ChatGPT can help with audience questions, messaging alternatives, first drafts, and analysis of supplied results. Choose the job you need first. Copy generation needs confirmed offer facts; campaign analysis needs actual data and metric definitions. A confident draft supplies neither.
Choose an input that fits the marketing job
| Job | Bring to the conversation | Ask for |
|---|---|---|
| Explore a message | Approved offer and permitted research notes | Alternative angles, each tied to supplied evidence |
| Draft an asset | Audience, channel, offer, voice examples and constraints | Copy plus a list of claims that need confirmation |
| Revise existing copy | Current text and specific reviewer feedback | A revision showing which request each change addresses |
| Review performance | Actual export, reporting window and metric definitions | Recomputed comparisons and limits on the conclusion |
OpenAI's prompting documentation recommends stating the goal, relevant context, and desired result. For a campaign, the important context includes the exact promise a customer will read. Paste a short approved excerpt if the whole brief is unnecessary. If you use files, check what was uploaded before relying on them.
Build the claim list before the headline list
Consider this original fictional offer:
- A 60-minute online workshop on organizing a personal photo collection.
- Participants receive a worksheet and a recording available for 14 days.
- The price and registration deadline have not been approved.
- There is no supplied certificate, testimonial, or measured result.
A usable first pass identifies the workshop length, topic, worksheet, and 14-day recording window as supplied facts. It leaves price and deadline open. “Certified,” “lifetime access,” and “organize every photo in an hour” have no basis in this brief.
Ask for different messages about the same offer rather than three synonyms for “easy.” One angle could emphasize choosing a folder structure; another could emphasize deciding what to keep. Those are proposed messages to evaluate against the actual workshop content, not evidence that an audience prefers them.
Draft with the missing facts visible
Help me prepare marketing copy for this offer.
Approved offer facts: [PASTE]
Intended audience and what we actually know about them: [PASTE]
Channel and length limit: [ENTER]
Approved voice examples: [PASTE OR NONE]
Required terms or prohibited claims: [PASTE]
First list the usable claims and the missing facts. Keep supplied
customer evidence separate from proposed messaging hypotheses.
Then draft two substantially different message angles, explain the
reader question each addresses, and write one asset for each angle.
Use only supplied product facts. Do not invent research, testimonials,
scarcity, prices, guarantees, credentials, deadlines, or results.
If a missing fact is essential, show [CONFIRM: fact] in the draft.
Finish with the exact questions I should send to the offer owner.
Do not publish or send anything.
For the fictional brief, a bounded draft line is: “Work through a folder plan for your photo collection in a 60-minute online workshop. Includes a worksheet and 14-day access to the recording.” This is an authored illustration of acceptable scope, not a recorded ChatGPT response or a tested ad.
The line still needs the real registration destination and the confirmed commercial terms before it can become an asset people can act on. A bracketed price is useful during review; it is a defect in released copy.
Review the promise across the whole asset
Compare the headline, body, button, landing-page destination, and any image text. A careful paragraph does not fix an unsupported promise in the headline. Read each number and condition against the approved source, including how long access lasts and what the purchase includes.
If a reviewer changes the recording window, update every asset that states it. OpenAI's creative-request workflow treats briefs, drafts, reviewers, and final assets as connected records. You can apply that discipline in the tools your team already uses; a generated draft does not require building a new request app.
When results arrive, use a consistent marketing denominator before comparing campaigns. Draft variations are candidates to test, not proof of higher conversion.
Finish by giving the reviewer the two message angles, their factual basis, and the unresolved offer questions. After those questions are answered, put the approved terms into the actual asset and check its destination before release.