On May 5, 2026, OpenAI published a blog post titled "New ways to buy ChatGPT ads" and quietly rewrote what its ad business is. The self-serve Ads Manager at ads.openai.com is now open to any US advertiser. The $50,000 minimum spend that gated the pilot is gone. Cost-per-click bidding is live alongside the original CPM auction. A conversion pixel and Conversions API are live for tracking post-click events. Cost-per-action bidding is in development. Agency partners — Dentsu, Omnicom, Publicis, WPP — are joined by ad-tech vendors Adobe, Criteo, Kargo, Pacvue, and StackAdapt.

That is the entire announcement, and it is bigger than it looks. Three months ago ChatGPT had no ads in production; six weeks ago the pilot needed a $250,000 minimum and a meeting with an Omnicom rep. As of yesterday it is a credit-card product with a $3-to-$5 starting bid recommendation and a pixel you paste into your site footer.

The internal target — reported across Axios and TheNextWeb — is $2.5 billion in ad revenue this year, $11 billion in 2027, and $100 billion by 2030. The shape of the company is changing in public.

The chronology

A useful way to read this week's update is against the four-month timeline of how fast the floor has fallen. The numbers are pulled from PPC Land's tracker and Digiday's coverage, with cross-checks against TheNextWeb.

• February 9, 2026 — ads launch in ChatGPT Free and Go tiers. CPM model only, $60 starting CPM, $250,000 minimum spend, agency-mediated buying.
• April 10, 2026 — minimum spend cut to $50,000.
• April 15, 2026 — CPC bidding announced. CPM erodes to the $25–$45 range, down from $60 launch.
• May 5, 2026 — self-serve manager open to all US advertisers; $50K minimum gone; pixel and Conversions API live; agency and ad-tech partner roster expanded.

That is a 100% reduction in floor, a CPC option grafted onto a CPM auction, and a measurement stack assembled in twelve weeks. The pilot is over. What is shipping is the product.

What CPC means here

CPM auctions are the traditional ad model: pay for a thousand impressions, hope someone notices. CPC means pay only when somebody clicks the link OpenAI rendered next to a model response. The shift matters because CPM ads in a chat surface are an awkward fit — the user did not come to read banners — while CPC ads can ride the back of a question the user already asked. "Many ChatGPT conversations are active and decision-oriented," OpenAI wrote in its post, "a click can be a meaningful signal."

The auction underneath is a relevance-weighted, second-price auction, which is the same architecture that runs Google AdWords. Recommended starting bids of $3–$5 per click slot ChatGPT in roughly the same range as competitive Google Search keywords, well above Meta's CPCs (which run three to five times cheaper than Google Search, per Digiday's reading) but below the very top of the Search market. That positioning is the thesis: a ChatGPT click should be worth what a Google click is worth, because the intent behind a chat question is at least as concrete as the intent behind a query.

Whether that holds in the data is the open question. Claire Holubowskyj, senior research analyst at Enders Analysis, told Digiday that "OpenAI's experimentation with CPC is driven largely by its need to maintain demand growth and build trust." The polite reading. The blunt reading is that $60 CPMs collapsed to $25 in ten weeks because advertisers did the math on impressions in a chat surface and did not love the answer.

The pixel and the API

The other half of the announcement is the measurement stack. A site-side conversion pixel went live, alongside a Conversions API for server-to-server event posting. Reporting in the manager now covers impressions, clicks, spend, click-through rate, average CPC, average CPM, and conversions — the same column set every advertiser already runs on Meta and Google.

OpenAI says individual conversation data is held back from advertisers; only aggregated metrics flow out. That is the right answer on privacy and also the only answer that survives a regulatory lookup, given the existing scrutiny over how ChatGPT logs personal context. It is also the answer that makes the pixel the load-bearing piece of the new system. Without conversation-level signals to share, OpenAI has to prove its click value through standard last-click attribution, the same way a smaller display network has to.

Hence the hire OpenAI quietly placed two weeks ago: an advertising marketing science leader, brought on to build attribution models, incrementality testing, media-mix modelling, and geo experimentation. Nicole Greene, VP analyst at Gartner, told Digiday what that role is for: "This consistent measurement will help advertisers justify reallocation of spend to OpenAI." In plain English, somebody has to convince a CMO that a ChatGPT click is worth a Google click. That argument is made in spreadsheets, not in keynotes, and OpenAI just hired the person who builds the spreadsheets.

