We ran an AI content check on four ads. One of ours was flagged.

Last week we wrote an “appropriate-content” checklist, loaded it into ReviewAI, and pointed it at four pieces of creative. We wanted to see if plain-English checks catch the stuff that gets past a gauntlet of reviewers.

Apparently they can (fantastic news). But one of the things they caught was in our own ad (less fantastic).

Appropriate-content checklist TL;DR

The checklist we put together was intended to flag harm, dignity, consent, representation, sexual content, safety, distressing imagery, and vulnerable subjects. Plus we made a catch-all for anything those might not have covered. Every one of these is something that should be checked before you sign off on creative.

What we'll cover

What it found in our ad

In our product demos we use a fictional headphone brand called groove4U, and the original print ad shows a person in workout gear pulling a training rope. That ad passed all nine checks without any issues.

But then we ran the revised version, where the rope shot is replaced by a woman running outdoors in headphones.

That one failed “Safety and imitability.” The checklist flagged that running outdoors in headphones is copyable behavior that reduces awareness of traffic and surroundings. The analysis even cited our own footer line (always be aware of your surroundings when wearing audio accessories) as evidence the risk was real, then flagged it anyway rather than treating the disclaimer as a fix.

Our creative team is absolutely amazing, but nobody thought to bring that up. The revision also resolved an open question about crop and pose in the original (does it cross from fitness depiction into sexual presentation?). The edit fixed one thing and introduced another, and without checking both versions, that trade would have been invisible.

What it found in the Callaway ad

You’ve probably seen the Good Good Golf × Callaway driver ad that made the rounds last week. Out of curiosity, and a desire to constantly improve our platform, we ran the same content checks on it.

Source: Wall Street Journal - Good Good Golf and Callaway Apologize for ‘Scary’ Ad

There were nine checks, immediately followed by nine flags, all pointing at the same two-and-a-half-second window about seven seconds in — the point where a man shoves a woman to the ground to keep her away from a golf bag. ReviewAI flagged harm, dignity, consent, safety, distressing imagery, and vulnerable subjects. Each analysis noted the comedic framing, then cited the instruction that matters most in this checklist, judge the asset as a first-time viewer with no knowledge of the brand or intended tone. Comedy might get recorded as context, but it doesn’t provide an override for a flag.

We noted that everything before and after those seconds came back clean.

Where it was off the mark

To be fully transparent, not every flag was a real problem. Our track4U fitness tracker demo failed “Distressing imagery” because of on-screen heart rate and sleep data. For a fitness brand, that’s not really an issue. But, issue or not, it’s honest. The check treats health content as flag-worthy by default and hands the call to a person instead of deciding quietly in its machine brain. One of the nine Callaway flags (sexual content) was a bit of a stretch, with reasoning visible enough to dismiss it quickly.

The point of all this

Across four assets we gave ReviewAI, thirty-six checks were answered. We saw eleven fail, twenty-five pass, and one asset come back clean and clear. Of the eleven fails, nine were Callaway (one of those was a stretch), one was ours (real, and we missed it), and one was a track4U false positive. Every result from ReviewAI came with a written analysis and a comment pinned to the exact frame or page.

ReviewAI didn’t block anything or slow it down. To us, it’s a reviewer in the approval chain who hasn’t been told what the joke is supposed to be. The humans still decide, of course, but they decide with the question in front of them instead of finding it in the comments after launch.