What ChatGPT Ads actually change for B2B SaaS

Your buyers have quietly changed how they build a shortlist. Instead of running three searches and opening eight tabs, a lot of them now ask ChatGPT which tools to consider, then go straight to two or three vendor sites. That behavior was invisible to advertisers until this year.

In February 2026, OpenAI started showing ads inside ChatGPT. By May there was a self-serve Ads Manager, which removed the large minimum spend that gated the early pilot. For the first time, you can buy placement inside the conversation where your category gets evaluated.

The targeting model is not keywords

This is the part most teams get wrong in their first test. There is no keyword list and no match types. Placement is contextual: the ad is matched to what the conversation is about, drawing on the current topic, previous chats, and how the person has interacted with ads before.

Practically, that means you are buying a topic and a moment, not a string. The implications:

You cannot negative your way to precision the way you can in Google Ads. Control comes from creative and landing page relevance, not exclusions.
Your ad copy has to make sense immediately after an AI has just explained the category. Someone arriving from a ChatGPT answer is usually better informed than someone who clicked a search ad.
Branded defense works differently. If a buyer is asking for alternatives to a competitor, that conversation is the placement.

What it costs right now

Bidding runs through a relevance-weighted auction, so creative quality affects your rank alongside your bid, and cost-per-click bidding is available. Early reported pricing has clustered around a few dollars per click, with thousand-impression costs well above display inventory.

Judge those numbers against your pipeline math, not against your Google cost per click. A more informed visitor who already has you on a shortlist is worth more than a cold click, and that difference shows up in close rate rather than cost per lead.

How to test it without ruining your reporting

The failure mode we see is simple: a new channel gets switched on, leads arrive with no source, and three months later nobody can say whether it worked.

Fix tracking first. Decide what the conversion is, make sure it writes back to your CRM, and confirm it fires before any budget goes live.
Set a fixed test budget and an end date. Treat it as an experiment with a decision at the end, not a line item that renews by default.
Give it a dedicated landing page. Write it for someone who just read an AI summary of your category, not for someone who typed your keyword.
Measure to pipeline. Clicks and cost per lead will tell you almost nothing in the first 60 days. Opportunities will.

Where this is heading

The deeper shift is that AI assistants are becoming a place where buyers form opinions before they ever reach your site. Ads are one way to show up there. Being the source the model cites is the other, and it is the one most B2B teams have no plan for yet.

If you are spending real money on Google and LinkedIn, ChatGPT Ads is worth a controlled test this quarter. Just run it like an experiment with tracking in place, not a land grab.