A maximum bid is a ceiling. A target is an average. On 9 November 2026 ChatGPT Ads swapped one for the other, and that single change decides whether the channel is safe for a small business to test.
OpenAI opened cost targets in Ads Manager that Monday: Target CPA for conversion campaigns, Target CPC for click campaigns, both replacing Max Bid and Max CPC on new campaigns. The old manual options retire on 1 January 2027, and campaigns still running them are converted automatically. The move itself is ordinary; Google and Meta did the same years ago. What is new is where it is happening — an ad surface inside an assistant that 1.2 billion people use every week, and that most small businesses have never bought, measured or sized.
What changed on 9 November?
Max CPC was a limit on what any single auction could cost you. Target CPC is not. OpenAI's own guidance says the system may bid above or below your target in each auction, working toward the average over time. The difference shows up on your worst day rather than your average one: a target of a few dollars a click does not stop one click from clearing well above it, and your only hard stop is the daily budget.
Target CPA leans on something most small accounts have never proven: that the conversion event you are bidding on actually arrives. OpenAI's requirement is explicit — the pixel or the Conversions API has to be sending the selected event to the data source linked to the campaign — and its warning in the same guidance is that restrictive targets or weak conversion signals can reduce delivery and leave budget unspent. Read that twice. The failure mode here is not an expensive click. It is a campaign that quietly does nothing while you sit waiting for data.
Why isn't this just another search line item?
Because of where the ads sit. The visual ads OpenAI launched on 5 October 2026 appear alongside images a user asked ChatGPT to generate, in the United States first. That is not a shopper typing a product query. It is a person in the middle of their own task, with the ad beside their work. OpenAI says the ads are clearly labeled and do not influence the answers ChatGPT provides — the right design, and also an honest description of what you are buying: attention, not intent.
Some share of those conversations does carry buying intent. Nobody outside OpenAI can tell you what share, because the surface publishes no query report of the kind you would read in a search account. So the sizing question is not what your cost per order will be there — you cannot know that before you spend — but what you are willing to spend to find out. Treat it as prospecting with an unknown intent mix, and the rest of the plan follows.
This is also not one vendor's experiment. Google is placing sponsored results inside AI Mode conversations, including the third and fourth follow-up questions, and testing summaries written above sponsored results. The assistant turn is becoming an ad slot across the board. Learning to measure one of them is the part that transfers.
A target is not a cap
The number in the bid field is now a direction, not a brake. Your real brake is the daily budget plus a conversion event you have read back yourself. Decide the week's loss you can live with, then launch.
What has to be true before you spend a dollar there?
Eight things, in order. None of them is about creative.
- Pick the single conversion event you will bid on.
- Fire it once by hand and find that row in the linked data source.
- Link the pixel or the Conversions API to the campaign before launch, not after.
- Fund the test with new money. Leave the search budget where it is.
- Set the daily budget to the weekly loss you accept, divided by seven.
- Start on clicks. Move to a cost-per-order target only after the event is proven.
- Keep a holdout: one region or one week in two with no spend.
- Write down today's orders per day before you launch anything.
A bid cap protected you from your own optimism. A target hands that job to your tracking, and most tracking has never been tested.
How do you know it worked?
Not from the platform's conversion column. Run two checks that can actually fail. First a read-back: fire the event you intend to bid on, then open the linked data source and find that one event. A tag reporting that it fired is a claim about intent; a row you can see is a claim about state. Second a holdout: keep one region, or one week in two, with no spend on this surface at all, and compare your own order count across the two. If your orders do not move, the channel did not work, whatever the dashboard says.
Then reconcile. Put the platform's claimed conversions beside your own records for the same window. If the platform claims more orders than you actually took, you have double counting, not growth — and finding that before you scale the budget is worth more than the test itself.
Audit before you add a channel
We write these guides from real client work — GEO, SEO, and builds we've shipped. If a piece raised a question about your own site, we're happy to look at it.
- Straight answers, no retainer required
- We reply in your language — EN · فارسی
- Based in Canada, working worldwide



