.
• Ask what it is assuming. The most useful follow-up in this entire article is: which of your recommendations depends on an assumption I have not confirmed?

A request looks like this:

Write me some ad copy for my product.

The same ask as a prompt:

Write six ad variants for [product], each leading with a different angle: problem, outcome, price, speed, social proof, and risk reversal. Keep them the same length and label the angle on each so I can test them separately. Do not use the words innovative or seamless.

Same request, completely different return. The second one goes straight into a test. The first one you have to sort through first.

The group most people skip, and the one that changes the campaign. Every prompt here produces something you can argue with rather than something you can publish. For brand-level work specifically, see our ChatGPT prompts for branding.

1. Audience objection map

Act as a sceptical buyer for [product] in [market]. List the eight objections you would raise before purchasing, ranked by how likely each is to stop the sale, and for each say whether it is a product problem or a communication problem.

2. Competitor message audit

Here is copy from three competitors in [category]. Produce a table with columns for company, the claim they lead with, who that claim excludes, and the claim nobody in this set is making.

3. The category's tired language

List the fifteen phrases that appear on almost every website in [industry]. Then tell me which three are load-bearing and which twelve I should ban from our copy entirely.

4. Positioning in one sentence

Here is what we do and who for. Write our positioning as one sentence in the form: for [audience] who [situation], we are the [category] that [differentiator]. Then write three alternatives with a different differentiator each, and say what each one commits us to.

5. Jobs to be done

From the description below, list the four jobs a customer is actually hiring this product to do, ranked by frequency. Distinguish functional from emotional jobs, and flag any we currently market to that nobody is hiring us for.

6. Where they actually look

For [audience], list the eight places they genuinely look for information about [problem], excluding anywhere they would only go once. For each, say what format performs there and what gets ignored.

7. Assumption audit of a brief

Here is our marketing brief. List every assumption it depends on, ordered from most to least load-bearing. For the top three, say what evidence would confirm or kill each, and roughly how long it would take to get.

8. The uncomfortable questions

Read the positioning below and ask me the five questions a good strategist would ask that I have clearly not thought about. Do not answer them.

9. Voice specification

Here are three pieces of our existing copy. Analyse the voice on four axes: sentence length, formality, use of humour, and how much it assumes the reader knows. Then write a one-page voice guide with three do and three do-not examples taken from the samples.

10. Segment differences

Split [audience] into three segments and, for each, write the single sentence that would make them care plus the sentence that would put them off. Same product throughout.

Run the objection map before you write a single line of copy. Most weak campaigns answer objections nobody has while ignoring the one that actually stops the sale.

Planning prompts fail when they produce a plan nobody can argue with. Each of these forces a trade-off into the open instead.

11. Campaign architecture

For a campaign with the objective [X] over [timeframe] and budget [Y], produce a table of channels with columns for role in the funnel, what it is measured on, minimum viable spend, and what we would cut first if the budget halved.

12. The brief nobody has to chase

Turn the notes below into a creative brief with exactly these sections: the one thing this must achieve, audience, the single message, mandatories, what success looks like numerically, and what we are deliberately not doing. Under 400 words.

13. Message hierarchy

We have five things we want to say. Rank them, then tell me which two make the third to fifth unnecessary. Write the campaign line for the top one.

14. Calendar from a theme

Build an eight-week content calendar around [theme]. For each week give the angle, the primary asset, two derivative assets, and the question that week answers for the reader. No filler weeks.

15. The pre-mortem

Assume this campaign failed badly. Write the post-mortem: the three most likely causes, the early warning signs each would produce, and what we should be watching in week two to catch them.

16. Repurposing map

Here is one long-form asset. Map every derivative we could produce from it across four channels, then mark which three are worth doing and which are busywork. Be honest about the busywork.

17. The competing priority

We want [objective A] and [objective B] from the same campaign. Explain where they conflict, and tell me what we would have to give up on each to get more of the other.

18. Budget reallocation

Here is last quarter's channel spend and results. Recommend a reallocation, state what you are assuming about each channel, and tell me which assumption you are least confident in.

19. Stakeholder objections

I am presenting this plan to [finance, leadership or sales]. List the six objections that specific audience will raise, ranked by likelihood, and for each note whether it is a real risk or a framing problem.

20. The minimum test

Before we commit to this campaign, design the smallest test that would tell us whether the core assumption holds. Specify what we would run, for how long, and what result would make us stop.

Expand

Always ask what you are giving up. A plan that costs nothing to choose is a plan nobody has thought about properly.

Asset production, which is where most prompt lists start and stop. These pair well with the channel-specific sets: ChatGPT prompts for SEO, social media and email marketing go deeper on each channel.

21. Landing page from a brief

Write a landing page for [offer] using this structure: headline, subhead, three benefit blocks each with a proof point, an objection-handling section, and one call to action repeated twice. Under 500 words total, and no adjectives in the headline.

22. Ad variants by angle

Write six ad variants for [product], each leading with a different angle: problem, outcome, price, speed, social proof, risk reversal. Same length, and label the angle on each so I can test them separately.

23. Email sequence skeleton

Design a five-email sequence for [scenario]. For each email give the single job it does, the subject line, the first sentence, and the one thing it must not say. Do not write the bodies.

24. Long-form outline

Act as an editor commissioning a 1,500-word piece on [topic] for [specific reader]. Give me a section outline with one line on what each section must accomplish, and flag every section that will need a source.

