The Last Resort Arrives: ChatGPT Gets Ads, and the Oldest Trap Repeats
In a nutshell
everything on the web starts with the domain
Two years ago, the man now selling the ads called the idea of advertising inside an AI assistant uniquely unsettling, a last resort. This week, that last resort quietly went live across five more countries. The most intimate technology most of us have ever used — the one we ask our real questions — is being turned into an advertising surface, and we have seen exactly this film before.
What OpenAI Confirmed
On 11 August 2026, OpenAI confirmed that its ChatGPT advertising test has expanded to the United Kingdom, Mexico, Brazil, Japan, and South Korea. The ads appear for logged-in adult users on the free and low-cost Go tiers; paying Plus, Pro, Business, Enterprise, and Education subscribers see none. The rollout has been methodical: a US launch in February, then Canada, Australia, and New Zealand, and now this latest wave, with more countries promised before year end.
OpenAI has wrapped the move in careful reassurances, and they deserve to be stated fairly. The company says ads do not influence ChatGPT's answers, that they are clearly labelled as sponsored and visually separated from the response, and that conversations stay private from advertisers, who receive only aggregate performance data rather than chat logs. Users can dismiss ads, see why one appeared, delete their ad data with a single tap, and review an ads history. Sensitive topics — health, politics, mental health — are excluded, as are minors. On paper, this is about as restrained as ad-supported anything has ever launched.
Altman's Reversal
The reason this matters is not the mechanics but the trajectory, and no one has mapped that trajectory more clearly than OpenAI's own chief executive. In 2024, Sam Altman described the combination of advertising and AI as deeply unsettling, something the company would turn to only as a last resort. The reversal is now complete, and the cause is no secret. OpenAI is spending on a scale that has no precedent for a company its age — over a trillion dollars in infrastructure commitments, vast ongoing losses, an IPO on the horizon — and advertising is the same cash machine that funded Google and Meta before it. With hundreds of millions of weekly users on free tiers, the temptation was always going to win. The "last resort" language was never a principle. It was a countdown.
The Trust Problem
Here is the structural tension no reassurance fully resolves. The value of an AI assistant rests entirely on trust — you tell it things you would tell no search box, and you act on what it tells you back. Advertising introduces a second master. OpenAI insists ads never shape the answers, and today that may well be true. But the whole history of the ad-supported internet is the story of that wall eroding under commercial gravity: once revenue depends on what users click, the pressure to optimise the experience toward what pays never stops pushing. Even if the answer text stays pure, the surrounding surface — what gets suggested, surfaced, nudged — is now partly shaped by who is paying. The concern is not that OpenAI is lying today. It is that the incentive has been permanently installed, and incentives, not promises, decide long-run behaviour.
Claim and Counter-Claim
The case for OpenAI's approach is genuine. Someone has to pay for giving hundreds of millions of people access to frontier AI, and ads on free tiers are a far more equitable funding model than locking capable AI behind a paywall only the wealthy can afford. The safeguards are real and unusually thorough, the labelling is clear, and the sensitive-topic exclusions are thoughtful. If advertising keeps powerful AI free for those who cannot pay eight dollars a month, that is a defensible trade.
The counter-case is that we are watching the original sin of the consumer internet being committed a second time, with full knowledge of how it ended the first time. We were promised clearly labelled, privacy-respecting, non-intrusive ads by the last generation of platforms too, and that promise decayed into surveillance advertising, engagement optimisation, and the attention economy that has done measurable harm. There is, as one privacy-focused critic put it plainly, no guarantee the wall holds. And the deeper objection is that an AI assistant is a far more intimate surface than a search results page — it knows your fears, your health worries, your half-formed plans — which makes it the most powerful targeting engine ever built, whatever restraint is shown at launch. The honest synthesis: the safeguards are real, and so is the slope beneath them, and history says the slope usually wins.
The European Perspective
Look closely at the map, because it tells the European story without a word of comment. The ads test just reached the UK, Mexico, Brazil, Japan, and South Korea. It has not reached the European Union, and it has not reached Switzerland. That is almost certainly not an accident. Advertising built on a system that knows this much about a user runs straight into the GDPR's rules on profiling and consent, into the AI Act's transparency obligations, and into the Digital Markets Act's scrutiny of how dominant platforms monetise attention. Europe is, once again, the market that comes last — not because it is behind, but because its rules make the easy version of this harder to ship.
This is the quiet vindication we keep returning to: the same regulatory friction that critics call a brake is also the thing standing between European users and the fastest, least examined version of ad-supported AI. But vindication is not victory, and Europe should be honest about the bind. The pressure to monetise will not stop at the EU's border; it will arrive, reshaped to fit the rules, and the question is whether those rules produce genuinely better advertising or merely slower advertising. The task is not to keep ads out forever — that is neither realistic nor entirely desirable if the alternative is AI only for the rich. It is to insist that when the ads do arrive, the wall between what the assistant knows and what the advertiser can exploit is built in law, not left to a company's promise and a chief executive's changing mind.
We are not first. We are right.