ChatGPT Ads Hit Logged-Out Users (Despite OpenAI's Own Help Page)
OpenAI has begun serving ChatGPT ads to users who are not signed in, per AdExchanger reporting from April 23, 2026. OpenAI's own help page still states that during the initial testing phase, ads will not appear to logged-out users. The shift pushes ChatGPT's ad surface from a roughly 900M-weekly-signed-in audience toward something closer to open web inventory, and the advertiser auction has not caught up.
The help page says one thing, the product does another
OpenAI has not announced the change. No blog post, no investor note, no update to the publisher docs. AdExchanger confirmed it the way most product changes get confirmed in 2026: a non-industry user noticed ads in ChatGPT while logged out and mentioned it to a reporter. She was served two ads during a conversation about resume advice, one for Canva and one for JobCopilot. Meanwhile, OpenAI's own help documentation continues to say ads will not appear to logged-out users during initial testing.
For most users this is a minor content policy detail. For anyone running a ChatGPT ad buy or auditing one, it matters more than the frequency cap. Your campaigns are now running against a population OpenAI's own docs say they shouldn't be. If your brand safety agreement cites "logged-in US consumers" (which several pilot agreements reportedly do), the default inventory routing is no longer what you signed for.
The inventory pressure made this inevitable
Go back three weeks. OpenAI cut its advertiser minimum from $200,000 to $50,000 and launched a self-serve manager. That's the kind of change you make when enterprise buyers are pulling back, which is exactly what happened. Digiday reported in the same window that ChatGPT ad prices fell fast because pilot advertisers could not hit frequency. They had committed to budgets the inventory could not absorb.
Adding logged-out users to the eligible pool is the easiest lever OpenAI has. ChatGPT reportedly hit 900 million weekly active users in February 2026, but the logged-out population is measurably larger. A meaningful share of ChatGPT traffic is anonymous, particularly on mobile web and embedded referrals. Flipping ads on for that cohort probably doubles available impressions overnight.
Which is also why it's happening quietly. Announcing "we're now running ads to anonymous sessions" gets picked apart by everyone from privacy advocates to AP style guides. Not announcing it lets the inventory scale first and the press release follow later. That's textbook inventory engineering, and I think it's the cleanest read of what just happened.
What logged-out inventory is actually worth
This is the part that should change your bid strategy.
Logged-out users give you less signal. No conversation history, no stored preferences, no custom GPTs, no account demographics. That's why display CPMs collapsed when Safari's ITP killed most third-party cookies, and a version of the same dynamic applies here. PPC Land reported ChatGPT CPMs dropping to around $25, down from earlier benchmarks above $60. Some of that is scale normalizing. Some of it is OpenAI adjusting to a CPC model. But some of it is almost certainly the logged-out cohort dragging the blended average down, because anonymous impressions are worth less in any auction.
From what I've seen in comparable shifts (TikTok's logged-out web experience, Reddit's non-authenticated pool), the CPM floor for anonymous inventory tends to sit 30-50% below the logged-in rate. If OpenAI is reporting a $25 blended CPM, the logged-in half is probably closer to $35 and the logged-out half closer to $15. That's an educated guess, not a leak.
The practical read: if you're bidding against a blended pool, you are overpaying for the logged-in half and underpaying for the logged-out half, and OpenAI controls whether you ever see that split.
The first buyers are paying to learn
AdExchanger named Canva and JobCopilot as two early advertisers appearing in logged-out inventory. Neither is a coincidence. Canva is a design tool with near-universal awareness and no prerequisite for prior context. JobCopilot is a direct-response resume assistant that wins on CTR in a resume conversation regardless of who the user is. Those are exactly the advertiser profiles that don't need audience signal to convert, which is why they're the right buyers for an anonymous-heavy surface.
If you are running performance ads that depend on intent modeling (B2B SaaS, high-ticket consumer, anything with a 14-day consideration window), the logged-out cohort is probably the worst inventory you can buy right now. If you're running broad-reach awareness for a tool with horizontal demand, it's the best deal in the room. Same $25 CPM, very different outcomes.
The first wave of pilot advertisers overpaid for logged-in ChatGPT; we covered the CPM compression from $60 to $25 in nine weeks last week. The second wave is paying to learn what logged-out intent is worth. The third wave will know.
How to bid this week if you're still in the pilot
Three things to check before Friday.
First, ask your OpenAI rep for a logged-in vs logged-out delivery split. They may not share it, but ask on the record. If the split is not in the reporting UI, treat that as a signal. Any channel that won't break out audience authenticity is a channel that's blending prices against your budget.
Second, look at your creative. Logged-out users have no conversation history, which means your ad's first 30 characters are doing 100% of the contextual work. If your copy references "as we were discussing" or assumes any kind of conversation thread, it's misfiring against a chunk of inventory. Rewrite to standalone.
Third, if you can segment bids, bid lower on any inventory segment flagged anonymous, broad-reach, or horizontal. The $5 CPC cap OpenAI introduced doesn't care whether the click came from a logged-in regular or a drive-by user searching for "resume tips." You should.
The part nobody is saying out loud
Running ads to unauthenticated AI queries is a different business than running ads to an authenticated assistant, and OpenAI hasn't really positioned this yet. It's not Google-scale search. It's not Meta-scale feed. It's closer to what Quora used to be before the AI walled garden ate it: a mass-reach surface where the intent signal sits in the query text itself and user identity is optional.
I don't know if that's good or bad for performance marketers. Probably both. What I do think is that the pilot agreements signed in Q1 2026 were priced for a product that no longer exists, and the renewals in Q3 will look very different. The smartest move right now is to document your current delivery mix, so you can price the next contract against the actual inventory and not the one the help page still describes.
Notice Me Senpai Editorial
