ChatGPT ads open self-serve in Europe, but your buyers are on the ad-free tier

OpenAI opened self-serve access to ChatGPT Ads across Europe on Monday, the same day it said the advertising business is pacing at a $1 billion annualised run rate less than six months after launch. The money number will get repeated. The access change is the one that lands on your media plan.

Until August 31, buying inside ChatGPT in European markets meant going through one of OpenAI's selected agency partners. Digiday reports that beta self-serve access to Ads Manager is now open directly to eligible advertisers across the same 31 European markets where ads went live earlier in August, so a small software company can run a campaign without an agency in the loop. The Austrian trade outlet Trending Topics puts the wider self-serve footprint at India, Europe, the Middle East and North Africa, and reports OpenAI's claim of tens of thousands of advertisers already on the platform.

Read the billion carefully

A run rate is not revenue. Digiday is explicit that the figure comes from multiplying current monthly ad revenue by twelve — a snapshot of today's pace, not money booked. On Digiday's own arithmetic that is roughly $83 million a month, implying closer to $330 million actually collected across the first eight months of the year. Dave Dugan, OpenAI's vp of global ad solutions, told Digiday that "Reaching $1 billion in ARR in under 200 days shows the scale of the opportunity ahead." OpenAI's own announcement, titled "A milestone in expanding access to AI," frames the revenue as what keeps the free tier viable.

It is a fast start for a channel that did not exist in January, and it behaves nothing like paid search. Ads appear only to users on ChatGPT's Free and Go tiers; Plus, Pro, Business, Enterprise and Education accounts stay ad-free. Targeting is contextual rather than keyword-based — conversation topic, approximate location, device and language, per Trending Topics — with personalisation promised later on an opt-in basis.

The eligibility problem is a B2B problem

That tier restriction is not a footnote. It is the structural issue for anyone selling software to people who use AI heavily, and the sharpest evidence comes from a company in exactly that position. SE Ranking — an SEO and AI-visibility tool vendor, so read its findings with that interest in mind — published a study on August 10 covering 50,006 commercial prompts across 20 US niches, alongside an account of running its own campaigns.

Its ads worked mechanically and failed commercially. Three campaigns across the US, Canada, Australia and New Zealand produced more than 97,000 impressions and 1,263 clicks at a 1.30% click-through rate — then almost no sign-ups, and the company stopped. Vadym Hryshchenko, a PPC specialist there, named the reason: "ads only reach Free and Go users, and the more someone looks like our ICP, the more likely they're on a paid, ad-free plan."

If your buyer is a developer, a security lead or an agency owner, that describes your funnel too. The people most likely to evaluate your product are the people most likely to have already paid to remove the ad slot.

Ad spend buys attention next to the answer. It does not buy a place inside it.

Ads do not buy you a recommendation

The same study measured whether an advertiser also showed up in the AI answer above its own ad. In 96.37% of placements it did not; the advertiser's domain was cited just 3.63% of the time, against 11.53% in Google's AI Mode. Ad frequency was comparable across the two, but 14.35% of ChatGPT ads had no more topical connection to the prompt than a randomly paired one would. Search Engine Land, owned by Semrush, covered the same research.

Reporting is thin, too. Jason Dodge of BlackTruck Media + Marketing, quoted in the study, said of the auction: "There's no Quality Score, no search terms report, and no way to know what prompt you appeared within." He put observed CPCs at $1.50 to $4.00. OpenAI points to its own case studies — Trending Topics cites a 3x return on ad spend for one ecommerce advertiser over 28 days — a plausible shape for retail, a poor proxy for a 90-day enterprise software cycle.

What a software marketer should actually do

Neither ignore this nor move budget into it wholesale. Self-serve access drops the cost of finding out to a test budget, and cheap attention in your category is worth measuring even when the audience is imperfect.

• Check tier eligibility before creative. Estimate what share of your ICP pays for an ad-free plan; if it is most of them, cap the test and say so up front.
• Install conversion tracking before the first euro. SE Ranking's team said they would start on a conversions objective rather than manual CPC bidding if launching now.
• Expect no query-level diagnostics. Build your read on post-click behaviour in your own analytics.
• Track ads and AI citations as separate programmes. They do not feed each other.

The planning point: a second AI-scale ad network now exists, it is open to anyone who passes a policy check, and its inventory is defined by who has not upgraded. That constraint will loosen as the free base grows, unevenly by market. Until it does, the billion-dollar number is a fact about OpenAI's business, not yours.