From sponsorship to open auction
For most of its existence, ChatGPT ads worked like a publisher deal: OpenAI selected who showed, set the floor price, and required $50,000 to $200,000 to participate. You weren't competing against anyone. You were buying a slot.
That changed this week. OpenAI began testing multi-advertiser ad units where multiple brands compete for a placement through a second-price auction. The mechanics are the same ones Google Ads and Meta have used for years: you set a bid, the system ranks by relevance-weighted value, and the winner pays one cent above the second-highest bid.
The entry minimum has dropped to $10,000. The self-serve Ads Manager is open to any business. And now there's a real auction underneath it.
The reason this matters: right now, there isn't much competition. That's the window.
What the format looks like in practice
Users see a single ad unit below a ChatGPT response. That unit can contain ads from one or more advertisers, with the most relevant match to the conversation appearing first. Placement is tied to commercial-intent queries: product research, comparison questions, purchase decisions.
The targeting surface is meaningfully different from keyword search. When a user asks ChatGPT "what's the best protein powder for lean muscle under $40," the conversation leading up to that question is part of the signal. That's intent qualified by context, not just keyword match.
Early performance data supports the distinction. Through Criteo's integration, brands are reporting CTRs roughly 3x higher than comparable formats in other environments. Conversion rates from ChatGPT-referred traffic run about 1.5x above other referral channels overall, and closer to 2x for consumer electronics, home goods, and lifestyle categories.
Those numbers will compress as the auction fills. They always do. But you act on signals like that before compression, not after.
Custom audiences: what's live right now
The second significant change arrived alongside the auction: custom audience upload. Advertisers can now upload customer email or phone lists under an "Audiences" tab in the Ads Manager and use them to control who sees their ads.
Two immediate uses:
Exclusion: if you're running acquisition campaigns, uploading your existing buyer list and excluding them from targeting removes obvious waste. Spending acquisition budget on people who already bought from you is the oldest problem in digital advertising. The tool to fix it on ChatGPT now exists.
Direct targeting: you can serve ads specifically to contacts from your list, useful for lapsed customers, high-AOV cohorts, or segments you want to reach with retention messaging rather than cold acquisition copy.
The match rate is partial. Only contacts whose details match an existing ChatGPT account will surface. Expect 20 to 40 percent match, not full list coverage. Treat the audience as directional, not complete. That's normal platform behavior for a network that's still building its logged-in base.
ChatGPT also added AI-generated ad creative this month, letting the platform write ad variations itself from your product information. You review and approve before anything serves. It's a workflow shortcut, not a creative strategy.
The platform's progression and what it signals
Where ChatGPT ads started versus where they are now:
CPMs launched at $60 and dropped to roughly $25 as supply expanded. CPC bidding launched at $3 to $5 per click. Daily budgets and DMA-level geo targeting arrived in May. CPA optimization came in June. Now: second-price auction, custom audiences, and AI-generated creative.
Over 2,000 brands are live through Criteo. The pilot crossed $100 million in annualized revenue within six weeks of the Criteo integration launching. OpenAI has expanded from four English-speaking markets to the UK, Japan, South Korea, Brazil, and Mexico, with EU markets requiring explicit consent under GDPR.
Platforms don't build audience targeting infrastructure for a test. They build it when the test becomes a channel.
Who this matters most to right now
If you sell considered purchases where the buyer researches before deciding (supplements, home furnishings, software, consumer electronics, high-AOV beauty), ChatGPT's conversational surface aligns with your category in a way keyword-triggered search ads don't. The user is already in a decision-making conversation before your ad appears.
If you have a real customer list of 50,000 or more emails, the custom audience capability gives you a structural advantage over advertisers that don't. Exclusion and targeting that your competitors can't yet run because they haven't set up the infrastructure.
If you sell impulse purchases or commodity products under $30, the economics probably don't work yet. The intent context is real but the conversion math only closes at higher margins.
The one thing to do before you spend anything
Install the OpenAI conversion pixel and configure the Conversions API before running any budget. Without return signal, the platform optimizes at CPM level and you'll have no defensible measure of actual CPA.
Once tracking is live, uploading your customer list takes under an hour. Set up exclusion first. Then run a narrow test on your highest-intent, highest-AOV product category with the auction bidding, and let the platform run for at least two weeks before reading results.
The auction just opened. Most advertisers haven't walked in yet, and the ones who have are mostly running without audience data. That combination produces the best performance windows in digital advertising — and those windows close fast.
If you want a clear read on where your tracking and conversion setup stands before testing a new channel, the free audit at Gromerce shows you in three minutes.
Sources: Search Engine Land, MediaPost, Criteo Press Releases, OpenAI Blog, MarTech, Adsroid, June–July 2026
