OpenAI emailed every free ChatGPT user in Europe: ads are coming, and they will use only three data points. Their own privacy policy lists seven.
The email undersells the data collection by half. The policy discloses purchase data from advertisers, prior ad interactions, behavioral signals, and contact sync on top of the three the email names.
"Ads are coming to ChatGPT Free/Go in the EEA later this month; selection uses only the current chat topic, general location and device type; Plus/Pro stay ad-free." [SOURCE ↗]
THE CLAIM. OpenAI Ireland emailed EEA Free and Go users on August 15 that ads will arrive later this month, selecting on "only" the current chat topic, general location, and device type. Plus and Pro stay ad-free.
THE CHECK. the updated privacy policy lists seven distinct ad-related data flows: conversation context, general location, device type, purchase data received from advertisers, prior ad interactions, behavioral signals, and optional address book sync. Four of those seven do not appear in the email. The word "only" is doing load-bearing work it cannot support.
THE TWIST. OpenAI called this a "test" in the US in February. By August it has rolled through the UK, Mexico, Brazil, Japan, South Korea, and now the entire EEA. A test that expands to 30+ countries is a launch.
The full receipt trail
On August 15, OpenAI Ireland Ltd sent an email to every Free and Go user in the European Economic Area and Switzerland. The key sentence: ad selection "starts with the current chat topic, general location and device type only."
That word "only" is the entire story.
What the privacy policy actually says
The updated OpenAI privacy policy (effective February 9, 2026, last revised April 30) describes seven distinct data categories involved in ad delivery:
1. Current conversation topic (in the email) 2. General location (in the email) 3. Device type (in the email) 4. Purchase data received from advertisers (NOT in the email) 5. Prior ad interactions (NOT in the email) 6. Behavioral signals (NOT in the email) 7. Optional device address book data (NOT in the email)
Four of seven data flows are absent from the user-facing email. The email is not lying, but it is presenting fewer than half the relevant facts.
The "test" that became a launch
OpenAI's language has consistently called this a "test." The timeline tells a different story:
- February 2026: US launch
- August 11: UK, Mexico, Brazil, Japan, South Korea
- August 15: Entire EEA + Switzerland
A test that expands to 30+ countries across four continents in six months is a phased global launch with plausible deniability built into the language.
The steelmanned counter
OpenAI could argue the email describes the ad *selection* process (what determines which ad you see), while the policy covers the broader *measurement and infrastructure* stack (what happens after). This is a reasonable distinction if the four omitted categories genuinely do not influence which ad appears. But "behavioral signals" and "prior ad interactions" are selection inputs at every other ad platform. OpenAI has not explained why they would collect these without using them for targeting.
What to do
Read the privacy policy, not the email. If you are a paying Plus or Pro user, you are not affected yet. If you are on Free or Go in Europe, your conversations are now ad-adjacent data. The opt-in for full personalisation is coming; the infrastructure is already in place.
Our call
Within 12 months the opt-in for personalised ads becomes the default, and the three-point framing quietly expands. Every ad-funded platform has followed this exact trajectory: launch with minimal targeting, expand behind policy updates, and let the original "only" language fade from institutional memory.
Flips if OpenAI freezes ad data categories at the current policy level or tightens the policy to match the email.
If you use ChatGPT Free in Europe, you are about to see ads. The email frames the data collection as minimal. The policy says otherwise. Read the policy, not the email.
Within 12 months the opt-in for personalised ads becomes the default, and the "only 3" framing quietly expands. The staged global rollout pattern matches every ad-funded platform before it.
Flips to holds if OpenAI tightens the privacy policy to match the email's three-input targeting description. Flips to failed if OpenAI confirms that purchase data, prior ad interactions, behavioral signals, or contact sync are ad-selection inputs.
RECEIPTS (6) · CONFIDENCE MEDIUM · every URL below answered a live HTTP check before publish · sweep 2026-08-25
- ▲ ppc.land ⧉ · "current chat topic, general location and device type only"
- ▲ ghacks.net ⧉ · "They can dismiss ads, give feedback, see why an ad was shown, delete ad data with one tap, and manage ad personalization"
- ▼ ppc.land ⧉ · "Rather than only data going out to advertisers in aggregate form, the updated policy confirms that advertiser-sourced data - including purchase data - now flows in to OpenAI."
- ● ghacks.net ⧉ · "expanding its ChatGPT ads test to the United Kingdom, Mexico, Brazil, Japan, and South Korea"
- ● almcorp.com ⧉ · "advertisers do not receive access to user conversations, chat history, personal details, or memories"
- ● leanware.co ⧉ · "OpenAI says targeting will be contextual rather than based on selling your data"