How to QA AI-generated ad creative before launch
Ad ops teams now generate campaign creative in batches — hero images, lifestyle shots, product visuals, every placement size, all from a prompt. The generation pipeline got fast; the review step mostly didn’t change at all. And unlike a bad organic post, a bad ad creative ships with a media budget behind it: a warped logo doesn’t just exist, it gets a million paid impressions.
The defects that survive an automated pass
Automated checks are genuinely good at what machines get technically wrong: resolution, file specs, obvious NSFW, gross artifacts. Run them first, on everything. What they miss is judgment — and judgment failures are exactly the ones audiences screenshot:
- Logo and brand-mark damage. Generators approximate marks instead of reproducing them. Blur, wrong proportions, invented details — often invisible at full size and obvious at 300×250.
- Garbled in-image text. Models draw what text looks like, not what it says. Price callouts, packaging, CTAs inside the image are the highest-risk pixels in the frame.
- Warping and artifacts under resizing. A creative that passes at 1080×1080 can shear at 970×250. Every placement size is its own review, not a crop of an approved one.
- Anatomy in lifestyle shots. Hands holding the product, faces in the background. One six-fingered hand turns a paid impression into organic mockery.
- Style and brand drift across the set. Item 40 of 300 quietly stops matching the palette and grade of item 1. No single-image check catches a set-level defect.
- Claims and compliance. Generated visuals can imply claims legal never approved — a “result” that overpromises, a lookalike of someone else’s trade dress.
None of this is hypothetical. Coca-Cola shipped an AI holiday ad where the truck wheels changed shape between frames, and the coverage wrote itself. The defect wasn’t subtle — nobody with authority looked frame by frame before it ran.
The pre-flight process
- Automated technical pass on 100%. Specs, resolution, safety, duplicate detection. Fractions of a cent per image; this is the floor, not the review.
- Human per-item review on everything that survives — in every size. Several independent reviewers score each creative against a fixed rubric: brand marks intact, text says what it should, anatomy clean, artifacts, on-brand look. Every failure gets a fix-note, so the retouch queue is a list, not a mystery.
- Senior eyes on the flagged slice. Brand or creative lead reviews only what the verdicts flagged, plus hero placements. Senior time goes where the disagreement is, instead of scrolling three hundred thumbnails at 11 pm before the flight date.
The math next to the media spend
A 300-creative campaign reviewed per-item by humans costs about $57 at Meldar’s on-demand rate of $19 per 100 assets. That is a rounding error against even a small flight’s media budget — and one caught logo defect pays for years of it. The asymmetry is the whole argument: generation made creative volume cheap, which made the cost of shipping one bad item relatively enormous, because it now ships with paid distribution and a screenshot-ready audience.
The pre-flight question worth institutionalizing: “who looked at every size of every creative in this flight?” If the answer is a pipeline, not a person, the campaign is one warped wheel away from being the case study.
Related reading: How to review AI-generated images before you ship them · What is AI slop? And how studios catch it before customers do