25. Copy without the clichés

Write product copy for [product] in under 150 words without using innovative, seamless, robust, leverage, empower or game-changing. If you reach for a synonym of any of those, be more specific instead.

26. Eight posts, eight structures

Write eight posts for [platform] about [topic] where no two use the same opening structure. Label each structure so I can see the range you covered.

27. Tone shift, claims locked

Rewrite the copy below from formal to conversational. Keep every factual claim and every number exactly as written, keep the paragraph count identical, and add no humour.

28. The thirty-second explainer

Explain [our product] to somebody who has thirty seconds and one decision to make. No history, no background, only what bears on that decision. Under 200 words.

29. Localisation check

Here is copy written for [market A]. List everything in it that would not land in [market B]: idioms, cultural references, assumptions about how people buy. Do not rewrite it yet.

30. The version legal would sign

Write a version of this copy in which every claim is one a lawyer would approve, then tell me which claims we lost and what evidence would let us keep them.

Ban the category's own vocabulary. If your copy could appear on a competitor's site with the logo swapped, the prompt did not have enough constraint in it.

Optimisation prompts should produce a hypothesis and a stopping condition, not a list of tweaks. Anything else is guessing with extra steps.

31. Hypothesis, not idea

Turn the optimisation idea below into a testable hypothesis in this form: we believe [change] will cause [effect] for [segment] because [reason], and we will know we are wrong if [metric] does not move by [amount].

32. What to test first

Here are eight things we could test on this page. Rank them by expected impact divided by effort, and tell me which three are not worth testing at all and why.

33. Where they hesitate

Review the landing page copy below and list every point at which a visitor would hesitate, in reading order. For each, say whether it is a clarity problem, a trust problem or a relevance problem.

34. Friction inventory

Walk through this signup flow as a first-time user and list every point of friction, marking which are necessary and which exist only because somebody forgot to remove them.

35. Subject lines by lever

Write ten subject lines for [email], grouped by the psychological lever each uses. Under each group, say who it works on and who it annoys.

36. The flat test

This test showed no difference. Give me four possible explanations ordered by likelihood, including the possibility that the test was badly designed, and say what I should check first.

37. Signal or noise

We ran a test for [duration] with [traffic] and saw [result]. Tell me whether this is signal or noise, what would make it conclusive, and whether stopping now is defensible.

38. The segment nobody looked at

Here is our conversion data by segment. Identify the segment with the largest gap between traffic and conversion, and give me three hypotheses about why.

39. Three ways to handle an objection

Take the top objection from our research and write three different ways of handling it in copy: address it directly, pre-empt it, or reframe it. Say which is riskiest and why.

40. Kill criteria

For the initiative described below, define the conditions under which we should stop. Be specific about metric, threshold and date, and include one condition that would be uncomfortable to act on.

Expand

Define what would make you stop before you start. It is the single discipline that separates testing from redecorating.

The other group nobody writes prompts for. These are about turning numbers into a decision, and several work far better with the data attached. Our ChatGPT prompts for data analysis go deeper on the analytical side.

41. The metric that cannot lie

For the objective [X], list the metrics people usually report and tell me which of them can move without the objective being met. Then name the one metric that cannot.

42. Anomaly check

Review the data below and identify anomalies: outliers, impossible values, suspicious duplicates, and any column whose distribution suggests a tracking problem rather than a real change. Explain your reasoning for each.

43. Attribution honesty

Here is our channel performance under last-click attribution. Explain which channels this model systematically over-credits and under-credits given a sales cycle of [length], and what we would see differently under another model.

44. The one-page report

Turn the data below into a one-page report with these sections: what happened, what caused it, what we are doing about it, and what we need a decision on. Under 400 words.

45. Explaining a bad quarter

Write the section of a report that explains underperformance, in which every claim is supported by data in the material provided and nothing is blamed on the market unless the data shows it.

46. Leading indicators

For the outcome [X], which we can only measure quarterly, identify four leading indicators we could watch weekly and say how reliably each has predicted the outcome historically.

47. The question behind the request

A stakeholder has asked for [specific report]. Tell me what decision they are probably trying to make, and what report would actually help with that instead.

48. Second-order effects

If we increase [activity] by 40 per cent, list the first-order consequences and then the second-order ones those would produce, including any that show up in other teams.

49. What the dashboard hides

Here is our reporting dashboard. List what it does not show that would change how we read it, and name the single metric you would add.

50. Forecast with its assumptions

Build a forecast for [metric] over [period], showing your reasoning as a chain of stated inputs and multipliers. Flag the input the forecast is most sensitive to, and give a range rather than a single number.

Ask what the dashboard does not show. The most expensive marketing mistakes come from metrics that were accurate and incomplete.

For this kind of work the input matters more than the model, but it is worth knowing where they differ. GPT-5.2 is quicker at generating structural options - eight angles, six variants, four ways to segment. Claude Sonnet 5 tends to hold restraint on the analytical prompts, which matters when you want an honest read on a bad quarter rather than an encouraging one.

One habit worth forming that is not in the list above: after any recommendation, ask which part of it depends on an assumption you have not confirmed. It takes one message and it will catch more bad decisions than any single prompt on this page. Chat Smith is free to try if you want to run a few of these before committing to a workflow.