On July 24, OpenAI pushed eight new features to beta advertisers inside ChatGPT Ads Manager: conversion-based bidding, budget flexibility, bulk campaign tools. On August 3 it went further, switching the daily budget rule to a seven-day rolling average. Advertisers running campaigns in that window are reporting something specific: the platform's stated $60 CPM ceiling is clearing at roughly $25 in practice, with recommended CPCs sitting at $3 to $5. You cannot buy an impression that cheap on Google right now. You can on ChatGPT. For now.
Why this price has an expiration date
This price won't hold because new ad inventory always sells cheap while demand is still catching up to supply, and it stops being cheap the moment that gap closes. A $25 CPM on a platform with over 900 million weekly users isn't a surplus of impressions, it's a shortage of advertisers relative to the traffic. OpenAI is still filling the pipe. The bidding model itself is a tell: billing still runs per click while the system optimizes toward conversions, which means the platform is currently gathering conversion data at your expense, cheaply, before it prices that data back into the auction. Every major ad platform has run this exact play once: cheap reach while the sales team builds a self-serve auction, then a multi-year climb once enough advertisers are bidding against each other for the same slots. The seven-day budget averaging OpenAI shipped on August 3 is not a courtesy, it's plumbing for the demand that's coming.
Why your brain can't flag this as an ad
Your brain can't flag it because the reflex that spots an ad in a feed developed around a visual boundary: a banner, a sponsored label, something that looks different from the content next to it. ChatGPT has no such boundary. The recommendation and the ad arrive in the same sentence, the same voice. A January 2026 Ipsos survey found 63% of US adults say ads inside AI search results lower their trust in the output itself. A separate Partnercentric study put the share who say a chatbot ad lowers their trust in the advertised brand at 57%. Add two smaller but sharper numbers: 42% of users cite simply seeing too many ads as a concern, and a third worry the model's answers are quietly leaning toward whoever paid. Zappi's May 2026 research draws a finer line: consumers aren't reflexively against the format, their trust breaks specifically when an ad interferes with a recommendation. The problem was never the ad. It's the collision.
Why OpenAI's own trust numbers deserve a discount
OpenAI's trust numbers deserve scrutiny because the company selling the ad is the same one measuring its effect on trust. Publicly, OpenAI reports no measurable dip in trust metrics and low ad-dismissal rates. That claim may be accurate. It is also not neutral. Independent research, Ipsos, Partnercentric, Zappi, tells a different story: users notice the ads and pull trust back proportionally. Both readings are measuring the same event through a different lens, and which lens your budget follows is a decision, not a default. Treat the gap between the two readings as a cost you haven't priced yet. It will show up eventually, either as a platform-wide trust correction that raises everyone's CPC, or as a brand-specific one the first time your ad breaks the conversational register in a screenshot that circulates.
- Test now, scale later. A $25 CPM is a learning budget, not a scale budget. Don't size the test like it's the program.
- Know the billing gap in the new bidding model. The system optimizes toward conversions but still charges per click. Build your margin math around that gap, not around the goal.
- Write for the advisory tone. An ad that breaks the conversation's register loses more trust than the format itself does.
- Never take trust data from one source. Run the platform's self-reported numbers against independent research and track the delta.
In a new ad channel, the cheap part is never the CPM. It's the question nobody has priced in yet.
What this means for your next campaign
Put a small slice of budget in now, run the new conversion bidding, and feed real conversion data through OpenAI's Conversions API early. Advertisers feeding data in early get a better optimization curve than the ones who show up after demand catches up. But run the test with copy written for the register of the conversation, not lifted from your display ad library, because that's exactly where Zappi found the trust fracture. Collect the price advantage and the data advantage before the window closes. Neither comes this cheap again.
Want this kind of thinking on your brand?
We build brand strategy, AI content and performance for the AI era.
START A PROJECT →