What AI is actually being used for in advertising
Written from the inside of live paid accounts rather than from a trend deck. Where AI has genuinely changed the work, where it has not, and what that means if you are the one answering for performance.
The four places it has actually landed
1. Creative production
The clearest win. Producing a video ad used to mean a brief, a shoot and an edit. Now the bottleneck is deciding what to make, not making it. This is where nearly all of the real change has happened.
2. Variation and iteration
The second win, and the underrated one. When a variation costs a render, you can afford to change one variable at a time and actually learn something. Most accounts still test three assets and call it a test.
3. Reporting and diagnosis
Useful, and further along than most teams realise: reading an account and saying which creative is carrying it and why is work a model does well. Deciding what to do about it is still judgement.
4. Media buying itself
Least changed, despite the marketing. The platforms' own automation absorbed most of this years ago. If someone is selling you AI media buying, ask what decision it makes that Advantage+ does not.
Where it still fails
- Anything that depends on a viewer believing a specific real person used the product.
- Products whose texture, fit or scale is the whole sell.
- Claims. A model will write a claim you are not allowed to make, confidently.
- Brand voice over a long run — it drifts, and nobody notices until it has.
What this changes about how you work
When creative is no longer scarce, the scarce thing becomes knowing what to test and being able to read the result. That is a media buying skill, not a production skill, and it is why the teams getting the most out of this are not the ones with the best video tools.
Our own creative format benchmark shows why. Across twelve live accounts, offer-led ads indexed 155 against their account average, while styled lifestyle photography, where most production budgets go, indexed 66. And in four accounts running both, unrehearsed real footage beat paid creators reading a brief in all four. The full method and numbers are on the creative format benchmark page.
Questions, answered.
What is AI advertising?
In practice it means three things: generating ad creative, generating variations of creative that works, and using models to read account performance. The buying itself is largely automated by the ad platforms already.
Is AI-generated advertising allowed?
Yes. Both Meta and TikTok allow it and both maintain rules on labelling synthetic media, which have changed more than once — check the current policy for the placement you are buying.
Will AI replace media buyers?
It has replaced a lot of production work and very little decision work. When creative becomes cheap, knowing what to make and how to read the result becomes the job.
What is the best AI tool for advertising?
Depends on what is capping your account. If it is creative supply, a video ad generator. If it is knowing what to test, that is not a tool problem yet.