The Eager Seller

The approach came on Friday afternoon. Three days before launch.

Not through our agencies, though they and others have been appointed to sell this thing across Europe. Directly. As one of Europe's larger advertisers, we received an approach on the platform's behalf, ahead of Monday's rollout, asking whether we wished to participate in ChatGPT's European advertising launch.

I'll tell you what the approach made me think. I'm deliberately not going to tell you what we decided. By the end, you'll understand why.

From 24 August, ChatGPT Ads went live in 31 European countries, taking the platform to 40 markets worldwide, six months after the US pilot began. Europe's version is not America's version.

Ads appear only on ChatGPT's Free and Go tiers; Plus, Pro and Enterprise stay clean. Personalised advertising is switched off across the EEA and Switzerland at launch. OpenAI's European legal basis is consent under GDPR, and rather than build the consent machinery first, they launched without personalisation. Under-18 accounts are excluded. The Digital Services Act requires every ad to be labelled, with disclosure of who placed it and why you're seeing it. And there is no self-serve access: buying runs through OpenAI's own ads team and its appointed intermediaries (Publicis, Omnicom, WPP, Havas, Dentsu and MediaPlus) with self-serve Ads Manager promised "later this summer," date unspecified.

This is arguably the most compliance-forward major ad launch the industry has seen. Consent named as the legal basis up front. Personalisation off, not fudged. Paid tiers ad-free, so the ad-funded relationship is opt-out-able at €23 a month. Minors excluded. Ads visually separated from answers, with OpenAI stating plainly that advertisers get no access to conversations. Compare that with how search and social advertising actually arrived in Europe, being retrofitted onto products already at scale then litigated into compliance over a decade and this is a different posture. The revenue, OpenAI says, funds free access.

Now the cloud. The EU's AI Act bans systems that deploy manipulative techniques and regulators haven't yet decided what that means when the advertising surface is a conversation that feels private. Separately, the European Commission is weighing whether ChatGPT's search function crosses the 45-million-user threshold for designation as a very large online search engine under the DSA, which would bring annual risk assessments, independent audits and a public ad repository. In other words: the most compliance-forward launch in memory may be arriving just before the compliance bar itself moves. Both readings will run all autumn.

Amongst the problems I turned over in considering this is that there are no benchmarks. Not "few." None that count. OpenAI's own official position is that no cross-advertiser performance benchmarks exist, and that its $3–5 starting bid guidance is guidance, not an average. Everything else in circulation claiming ROI evidence and a model for how to make the most of the platform comes from people selling something.

With that health warning, the American evidence so far: the early pilot recorded a 0.91% click-through rate against roughly 6.4% for Google search ads. A real gap, though CTR is arguably the wrong lens for a format where the user carries on talking. Criteo data shows LLM-platform referrals converting at 1.5x other digital channels, with visitors spending 60–80% more time on site. So when you are ready to click, you are acting with more intent than just exploration. The most granular public account (one US agency's fifteen days of live spend) ran at roughly $1.72 per click, under half the guidance range, at 1.49x blended return with daily swings from 0.2x to 2.9x. Their own conclusion: thin competition is flattering both numbers, and costs will travel toward the guidance range as the auction fills.

So the early-mover case is genuine (soft auction, cheap clicks, unusually high-intent traffic) and it is time-boxed, because pre-compression pricing is a temporary market inefficiency, not a feature. The honest way to say it: the first buyers into this channel are not exploiting price discovery. They are funding it. Every campaign booked this quarter is a data point the platform needs and the buyer pays for.

And look closely at what actually launched, because you will see the stitching. A piece of search (the intent, the moment of the query) but with the personalisation removed. A piece of display (the adjacent placement, the premium pricing) but assembled in months against a deadline. The sceptic's reading is that Europe has been sent a Frankenstein product: parts harvested from every existing format, sewn together at speed, and quite possibly the worst of each.

