The short version
- Review for "can we ship this" before you review for "does this look good."
- Wrong product facts, wrong prices, mangled logos, unreadable text, and unsupported claims are stop-ship items, not fix-later items.
- A score can't replace the defect list. Reviewers need to know what's wrong, who signs off on it, and what to recheck after the fix.
Review AI creative for whether it can ship before you review whether it looks good. Wrong product facts, wrong prices, mangled logos, unreadable text, and unsupported claims are stop-ship items, not scores to weigh against composition. And a total score can't replace the defect list: name what's wrong, who signs off on it, and what to recheck after the fix.
A batch finishes generating and everyone goes straight to the art direction. Is the color right, is the composition interesting, does it stop the scroll.
Fair questions. They just come later.
If the product UI is drawn wrong, the price is off, the logo is distorted, or the headline promises something you can't back up, none of the visual craft matters. That creative can't run.
Three verdicts, not one score
Keep the outcome simple:
- Ship it. Nothing blocking turned up. The owner still gives the final nod.
- Fix and recheck. The direction works, but there are specific defects to correct.
- Stop. Something is wrong with the facts, the brand, the text, or the risk profile. Nothing ships with the defect still attached.
Resist the urge to reduce this to a number. "82" could mean the headline runs a little small, or it could mean the price is wrong, and those are two very different fixes.
The seven checks
Product truth
- Does the product's shape, UI, and structure match the real thing?
- Do the features shown actually exist?
- Are prices, plan names, dates, and availability accurate?
- Is anything on the roadmap being shown as already shipped?
This is the first gate. When something fails here, stop. Don't start adjusting color on a creative that is describing a product you don't have.
Brand consistency
- Is the logo complete, sharp, and correctly proportioned?
- Are the brand and product names spelled right?
- Have the brand colors, typefaces, or overall tone drifted?
- Has anything from another brand's identity crept into the frame?
Generated logos are the repeat offender here: a wordmark with a duplicated letter, an extra stroke on the mark, a brand color one step off the real value. All of it survives a casual glance at thumbnail size.
Copy and text rendering
- Can every word in the image be read cleanly?
- Any garbled characters, duplicated words, or awkward line breaks?
- Does the headline carry exactly one idea?
- Does the CTA match what the landing page actually offers?
- Has localized copy been confirmed by someone who speaks the language?
Visual quality
- Is the subject unmistakable?
- Can someone follow the hierarchy in a couple of seconds?
- Do the lighting, scale, and edges between product and background hold up?
- Is decoration competing with the product or the message?
- Does it still hold up scaled down to phone size?
Placement fit
- Do the ratio and pixel dimensions match the target placement?
- Are the headline, logo, and product clear of the platform's own UI overlays?
- Have the square, vertical, and horizontal versions been re-composed rather than just cropped?
- Do file size and format meet the spec?
Specs change. Treat the platform's own ad preview and current documentation as the source of truth, not a spec sheet someone saved last quarter.
Risk and compliance
- Are performance, comparison, and outcome claims supported by something you could produce on request?
- Any absolute language or promises you can't substantiate? (The FTC's advertising FAQ for small business is the short version: ads have to be truthful and claims need substantiation before they run.)
- Do you have rights for the people, trademarks, stock images, and fonts in the frame?
- Are offer conditions and limitations legible and complete?
- Is this a regulated category with extra requirements, like health or finance?
A quality review is not legal advice. Anything high-risk goes to whoever owns compliance or legal.
Novelty and repetition
- Is this version different from what you already have in any way other than color?
- Does it bring a new audience, situation, or angle of persuasion?
- Is the visual sitting too close to its reference material?
- Across the set, can you articulate what each version is testing?
Novelty comes last on purpose. An accurate but ordinary creative can be improved on the next pass. An interesting but wrong one can't run at all.
Hard stops
Any one of these means stop. Not "flag it and fix it after launch":
- Wrong product, price, feature, or availability.
- Damaged logo, misspelled brand name, or altered packaging.
- Text in the image you can't fully verify.
- A core claim with nothing behind it.
- Unclear provenance or usage rights on a source asset.
- Serious occlusion or cropping in the real placement preview.
- A landing page that doesn't deliver what the creative promised — Google Ads' destination requirements make this a policy issue, not just a conversion issue.
Who checks what
Route each area to the person who would notice the problem first:
- Product owner confirms features, UI, and limitations.
- Brand or design owner confirms the visual system.
- Marketing owner confirms audience, message, and action.
- Local team confirms language and cultural read.
- Compliance owner confirms high-risk claims.
Having everyone review everything feels safer, and isn't. It creates the gap where four people each assumed one of the others was checking the price.
Creative Factory surfaces quality problems by dimension, so the team can locate the defect first and then decide whether to fix, recheck, or drop it. What the system returns is a risk signal, not an approval.
FAQ
Does every creative need all seven checks?
Anything going out publicly, yes. Early direction sketches can get the short version: product facts and obvious risk. But nothing skips the full pass on its way to delivery.
Our automated score is high. Can we ship?
No. An automated review can't own product facts, brand permissions, or legal exposure. A high score means the system found nothing obviously broken. That is a long way from approved.
Doesn't all this slow us down?
Unordered revision cycles are what actually slow you down. Blockers first, then copy and visuals, then novelty. That sequence usually beats five people dropping scattered notes into a thread at the same time.