The assistant grew an ad business — and the EU called it what it is: a search engine

On 31 August, two announcements landed hours apart and said the same thing in different languages. From Brussels: the European Commission designated ChatGPT a Very Large Online Search Engine under the Digital Services Act, after OpenAI reported 159.1 million average monthly users for ChatGPT search in the EU — more than three times the 45-million threshold — starting a four-month clock to full DSA duties in January 2027. From the business press, the same day: OpenAI’s advertising business crossed a $1 billion annualised run rate, roughly 200 days after the first ad appeared in a ChatGPT session in February. Google’s AdWords needed four years to reach that figure; the assistant did it in six months, and the stated target for 2026 is $2.5 billion. Put the two side by side and the week’s real headline appears: the thing sitting in front of a billion questions is now, officially, an ad platform by revenue and a search engine by law. A third data point completes the picture: Apple’s rebuilt Siri answers on a custom 1.2-trillion-parameter Gemini rented from Google for about $1 billion a year. The intelligence in the middle is rented everywhere — and the rent is paid in different currencies. Apple pays cash. Free users pay attention.

The business model arrived before the category did

The mechanics, as OpenAI describes them, are deliberately conservative: sponsored links beneath the answer, matched to the conversation rather than to a behavioural profile, shown to Free and Go users while paid tiers stay ad-free — framed, in the company’s own words, as expanding access to AI. Take all of that at face value, and what remains is still the oldest structure on the internet reassembling itself one layer up. An intermediary that holds everyone’s questions, funded by third parties who pay to be near the answers, will be optimised — gradually, rationally, quarter by quarter — for the funders. That is not a scandal; it is a gradient, and $1 billion in 200 days is a steep one. It is also exactly why the Commission’s designation matters: the DSA now obliges the assistant to do what very large search engines must — assess systemic risks, submit to audits, open data to researchers, explain its recommender logic. But transparency is a window, not a steering wheel. It lets Europe watch the middle. It does not make the middle yours.

What “free assistant” means at work

For a consumer the trade is at least legible: no invoice, ads under the answer, the oldest deal on the web. For an organisation it is a different shape, because at work the questions are the sensitive part. The unreleased product name is in the prompt. The draft contract is pasted in for a summary. The strategy is legible from the query. Piped into an ad-funded intermediary, those questions leave your perimeter, get answered next to an auction, and come back with sponsored suggestions carrying the assistant’s borrowed authority — an authority ten blue links never had, because a recommendation inside a conversation reads like advice, not inventory. Even the measurement is contested: within a day of the run-rate milestone, trade press was reporting advertisers struggling to reconcile the clicks they were billed for with anything their analytics could see. None of this requires bad faith anywhere in the chain. It only requires incentives to keep compounding in the direction they already point. The organisational question is therefore not “are the answers good”. It is: who stands in the middle of our questions, what do they get out of it, and what can we see.

Where the platform stands

Our answer is unfashionably literal: at work, pay for the middle in money — and own as much of it as you can. Qevron is that posture productised. It is one OpenAI-compatible gateway where every model call is yours to route, meter and log: our five in-house model families sit beside 43+ external providers, and every answer is priced per token instead of subsidised per impression. There is no ad layer anywhere in the stack and nothing for a third party to bid on — the only party the answer is optimised for is the one paying the meter, which is you. Because the models underneath are ours, the whole path can run in the cloud, on-premises or fully isolated: the question that contains your roadmap need never leave the building that owns the roadmap. And the observability the DSA will extract from the public middle by January — audits, explained rankings, visible logic — the gateway gives you today, for your own traffic: every token visible, every route explained, every cost attributed to the team that spent it.

What we will not claim

Three honest limits. First, ads funding free access is a real trade, honestly disclosed: hundreds of millions of people get a capable assistant for nothing, OpenAI’s stated rules — contextual matching, labelled placements, free tiers only — are defensible, and if they hold, the free tier remains a fine place to ask for a recipe. Our argument is about gradients and workplaces, not villains. Second, the VLOSE designation is a set of duties, not a condemnation — and it cuts both ways, because 159.1 million people voting with their questions is also the official confirmation that chat won the search gate. Third, we are not selling you a living-room ChatGPT replacement. The claim is narrower: for work, an auditable meter beats a free middle — and if consumer assistants do stay in your building, route them like the third-party processors they are: through policy, not habit.

Every question now passes through a middle. The only middle you can audit is the one you pay in money — because the one you pay in attention is optimised for someone else.

In Türkiye the frame writes itself. Staff pasting work data into a consumer assistant was a KVKK question from the start: the questions travel to infrastructure abroad — under a regime where the Board has issued no adequacy decision for any country, so transfers run in practice on standard contracts notified to it — and an ad business beside the answers means the path your questions take now carries commercial incentives of its own. For institutions under the 2019/12 Presidential Circular the picture is starker still: critical data may not sit in clouds outside institutional control in the first place. Europe’s response is oversight — a designation, audits, a January deadline. Yours can be architecture: a gateway you meter, models you can own, deployment where your data lives. Qevron, from Arpanet Bilişim A.Ş., was engineered for the KVKK from its first line. Contact us and we will scope it with you.