On one dimension the European product may be better than its restrictions suggest. On display or social, stripping personalisation guts the targeting. Here, the user has just told the system exactly what they want, in their own words, seconds before the ad appears. The conversation is the profile. It may just be that Europe's "hobbled" contextual-only product performs far closer to the American one than either regulators or sceptics expect, which softens the performance objection and sharpens the regulatory one in the same breath.

So the assembled thing probably lives. The intent is certainly real. And I still think the European launch may not be for you. Here's why.

Ask why a platform launches its most complex product expansion (in 31 privacy-regulated markets at once) before its self-serve controls exist, before its consent machinery is built and before a single benchmark is published.

The answer has a date on it. OpenAI confidentially filed for a public listing in late spring (with Goldman Sachs and Morgan Stanley reportedly leading), targeting a debut as soon as the fourth quarter of this year, at a valuation reported as high as $1 trillion, though the timing may yet slip into 2027. The company reports $2 billion in monthly revenue, growing faster at this stage than Alphabet or Meta did, and is not yet profitable.

Into that context, read the launch week again: the largest geographic expansion in the product's history, an unaudited "+25% ad revenue this month" figure handed to the press alongside it, and a sprint from pilot to 40-market network in six months. That is the cadence of building a revenue line to a filing deadline. "40 markets" is a prospectus sentence.

And the eagerness has deeper roots. OpenAI is what it is because of a constitutional reflex for being a first-mover. The company's every defining moment has been a launch nobody else had yet dared. Falling forward has been the strategy, and let's be fair to it: falling forward built the most successful consumer software launch in history, and everyone who has bet against OpenAI's velocity has so far lost. But the firsts bought audience, not economics. While OpenAI was winning the consumer war, the enterprise market flipped to its rival: Menlo Ventures' enterprise survey now puts Anthropic at 40% of enterprise LLM API spend against OpenAI's 27%, a near-perfect inversion of 2023. OpenAI's rebuttal is that enterprise is now over 40% of its revenue and heading for parity, but a billion free users generate the world's largest inference bill either way, and ads to those users is where the spiral was always going to land. The deadline, the reflex, and the bill: three reasons the seller is eager, and none of them is your media plan.

Here is what makes this first different from all the previous ones. Being first with GPT risked OpenAI's money. Being first in its ad auction risks yours. Every prior leap, the falling company absorbed the fall. This time the landing mat is the advertiser's budget.

One more reading of the week, from experience rather than suspicion. The direct approach itself didn't surprise me since when new media products launch, going straight to the biggest advertisers is normal practice, and anyone who has spent a June on the Croisette knows the media-owner conversation is always direct. What deserves your attention is the structure around it. OpenAI needs those six appointed agencies today, because agencies are how you unlock clients' money at scale in thirty-one markets at once. And (like Google before it, like Meta before it) it has already published its intention to need them less: self-serve is on the launch roadmap, in writing, from day one. The doormen are being paid in the currency of a relationship the platform intends, in time, to own.

You now have everything I had this weekend. The asymmetry: they need this quarter; you don't. The evidence: promising, vendor-published, unaudited, time-boxed. The product: better than its restrictions suggest, which comforts the buyer and should worry the regulator. The seller: eager for three reasons, none of them yours.

We made our decision at Nomad. I'm not going to tell you what it was. Not because it's confidential strategy, though it is, but because the disclosure would defeat the purpose of the piece. The entire point of a channel with no benchmarks is that the only defensible decision is the one you build from your own mathematics: your categories, your margins, your appetite for funding someone else's price discovery. If you find yourself deciding because of momentum, you have already answered the wrong question.

When the seller is more eager than the buyer, the buyer sets the terms. That is the whole negotiation, and this week, Europe's marketers are the buyer.

So here is my question, and I mean it practically, not rhetorically: you've now seen the same maths I saw this weekend. Which way did you decide, or which way would you? And what number, if OpenAI published it, would change your answer?

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 27 August 2026 in the Brandflow newsletter on LinkedIn.