OpenAI Is Running Ads Now. The Early Numbers Are Not Good.

OpenAI spent most of 2024 insisting ads were a last resort. Then it paused its own ad rollout in December after a company-wide code red to fix product quality. On January 16, 2026, it announced ads anyway — for the free tier and a new $8/mo Go plan — and has been quietly expanding the test ever since. Reuters reported Thursday that the rollout is weeks away from reaching all free and Go users in the US.

The question that most coverage has not answered: is this actually working?

The early data says no. Adthena, an ad intelligence firm tracking the ChatGPT pilot, found that one brand in the test had a click-through rate of 0.91% — against a 6.4% benchmark for Google search ads in the same sector. That is nearly seven times lower. A separate enterprise advertiser committed $250,000 to the pilot and spent only 3% of it. And the Ad Manager tool OpenAI provides to advertisers reportedly has a reporting glitch that blocks campaign visibility. Advertisers are flying partly blind.

The pilot officially ends March 31. WPP, Omnicom, and Dentsu — three of the largest agency holding companies — are in it. Individual brand commitments have run $200,000 to $250,000, roughly double a typical experimental ad buy, which suggests serious intent from the agency side even as performance has been modest. Criteo, the ad tech firm best known for retargeting display ads, has been integrated as the backend.

The Premium Paradox

OpenAI's pitch to advertisers is theoretically compelling: conversational AI generates high-intent moments. When you ask ChatGPT how to design a kitchen, you are not idly scrolling. The intent signal should, in theory, beat keyword-based search.

The execution has a structural problem. The most commercially valuable ChatGPT users — the ones paying $20/month for Plus, or more for Pro, Business, or Enterprise accounts — are permanently excluded from ad targeting. OpenAI explicitly guarantees them an ad-free experience. Advertisers are paying to reach the users who decided not to pay, which is, by definition, the lower-engagement end of the user base.

This is the inverse of how most ad platforms work. Google charges a premium for users who search with high commercial intent. OpenAI is charging for access to users who could not afford or did not want to pay eight dollars a month.

That does not make the model unworkable. Free-tier users number in the hundreds of millions — OpenAI reported 400 million weekly active users as of February 2025, and the platform has grown since. Scale can compensate for lower per-user value. But the relevance problem still has to be solved. StubHub ads appearing during an AI conversation about something unrelated to events, Ancestry ads during a name etymology query — these are the placements a PCMag reporter documented in February. They read less like search-intent targeting and more like contextual keyword matching with the guardrails off.

What Last Resort Actually Meant

In October 2024, Sam Altman said in an interview that advertising was kind of like a last resort for OpenAI's business model. The framing was easy to misread as principled resistance. The more accurate read, with hindsight, was sequencing.

OpenAI's revenue grew from $2 billion in 2023 to $6 billion in 2024 to $20 billion in 2025 — a 3x year-over-year jump driven almost entirely by subscriptions and API fees. CFO Sarah Friar has framed advertising as a third leg of a diversified revenue model, not a replacement for subscriptions. When Altman paused the ad program in December during the quality code red — delaying not just ads but AI shopping, health agents, and a personal assistant product called Pulse — he was making a reasonable call: ads next to a degraded product are worse than no ads.

The January relaunch came after that quality work. The sequencing, at minimum, was intentional.

The financial logic is clear enough. OpenAI burned roughly $8 billion in cash in 2025 on compute and infrastructure. Projected cumulative losses by 2026 are around $14 billion. Subscription revenue at $20 billion ARR is strong, but the compute costs scale with capability. Truist analysts have modeled OpenAI at under $1 billion in ad revenue for 2026, growing toward $30 billion by 2030 if adoption tracks. The $30B figure assumes the relevance and measurement problems get solved.

The Competition That Is Not Running Ads

Anthropic bought a Super Bowl ad slot in February to say it was not going to run ads. Claude will remain ad-free — as a positioning statement, it was direct. Altman called Anthropic dishonest, which suggests the competitive sting landed.

The more structurally interesting holdout is Google. Gemini, Google's AI assistant, has not introduced chatbot ads. Google does run AI Overview ads in search — small business ads placed beneath AI-generated summaries — but has not extended ads into the conversational Gemini product. This is notable because Google has $252 billion in annual search ad revenue at stake. If conversational AI replaces search, Google needs to have an ad model for it. But if Google puts ads in Gemini, it is simultaneously validating OpenAI's model and risking the perception that AI assistants are just glorified ad delivery systems.

Demis Hassabis in January criticized OpenAI's monetization rush. That framing, from the CEO of a company sitting on top of the world's largest ad business, deserves scrutiny.

Perplexity launched ads in 2024 and quietly removed them. No one has said whether they are planning to try again.

What the March 31 Deadline Means

The pilot ends this month. The question for WPP, Omnicom, and Dentsu is whether performance data from the test — even with its reporting gaps — is good enough to commit to the next phase. Early CTR numbers alone would argue for walking away. But the agencies are also reading a platform that went from 1% of mobile users seeing ads on March 1 to 5% by mid-March, with Sensor Tower clocking a 600% volume increase in three weeks. Scale is arriving fast. Relevance may follow.

The real signal will come from whether any of the three major agencies renews after March 31, and at what commitment level. If they pull back, OpenAI will have confirmed that the first serious test of conversational AI advertising produced mixed results and a broken measurement tool. If they stay, OpenAI will have won the agency relationship even before it has proven the format.

The data so far suggests the format needs work. The business pressure suggests it is going to get tested at scale regardless.

Primary sources: OpenAI blog posts January 16 2026 — our approach to advertising (https://openai.com/index/our-approach-to-advertising-and-expanding-access/) and testing ads in ChatGPT (https://openai.com/index/testing-ads-in-chatgpt/); Reuters March 21 2026; Adthena via Winbuzzer; Sensor Tower mobile data; CNBC pilot partner reporting March 20.