25 prompts for marketing tasks that actually work
Prompt collections mostly fail because they hand over finished text instead of a construction principle. Here each one states why it looks the way it does — so you can phrase the twenty-sixth task yourself.
In short
- Every usable prompt has four building blocks: starting point, task, rules, output format.
- The single most effective addition is a ban on invention — the instruction to mark unknowns as unknown rather than filling them in plausibly.
- Prompts meant to criticise need an explicit instruction not to praise. Otherwise you get agreement instead of review.
- Long prompts are not a problem; badly ordered ones are. The most important part belongs at the beginning and at the end.
The construction principle
Before the collection, the pattern behind it. Four building blocks, always in this order:
1. Starting point
Who you are, who you work for, which figures and circumstances apply. Without this part, every model produces the average of all companies — and that is useful to nobody.
2. Task
Numbered, one action per point. Five numbered tasks get worked through more reliably than the same content as flowing prose.
3. Rules
What should not happen. This is where the ban on invention goes, where the style frame goes, where what to leave out goes.
4. Output format
Table, list, prose, length. Without a specification you get prose of medium length — rarely what was needed.
Worth knowing
Details in the middle of a long prompt are taken into account less reliably than those at the beginning or the end. The effect is well documented and affects all common models.
Practical consequence: the most important instruction belongs at the close, not in the middle of a list. And when a prompt gets very long, it pays to repeat the central rule at the end — that works better than any rephrasing.
Research and planning
1. Cluster search terms
Paste the raw list, have it sorted by intent, group variants of the same question into clusters. Why this way: sorting is the strength; inventing search volumes is the weakness — hence the explicit ban.
2. Sharpen the audience
Paste the description, have ten follow-up questions raised that narrow it. Why this way: questions deliver more than a rewritten version — you keep the decision.
3. Collect objections
Describe the offering, have the twelve most common objections phrased in the audience's language. Why this way: that list is the outline for the landing page and the questions section.
4. Find competitor gaps
Paste three competitors' tables of contents, have the questions named that none of them answers. Why this way: comparing is a strength; judging would be guessing.
5. Topics from customer questions
Paste twenty real customer questions, have them grouped into topics, one heading per group naming the result. Why this way: you supply the substance, the model supplies the structure.
6. Draw up the quarterly plan
State the topic list and your capacity, have it sorted by proximity to a buying decision. Why this way: sorting against a stated criterion is reliable; prioritising "by feel" is not.
Writing and revising
Prompts 7 to 14 all follow the same principle: the model receives your substance and a tight frame, not an open assignment.
| No. | Task | The decisive addition |
|---|---|---|
| 7 | Draft from bullet points | "use no figure that is not listed above" |
| 8 | Cut the text by 30 per cent | "remove no statement, only words" |
| 9 | Technical text for lay readers | "without simplifying – where it is complex, say why" |
| 10 | Five title variants | "name the result, not the topic" |
| 11 | Generate a questions section | "every answer understandable without the rest" |
| 12 | Summary at the top | "four points, each a complete statement" |
| 13 | Translation with adaptation | "adapt amounts, legal position and examples to the country" |
| 14 | Put text into the brand voice | supply the style document, do not describe it |
Checking and criticising
The most useful group — and the one where the phrasing matters most.
Review the following draft. Be strict; praise nothing. Draft: [paste text] Tasks: 1. Mark every sentence that could equally appear on any competitor's website. State the percentage. 2. List every verifiable claim and mark whether I have to check it myself. Do not invent sources. 3. Name the places where personal experience or original figures are missing – and for each, ask me one concrete question whose answer would fill the gap. Do not rewrite the text. I want to see the gaps.
Prompts 16 to 20 follow the same pattern for other checks: quotability for answer engines (16), comprehensibility for lay readers (17), objection coverage on a landing page (18), agreement between structured data and visible content (19), completeness of a privacy notice against the list of services in use (20).
The line "be strict; praise nothing" is not a stylistic flourish but functionally the most important sentence in any review prompt. Without it, nearly every answer opens with an appreciation of what works — and that dilutes the criticism, because it creates the impression that things are broadly fine.
Second most important line: "do not rewrite the text." Receive a revised version and you can no longer see what was wrong with it — and learn nothing for the next piece.
Sales and following up
21. Meeting preparation
Paste the contact history, have three open points and two likely objections named. Why this way: summarising from supplied material is reliable — assessments without material are not.
22. Draft a follow-up email
Paste the meeting note, have a draft produced under the rule: no pressure, one concrete next step, at most 120 words.
23. Make a quote understandable
Paste the quote text, have the three places named where a customer would stumble. Why this way: do not have it rewritten — have the places pointed out.
24. Analyse rejection reasons
Paste the list of rejections with reasons, have patterns named and, per pattern, a possible cause in marketing.
25. A reference from notes
Paste the project notes, have a draft customer reference produced — under the rule to use no figure not in the notes, and to mark the text as a draft for the customer's approval.
In closing
A prompt collection goes out of date; a construction principle does not. Starting point, numbered tasks, rules, output format — plus a ban on invention, and for review tasks the instruction not to praise and not to rewrite.
Know those six elements and you no longer need a collection. The 25 above are examples of what it looks like assembled.
Common questions
Which prompts help in day-to-day marketing?
The review prompts most of all: substance check on a draft, quotability for answer engines, objection coverage on a landing page. After those come research and sorting — clustering search terms, collecting objections, deriving topics from real customer questions. Least useful are open writing assignments with no substance of your own.
How do you build a good prompt?
From four building blocks in this order: starting point (who you are, which figures apply), numbered tasks (one action per point), rules (what should not happen, including a ban on invention) and output format (table, list, length). The most important instruction belongs at the close, not in the middle.
Why does the model only agree instead of criticising?
Because the instruction not to praise is missing. Without the addition "be strict; praise nothing", nearly every answer opens with an appreciation of what works, and that dilutes the criticism that follows. Equally important is "do not rewrite the text" — a finished revision conceals what was wrong.
How do you prevent invented figures?
With an explicit rule in the prompt: use no figure that is not in the supplied starting point, and mark unknowns as "not measured" or "[detail missing]". Without that rule, every model fills in plausible-looking values — particularly for search volumes, market figures and studies.
May customer data go into a prompt?
Only where the processing arrangement is settled: a contract with the provider, naming them in the privacy notice, a basis for any transfer abroad, and a check on whether inputs are used for training. If that is not settled, only anonymised details belong in the prompt.
Marketing that sets itself up
The Studio Engine beta is live. Claim your spot and help shape it from the start.
Join the beta →