“Make it clearer” sounds like feedback, but it leaves the designer to discover both the problem and the intended result. Useful feedback on ChatGPT ad concepts identifies something in the material, explains the possible misunderstanding and gives the next version a job to do. That makes the review actionable without turning every reviewer into an art director.
Start by agreeing what is being reviewed. An image in a presentation, an old export in a message and the latest ad draft are different objects. Even excellent comments can produce the wrong revision when they refer to different versions.
Put one task beside the concept
Write a short statement of the customer decision that the ad supports. Include the image, proposed title, body and destination in the review material. Add a version label that will remain visible when someone forwards the document.
Consider a fictional booking software company advertising to workshop operators. The concept should help a receptionist recognise a way to find available appointments. The image shows a calendar. The task is not to explain every feature in the software or demonstrate that the brand is technologically advanced.
A reviewer can now ask whether the calendar makes availability understandable. That is a narrower and more useful question than whether the image feels exciting. It also gives the product owner and designer a shared basis for discussing changes.
Context hints add context; they are not exact keywords or placement guarantees. OpenAI: Targeting Keep the imagined customer situation in the brief, but do not review the concept as though a particular conversation is certain to display it.
Build a comment from three parts
Use an observation, a consequence and a requested decision. This is an editorial review method, not a requirement imposed by the advertising platform.
An observation points to material that another person can inspect: “Available and booked appointments have similar visual weight.” The consequence explains the concern: “It is difficult to see what the receptionist can find using this screen.” The request defines the next job: “Show an option where availability is understandable without reading a separate legend.”
That is more useful than “Make the boxes green.” Green may already mean a confirmed appointment in the real product. Describing the communication problem allows the designer to investigate a solution that respects the product.
Be equally careful with claims about readers. “Customers will be confused” sounds like an observed outcome when it may be one reviewer’s prediction. Write “I interpreted this as a confirmed booking; please check whether the intended meaning is available time.” The uncertainty is now visible and can be resolved.
Give comments different levels of urgency
Use three categories: must change, needs an answer and optional improvement. Each item also needs a person who can resolve it. Without an owner, a comment can circulate through several revisions without becoming a decision.
| Fictional review comment | Category | Responsible next step |
|---|---|---|
| The screen shows a feature the product does not offer | Must change | Product owner identifies an accurate screen |
| The colour could indicate either availability or confirmation | Needs an answer | Designer checks meaning with the product owner |
| A different background feels more contemporary | Optional improvement | Review whether the current background obstructs the task |
Start the meeting with potentially misleading information. Continue with unresolved meaning, then consider preferences. A decorative suggestion should not silently acquire the same priority as an inaccurate product claim.
If a comment concerns a promised benefit, request evidence through the claim evidence matrix. Designers can make a statement prominent, but prominence does not establish that the statement is true. Resolve that question before asking for a more persuasive treatment.
Reconcile the feedback before assigning revisions
One person should assemble the review into a coherent change list. Several comments may describe the same problem. “Too busy”, “I cannot find the openings” and “Too much calendar” may all refer to the same missing visual priority.
Combine those into one problem with one owner. Keep separate issues separate when they require different decisions. Product accuracy and small text should not disappear inside an instruction called “Improve the graphic”.
Conflicting requests also need a decision before production resumes. If one stakeholder requests more features while another requests a simpler image, return to the stated task. Which extra feature helps the receptionist recognise available appointments? If none does, adding the feature belongs to a different concept.
Record the reason for rejecting a suggestion without judging the person who offered it. “Outside this concept’s availability task” is more useful than “Does not fit the brand”. The former explains a boundary that another reviewer can apply consistently.
Decide what closes the comment
A review item should have an acceptance condition. For the calendar, it might be that a colleague unfamiliar with the project can point to an available appointment and describe what the image represents. This is a practical comprehension check, not evidence of future campaign performance.
Check the revised image in the complete ad. Enlarging the calendar might solve one problem while pushing the product name into an unreadable corner. Use the small-size review when the issue depends on detail, rather than closing it after inspecting a zoomed design file.
A preview does not establish serving eligibility, as OpenAI’s campaign documentation explains. Give internal approval a precise label, such as “creative comprehension reviewed”, so that it cannot be mistaken for a platform status.
From reaction to resolved comment
- Observe
Available and booked appointments look alike.
Reconsider - Explain
The receptionist cannot tell which appointment can be chosen.
- Decide
Make availability understandable without a separate legend.
- Verify
A fresh reader identifies an available time in the revised ad.
Checkpoint
Preserve the decision, not an endless discussion
The final record can be brief: version, open questions, decision and owner. Keep the original comment beside the resolution. This lets a future reviewer understand why a detail changed without reconstructing a long conversation.
Once the blocking issues are closed, use that record in the final concept selection. It reveals which limitations have been resolved and which remain relevant to the choice. The value of the review is a clearer concept and a smaller set of unresolved decisions, not a larger collection of annotations.
Sources and scope
Turn creative review comments into prioritised, verifiable changes to a ChatGPT ad concept.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.