Ad Creative Feedback Examples: Turn Vague Notes into Useful Edits

Useful ad-creative feedback identifies the visible problem, explains what should become clearer and names the details that should stay. You can say this in ordinary language. “The bottle disappears into the background; give it more separation while keeping the amber glass and label” is enough to start a focused revision. It does not require a technical design specification.

When you are editing a complete AI-generated ad, also refer to the exact version. A clear note attached to the wrong image creates another round of confusion before any creative work begins.

Turn a reaction into an observable change

Asana's guidance on constructive criticism emphasizes feedback that is specific and actionable. For an advertisement, connect that principle to what the viewer should notice or understand.

Here are six original examples for product creatives:

The final column is especially helpful for AI revisions. It tells the system which parts of the current direction you value. It does not guarantee those pixels remain identical, so compare the new version afterward.

Explain the problem before prescribing every detail

You can give an exact instruction when the correction is factual: the name is misspelled, the cap is the wrong shape, or the set includes one cup rather than two. For a design problem, explaining the outcome leaves room for a better solution.

“The product name is hard to read on the phone preview” is more informative than “increase the type by 12 percent.” The designer might improve contrast, shorten the line or simplify the background instead of only enlarging the text.

Likewise, “the first frame does not show why someone would need this product” points to a creative issue. Moving a decorative element will not solve it. Separate that concept problem from smaller changes to spacing or color.

Keep one revision round focused

Suppose a fictional desk-lamp ad has three issues: the lamp has the wrong switch, the first frame is visually crowded and the final caption is too small. Start with the incorrect product feature and the main communication problem. A polished incorrect lamp is still the wrong advertisement.

Use a short priority order in your message:

1. Match the lamp's actual switch and base from the reference.
2. Simplify the first frame so the working desk is immediately understandable.
3. Make the final product name readable at the intended display size.

You do not need to hold back every smaller preference. But distinguish a required factual correction from an optional art-direction experiment. Otherwise, a minor color preference can obscure the edit that determines whether the creative represents the product correctly.

Send a natural-language revision that stands on its own

Here is a complete original message you could use in Panelly Studio:

Use the latest four-panel desk-lamp version as the starting point.
The mood is right, especially the warm pool of light on the notebook.
The lamp itself needs to match @image1: a flat circular base and a small switch on the stem, not a touch button on the base.
The first frame also feels crowded. Remove the extra plant and the loose cables so the desk and lamp become the focus.
Keep the cream finish, the working-at-home scene and the final line “Light for the work ahead.”
Generate the revised complete ad without adding new features or changing the offer.

This message combines a factual fix, a layout change and a preservation note. It keeps the conversation flexible while giving the next version a clear purpose. The image reference supplies visual evidence; the words explain which detail matters.

Integrated headline and CTA text can remain part of the generated image. You do not have to create a separate text-layer workflow to ask for precise wording. Verify the actual result instead of assuming that a successful generation followed every instruction.

Compare versions against the request

Put the accepted starting version beside the revision. First check the requested change: did the base and switch become correct? Then check the preserved details: is the cream finish still right, is the useful pool of light still visible, and does the final line remain accurate?

A simple comparison note can read: “Product shape corrected; first scene clearer; new version introduced a second lamp in frame three.” That separates progress from a new issue. Your next message can address the new issue without restarting the whole brief.

Avoid deciding only that the second image feels “better.” Save why it is better. That record makes later creative choices easier and helps a collaborator understand which parts of the direction are intentional.

Know when the request is actually a new concept

Changing the product audience from office workers to outdoor campers is not just a lighting adjustment. Replacing a quiet product demonstration with a humorous character story is a new concept. Treat it as another creative direction so the previous version remains available.

A useful sentence is: “Keep this version, and explore an alternative for…” You can then compare concepts separately from corrections. For planning that comparison, use the small-batch ad testing guide rather than assuming every new image belongs in one undifferentiated pile of variants.

FAQ

Do I need design vocabulary to give good feedback?

No. Identify what you see, what is not working for the viewer and what should remain. Concrete references are usually more useful than specialized terminology.

Should I ask for one change at a time?

Group changes that solve the same problem, and prioritize them. Several related edits can fit one message. A long list of unrelated preferences is harder to evaluate afterward.

Will a precise AI edit keep everything else identical?

Not necessarily. A complete new generation can change unrequested details. State what should remain and compare the final version with the reference before choosing it.

Recorded example: removing clutter also changed the setting

This is a controlled editorial exercise made on September 17, 2026 with the GPT Image 2.5 image models used by Panelly. We generated a fictional ATELIER perfume scene, then sent that image as the reference for one revision directly through the image provider. It is not a customer campaign or a recorded Studio session. The clutter in the starting image was deliberately requested; it is not evidence that the model normally produces this problem.

The revision request is shown below (translated in non-English editions; the run notes include the original English text):

Revise the supplied image as a focused editorial demonstration. Keep the single amber glass perfume bottle, short black cylindrical cap, rectangular ivory ATELIER label, original framing, beige stone plinth and warm sunlight from the left. Remove all foreground flowers, the small mirror and the folded fabric. The bottle is the only product and should be the clear focal point. Keep a visible, physically plausible contact shadow where the bottle touches the stone, with the cast shadow extending right. Keep the label unobstructed. Add no props or text. Do not redesign the bottle.

One text-to-image request and one image-to-image request completed. The requested setting was 1K and 3:4. The returned originals measured 768 × 1024 and 1086 × 1448 pixels respectively; requested resolution labels are not a promise of identical pixel dimensions. The images on this page are WebP copies for reading.

The first call took about 49 seconds and the reference revision about 79 seconds in this run. These are individual elapsed times, not typical latency or a service guarantee. We did not measure sales, conversion, or an acceptance rate. Download the exact prompts and run notes.

To try the same review process in Studio, type @, select the starting image on your canvas, and describe a focused correction. The next request here would explicitly restore the warm wall and original platform while keeping the cleared foreground. That is a proposed next step; we have not shown a third generated result.

Sources and further reading

• Asana constructive feedback guidance