Glossary

What is artificial intelligence in advertising?

Q: Is machine learning the same as AI?

[Machine learning](/resources/glossary/machine-learning) is the subset of AI that learns patterns from data rather than following written rules. In advertising the two terms are used interchangeably, and almost always mean machine learning.

In advertising, artificial intelligence means using machine learning models to automate decisions that were previously made by rules or by people — bidding, targeting, creative production, fraud detection — and, more recently, to generate the surface the advertising appears in.

AI and conversational advertising

All 220 terms

Where AI actually operates in the ad stack

The term covers so much that it has stopped distinguishing anything, so it is worth being specific about where models genuinely do the work.

Bidding. Predicting the value of an impression and setting a price — the oldest and most load-bearing application.
Targeting and lookalikes. Finding users resembling a converting set, without a human specifying the resemblance.
Creative. Generating and testing variants at a volume no studio would produce by hand.
[Fraud detection](/resources/glossary/mobile-ad-fraud). Classifying anomalous traffic, where the adversary is also adapting.
The surface itself. New in this generation: the model generates the content the advertising sits inside — see AI media.

The distinction that matters

There are two very different claims hiding under one word, and conflating them causes most of the confusion in the category.

AI-powered advertising means using models to buy conventional media more efficiently. The medium is unchanged; the buying got better. Almost every platform has done this for a decade.

Advertising in AI means the medium itself is a model's output. The user is in a conversation, the inventory is generated, and the mechanics of targeting, format and measurement all change. That is a different market, not a better version of the old one.

What models are bad at, in this domain

Two failure modes are worth naming because they recur.

Optimising into a biased measurement. A model will maximise whatever it is scored on, including artefacts of the attribution model. If last-click undercounts a channel, an optimiser will defund it faster and more confidently than a human would.

Confident wrongness at the edges. A generated recommendation that misstates a price or a policy is delivered with the same fluency as a correct one, which is why hallucination liability is a live commercial risk rather than a hypothetical.

Common questions

Is machine learning the same as AI?

Machine learning is the subset of AI that learns patterns from data rather than following written rules. In advertising the two terms are used interchangeably, and almost always mean machine learning.

Does AI targeting still work without cookies?

On AI surfaces it works better, because the signal is what the user just said rather than what they were inferred to be. Identifier deprecation removes an input the conversational surface never depended on.

Back to the glossary