What is targeting in advertising?

The main types

• Demographic. Age, gender, location, language. Cheap, available, and usually weakly predictive alone.
• Behavioural. Based on past actions. Historically the strongest, and the most affected by identifier loss.
• Contextual. Based on the content the user is engaging with right now. Needs no personal data.
• Lookalike. Modelled resemblance to a converting seed audience.
• Retargeting. Users who already interacted — see retargeting.
• Conversational. Based on what the user has just said they are trying to do — see synthetic audience.

Why third-party targeting collapsed

Behavioural and lookalike targeting depended on recognising individuals across sites and apps. Third-party cookie deprecation, ATT and platform restrictions removed the identifier that made it possible.

What survived is targeting that does not require cross-company identity: your own first-party data, contextual signals, and platform-internal audiences inside walled gardens where the identity never had to leave.

The commercial consequence is that context regained value it had lost for a decade, and that the largest platforms became relatively more attractive because their internal targeting was unaffected.

Over-targeting

Narrow targeting feels efficient and frequently is not. It raises conversion rate while shrinking reach, drives up CPMs through competition for a small pool, and deprives optimisation algorithms of the data they need to find people you did not think of.

Modern platforms generally perform better with broad targeting and a clear conversion objective than with narrow manual segments, because the model can explore. The instinct to define the audience precisely usually costs more than it saves.