AI products like ChatGPT are doing Google-style search ads, but they should learn from Instagram's discovery model instead. Discovery ads build user profiles to anticipate needs rather than react to what someone just asked, which can create a conflict of interest.
Podcast: Ben Thompson from Stratechery on AI ads, the end of SaaS, and the future of media
How should AI companies make money from ads?
Ben Thompson argues AI platforms like ChatGPT are doing it wrong. Currently, they are experimenting with ads targeted to the context of a user's conversation, much like Google Search ads.
They match ads to the current prompt. It is easier to implement, but it creates two problems:
First, trust: users may doubt whether answers are objective. They will wonder whether the AI shaped its answer to serve the ad.
Second, scale: ad opportunities are limited to narrow prompt contexts. You can only show an ad if the user happens to ask about something an advertiser targets.
The alternative is Meta's model, like Instagram and Facebook ads. Instead of reacting to what you are doing right now, it builds a profile of your broader interests and shows you products you didn't know you wanted.
In other words, the strength should be discovery, not intent.
The ideal AI ad model would not wait for you to ask about winter jackets. It would know your interests and the changing seasons and show you the perfect jacket before you even search.