The partner list

The agency roster — Dentsu, Omnicom, Publicis, WPP — is the same big-four cohort OpenAI lined up for the February pilot. What changed is the ad-tech partner row: Adobe, Criteo, Kargo, Pacvue, StackAdapt now plug in directly. Pacvue's Melissa Burdick told Adweek that "conversational AI is the most significant new channel since the rise of retail" — a quote with a comparable sales pitch energy to the one Pacvue's whole business depends on, but also a quote that exists because Pacvue now has commercial reasons to say it.

The roster reads as a deliberate hedge. The big four cover the brand-side trust and budget; Pacvue covers retail media; Criteo covers performance and its claimed 17,000-advertiser network; Adobe covers the creative tooling; Kargo and StackAdapt cover programmatic. Each one is a wedge into a different buying motion. Each one is also a vendor that wants ChatGPT to work, which is the polite version of saying nobody on this list will be the source of the first hostile leak.

What this means for the rest of the AI stack

This is the part of the story that will be replayed in the next earnings call.

OpenAI's public framing on ads in 2024 and early 2025 was that they were a contamination of the chat experience and probably never coming. The framing in 2026 is that they are a primary product line with a $100 billion target. The intermediate step — Fidji Simo joining as CEO of Applications, hired specifically to build the consumer business — was the load-bearing personnel decision the rest of this follows from. Anyone surprised by today's announcement is reading the press releases instead of the org chart.

The implication for Anthropic is more interesting. Anthropic shipped ten finance-agent templates the same day and pushed Opus 4.7 deeper into Microsoft 365. That is a B2B/seat-license bet against OpenAI's emerging B2C/ad-revenue bet. Both companies are now choosing different revenue mixes in public. Twelve months from now one of those mixes will look obvious in retrospect and the other will look like a category error, and the press releases this week do not tell you which is which. The Loop's view: a chat product where the answer ranks against ad bids is structurally different from a chat product where the answer doesn't, and the convergence everyone has been predicting between Claude and ChatGPT on UX is going to bend at the auction layer, not at the model.

The other implication is for Google. Google Search ads are the reference price for the auction OpenAI is now running. If ChatGPT clicks settle at parity with Search, OpenAI captures meaningful revenue without taking obvious share away from Google's existing book — the user who asks ChatGPT instead of Googling is a user Google was going to lose anyway. If ChatGPT clicks settle at a premium, OpenAI takes share. If they settle at a Meta-style discount, the $100B target is in trouble. The number that determines which scenario we are in is not in any blog post; it lives in the bid logs of advertisers who have access to both.

What to watch

1. What the May reporting looks like for advertisers running matched campaigns on Search and ChatGPT. The CMOs who will set the pace of OpenAI's growth are the ones running A/B tests across both channels. The first leaked CPC parity number — or the first leaked discount — will set the public narrative for the rest of the year. Watch the Digiday and PPC Land trade press feeds, not the OpenAI blog.
2. Whether ChatGPT Plus and Pro stay ad-free. Free and Go tiers are in scope today. The original Fidji Simo post said ads would not influence the answer ChatGPT gives you; it did not say ads would never appear next to a paying user's chat. The day the line moves on Plus is the day the user-revenue ceiling matters less and the auction matters more.
3. The Conversions API integrations Anthropic does not build. A measurement layer like OpenAI's is a competitive moat the labs without one will have to navigate. If Anthropic, DeepMind, or DeepSeek announce equivalent first-party conversion tracking inside their consumer surfaces, the ad model has won the strategic argument across the industry. If they don't, the seat-license thesis is still alive and the labs have actually picked different lanes.

The frame the trade press reached for this week was "OpenAI launches self-serve ads." That undersells it. The pilot was OpenAI building an ad business on a side desk with four agency partners. The product is OpenAI building an ad business on the same architecture Google did, in the same auction primitive Google did, with a measurement stack hired to compete with Meta's, and a target that says they think they can take a hundred billion dollars of media spend by 2030. The sentence that should follow every keynote about ChatGPT's product roadmap from this point is the one OpenAI did not put in the post: the model now has somebody pricing the back of every answer.