Skip to content

How-to · 8 min read

Prompt WritingBriefing a machine like a pro

Diagrams
02
Tools
02
Sections
08

The short answer

Prompt writing for marketers is the skill of briefing an AI model the way you would brief a capable freelancer: stating the goal, audience, facts, constraints and format, supplying examples, and iterating on the output. Good prompts are mostly good briefs. The best teams store proven prompts as shared templates.

A prompt is a brief

Marketers already know how to write prompts. They call them briefs. The habits that make a creative brief good, a clear objective, a defined audience, one message, mandatory inclusions, also make a prompt good.

The difference is that a model has no context at all. A freelancer can ask a follow-up question or remember last month's project. A model knows nothing about your brand unless you say it, every time or through saved context. Most weak output is the result of a brief that assumed knowledge the model never had.

If a new freelancer could not do the job from your prompt, neither can the model.

The anatomy of a strong marketing prompt

Fig. 01 · Stack

Six layers of a marketing prompt

  1. Role and task

    What you want done, and from what perspective

  2. Goal

    What the output must achieve for the business

  3. Audience

    Who reads it, what they know, what they doubt

  4. Source material

    Facts, product details, approved claims to use

  5. Constraints

    Voice, length, words to avoid, legal limits

  6. Format and examples

    Output structure and samples of good work

Not every prompt needs every layer, but most weak prompts are missing the middle three.

The source material layer is the one that protects you. When you supply the facts, the model rewrites them; when you do not, it invents them. Paste the product sheet, the approved claims list or the interview notes directly into the prompt or attach them.

The examples layer is the one that improves quality fastest. Two short examples of copy you love teach tone more reliably than a paragraph of adjectives describing it.

Step by step: writing the prompt

  1. 01State the task plainly. 'Write five subject lines for a re-engagement email' beats 'help me with email'.
  2. 02Say why. 'The goal is to win back customers who have not bought in six months, without discounting.' Purpose shapes choices.
  3. 03Describe the reader. Who they are, what they already believe, what objection they hold.
  4. 04Supply the facts. Paste the offer, product details or source notes. Tell the model to use only these facts.
  5. 05Set constraints. Length, voice, banned words, regulatory limits, reading level. Link or paste your brand voice guide.
  6. 06Specify the format. Numbered list, table, word count, headings. Ask for several options, not one.
  7. 07Ask for its reasoning or doubts. 'Flag anything you were unsure of' surfaces gaps you would otherwise miss.

Worked templates for common tasks

Compare scenarios

Prompt templates you can adapt

Task: write eight headline variants for a search ad. Goal: drive qualified enquiries for [service] from [audience]. Facts to use only: [paste]. Constraints: within the platform's character limits, no superlatives we cannot prove, British spelling. Format: a table with headline, angle used and the objection it answers.

  • Ask for distinct angles, not synonyms
  • Name the objection each variant answers
  • Check character limits in current platform documentation

Iterating: the second prompt matters more than the first

Treat the first output as a draft to react to. Specific feedback produces specific improvement: 'the third option is closest, make it less formal and cut the second sentence' beats 'make it better'.

Fig. 02 · Cycle

The prompting loop

Brief

Two or three turns of specific feedback usually outperform one long, perfect prompt.

Useful feedback phrases include: 'keep the structure of option two but use the tone of option four', 'this claim is not in the source, remove it', 'shorter sentences, fewer adjectives', and 'what would a sceptical buyer object to here?'. Each names a specific change the model can act on.

If the third round is still wrong, the problem is usually missing context rather than wording. Go back and add the source material or an example instead of rephrasing the same request.

Common prompting mistakes

  • Asking for facts from memory. 'Write about the benefits of our product' invites invention. Supply the benefits.
  • Adjective-only tone guidance. 'Friendly but professional' means nothing on its own. Show examples.
  • One option. A single output anchors you. Ask for five and choose.
  • Stuffing everything into one prompt. Long multi-part tasks degrade. Break research, outline and draft into separate steps.
  • Never saying what to avoid. Banned words, claims you cannot make and topics to steer clear of are as important as instructions.

Chaining prompts for bigger jobs

Large tasks work better as a sequence of small prompts than as one enormous request. Writing a landing page, for example, splits naturally into research, message hierarchy, section drafts and an editing pass. Each step produces something a person can check before the next step builds on it.

  1. 01Research: summarise the supplied customer quotes and objections.
  2. 02Message hierarchy: propose the main promise and three supporting points, using only the research.
  3. 03Draft: write each section from the approved hierarchy.
  4. 04Edit: tighten, check voice and flag unsupported claims.

The checkpoint between steps is the point. A wrong assumption caught at the hierarchy stage costs a minute; the same mistake discovered in a finished page costs an afternoon.

Building a team prompt library

Individual prompting skill becomes a team asset only when it is written down. A shared library of tested prompts, with an example of good output for each, lets a new hire produce on-brand work in their first week.

Keep the library small and curated. Each entry needs an owner, a date and a note on when to use it. Prune it quarterly, because a tool update can change how a prompt behaves. The library also feeds your AI content workflow.

Checklist

0/9

Prompt quality checklist

Key takeaways

  1. 01A good prompt is a good brief: task, goal, audience, facts, constraints, format and examples.
  2. 02Supplying source material is what stops a model from inventing facts.
  3. 03Examples of good work teach voice faster than descriptive adjectives.
  4. 04Iterate with specific feedback, and add missing context rather than rewording the same request.
  5. 05Store tested prompts in a small, owned, regularly pruned team library.

Frequently asked

How do I write a good prompt for marketing?
State the task in one sentence, explain the business goal, describe the audience and their doubts, paste the facts the model should use, set constraints on voice and length, and specify the output format. Ask for several options and tell the model to flag anything it is unsure about. Then refine with specific feedback.
What should a marketing prompt include?
At minimum: the task, the goal, the audience, the source facts, constraints such as voice, length and banned claims, and the format you want back. Adding two short examples of work you like is the fastest way to improve tone and quality in the output.
Why does AI give generic marketing copy?
Usually because the prompt gave it nothing specific to work with. Without your facts, audience insight, point of view and voice examples, a model falls back on the most common patterns in its training data, which reads as generic. Supply specifics and the output becomes specific.
Should marketing teams share prompts?
Yes. A curated library of tested prompts, each with an owner, a date and an example of good output, turns individual skill into a team capability. Keep it small and prune it regularly, because tool updates can change how prompts behave.
Is prompt engineering still a useful skill?
The tricks change as tools evolve, but the core skill of explaining a task clearly with the right context remains valuable. Marketers who can brief well, judge output and give precise feedback will keep getting better results regardless of which interface they are using.

Published by Fabulous.Media, a network of specialist marketing agencies. Updated 9 October 2026. Platform features change often; check current official documentation before acting on platform-specific detail.

Read next

Prefer a specialist to do this with you? The network has a house for every discipline in this library.

Request an Introduction