ChatGPT Ads do not read intent better than Google. The conversational format is why.

The pitch for ChatGPT Ads is that they read intent better than Google. The conversational format is the exact reason they do not.

OpenAI opened its self-serve Ads Manager beta this month, added CPC bidding, and the early coverage all repeats the same line: targeting plus intent, a signal Meta and Google cannot touch. That is the case being made across every feed right now.

Here is the problem.

Why search intent worked

Search intent works because the query is the intent. Someone types "best CRM for a 5 person team" and you bid on that exact moment. One query, one intent, one auction.

The narrowness is the value. The advertiser knows what the user is looking for because the user just typed it. The auction is for the specific moment of expressed need.

Why conversation breaks this

A ChatGPT session is not a query. It is six topics in one thread.

The user starts on CRMs, pivots to onboarding, ends up debugging a spreadsheet. And by OpenAI's own placement model, the ad surfaces later in the conversation, not at the moment the intent was expressed.

Business Insider's early data already named it: intent drift. The ad fires against the session, not the need. The thing that made paid search valuable was the narrowness. Conversation is wide by design, better for the user, worse for the targeting.

The user got better at exploring multiple topics in one place. The advertiser got worse signal about which topic the user was on when the ad fired.

What conversion tracking does not fix

Conversion tracking will tell you what happened after the click. It cannot tell you which of the six things the user was doing when your ad showed up.

You will see CPCs, click-through rates, and conversions. You will not see "this user was actively comparing CRMs when the ad fired" vs "this user had moved on to debugging a spreadsheet and saw the ad as an interruption."

Both produce a click. One converts. The other does not. The reporting layer does not distinguish them. The optimisation layer cannot learn from the difference.

Where ChatGPT Ads might still work

The format is not useless. The likely-good fits:

• High-consideration B2B where intent is sticky across a session (the user is researching the category for an hour)
• Categories with low query-intent variance (everyone asking about [product] is asking the same kind of question)
• Brand advertising where the goal is awareness, not direct response, and "they saw it" is enough
• Niche categories where the inventory is thin and any exposure beats no exposure

The likely-bad fits:

• Direct-response B2C where the auction value depends on tight intent matching
• High-volume search categories where the alternative (Google Search) has cleaner intent signal
• Anything where the conversion event is far downstream and attribution gets noisy

What to actually do

• Test ChatGPT Ads with a small budget on categories that match the likely-good fit profile.
• Build attribution that distinguishes ChatGPT-driven conversions from Google-driven ones, so you can measure cleanly.
• Set conversion-tracking expectations low. The attribution will be messier than search for the same reasons the targeting is messier.
• Do not redirect significant Google Search budget to ChatGPT without an apples-to-apples test on the same category.
• Read OpenAI's own placement and reporting documentation. The format is changing fast.

The cheap beta CPCs are tempting. Before you chase them, what intent are you actually buying?

If you want help running a structured ChatGPT Ads test against your current Google Search performance, book a free audit.