ChatGPT guide
Give ChatGPT a brief worth answering.
Learn how to prompt ChatGPT for useful writing, research and finished work, with copyable examples and checks for better results.
Updated
Identify the job behind the request
‘Help with marketing’ hides several possible jobs: choosing a message, writing copy, building a poster or measuring a campaign. Pick the deliverable first. If you need a decision, state the alternatives. If you need an asset, specify the channel, audience and next action.
A useful brief also says what you already know. A restaurant with a full dining room and quiet takeaway business needs a different campaign from a new restaurant seeking its first customers. That context changes the answer more than a request to ‘be an expert’.
Help me choose the message for [campaign]. We sell [offer] to [audience]. The current obstacle is [obstacle], supported by [evidence]. Give three message directions, explain the assumption behind each and recommend the first one to test. Ask about missing information that could change the recommendation.
Give references a job
Do not attach a pile of files and hope their meaning is obvious. Label which file supplies facts, which demonstrates tone and which contains the constraints. If your website is a tone reference rather than a factual source, say so. A previous campaign may be an example to avoid.
For long tasks, define the questions the sources should answer. Ask for references to the relevant file sections so you can inspect the basis of the result.
Use the product sheet for facts, the attached email for our voice and the brand guide for visual rules. If they conflict, identify the conflict before drafting. Do not use the tone reference as evidence for product claims.
Improve one dimension at a time
When an answer misses the mark, identify the dimension: too broad, too formal, wrong audience, unsupported or difficult to use. ‘Make it better’ leaves the model guessing. Give a concrete change and preserve the part that worked.
Keep the argument and facts. Replace the broad opening with the specific customer problem, cut the draft by a third and finish with one clear next step. Highlight any factual claim you cannot trace to my sources.
Judge the output against the intended use
Read the result as the intended recipient. Can a colleague act on it? Could a customer misunderstand the offer? Does a recommendation still make sense under the most plausible alternative assumption? A fluent answer can fail those checks.
- For writing: check evidence, voice and the reader's next action.
- For analysis: check definitions, calculations and source coverage.
- For creative work: check distinctiveness, brand fit and usability in the real format.
Common questions
Do I need a long ChatGPT prompt?
Only when the task needs the detail. A narrow rewrite may need two sentences; a campaign brief needs audience, offer, evidence, assets and constraints.
Should I ask ChatGPT to act as an expert?
A role can focus a perspective, but it does not supply evidence or qualifications. Explain the actual decision and quality criteria as well.
Keep exploring
Sources and editorial note
Guidance checked 9 October 2026. These AI-assisted guides and original templates are reviewed against the linked sources. Results depend on your inputs and available tools. Model recommendations are editorial starting points, not comparative test results.