The Maze: OpenAI is changing how advertisers bid for placements in ChatGPT. Since November 9, advertisers setting up a click or purchase campaign can tell OpenAI the average cost they want it to aim for. OpenAI then decides what to bid each time it has a chance to show an ad. Advertisers who previously capped each bid can set an average cost target or let OpenAI bid within their budget without one. The change affects advertisers with eligible ChatGPT Ads campaigns. OpenAI has rolled ads out across multiple regions, with access depending on the country and account.
The details
🎯 A target gives OpenAI room to vary bids. For a click campaign, Target CPC gives OpenAI an average cost to aim for per click. It weighs how likely a click is and what the bid would cost. Bids can go above or below the target. Target CPA does the same job for a conversion, an action such as a purchase or lead. Actual average cost can land above or below the target. Advertisers can also let OpenAI bid automatically within the budget without setting a target.
🧾 A purchase goal can still produce a click bill. A retailer can ask OpenAI to find more shoppers likely to buy while choosing to pay for valid clicks or for ad impressions. An impression is one showing of the ad. A purchase target guides where OpenAI competes for an ad slot. The billing option determines what generates a charge, including when someone clicks and leaves without buying.
📅 The campaign and software changes have separate timelines. Since November 9, advertisers have been able to move existing campaigns off bid caps in Ads Manager. The caps retire on January 1, 2027. For partners whose software manages campaigns through OpenAI’s API, the old bid fields keep their prior meaning through November 17. From November 18 through December 31, values in those fields become average cost targets only when a tool creates a campaign or changes an ad group’s bid type. OpenAI recommends using the dedicated target fields and will deprecate the old fields on January 1, 2027. Search Engine Journal reports that OpenAI’s guide leaves the starting target unspecified for existing campaigns that are still on caps at retirement.
Our take
Retail advertisers will get more out of the switch if they treat it as a change to how they measure a campaign. An average cost is useful when it comes with enough sales to justify the spend and when the store can see those sales in the first place.
📊 Start with what the campaign actually costs. OpenAI’s example puts $400 in spend against 10 purchases credited to the ads, or $40 per purchase. A $20 target in that situation could leave budget unspent if OpenAI finds too few promising auctions at that level. Compare spend, purchase count and cost per purchase over comparable periods. A better-looking average on fewer orders can still leave a retailer with less business.
🔍 Trace a purchase before raising the target. A website tracking tag can load without sending the completed purchase to Ads Manager. If the campaign is barely spending, check that the store sends the chosen purchase event and the campaign’s connected data source receives it. Weak signals can restrict delivery. A received purchase event confirms part of that tracking chain. The retailer still needs to check whether the platform credits the sale to the campaign.
Sources: OpenAI’s October ads expansion announcement · OpenAI’s ChatGPT Ads bidding guide · OpenAI’s conversion campaign guide · Search Engine Journal · OpenAI’s conversion tracking guide
