A user opens Meta AI, asks about a symptom, a mortgage or a holiday, and sees no advertising in the window. According to Japan’s communications ministry, the conversation still reaches the advertising system. The Ministry of Internal Affairs and Communications said Meta uses exchanges between users aged 18 and older and its Meta AI assistant to optimize advertising, and that the company has confirmed it does so.
The finding appears in the 2026 monitoring report of the ministry’s User Information Working Group, which reviews how large platforms handle personal data. The group settled its conclusions after deliberations held Oct. 19 to 21 and published them this week. This year the review added a question it had not asked before: how conversational AI is used in ad delivery.
Meta was not the only company named. LINE Yahoo serves advertising that matches the content of a user’s questions inside its conversational AI, and users have no way to opt out of that prompt-based targeting. As of August the company was considering using earlier conversations for advertising as well. LINE Yahoo has been building the model commercially since Sept. 1, when it expanded context-based ad delivery on its Agent i assistant and began charging advertisers for it.
Two other companies came out of the review differently. Google does not use Gemini conversations for targeted advertising, the report said, though its conversational products displayed search-related ads as of September. TikTok reported no conversational AI advertising and no use of AI conversations to personalize ads.
The report also compared what each company feeds into model training, and the differences are worth reading closely. Google listed Gemini prompts, uploaded files, feedback and information from linked apps. Meta listed posts, photographs, captions, Meta AI conversations and third-party information. LINE Yahoo reported no use of user information in the conversational AI training covered by the review, while TikTok said it uses user-generated content for moderation and advertising optimization.
The ministry’s conclusion was procedural rather than punitive. The working group recommended that the ministry verify early how conversational AI handles user information, both for training and for advertising. That is the limit of its authority: it can examine, publish and recommend, and it cannot fine. Nothing in the report is an enforcement action.
What makes the finding worth attention is the asymmetry it describes. A search engine’s advertising is visible beside the query, and users have learned to connect the two. A chat window shows no ads at all, so a user has no reason to think the message is being read for commercial purposes. The inference that the conversation is a targeting signal is correct and entirely invisible from the interface.
The commercial logic explains why companies do it. A search query is a few words and a click. A conversation carries the constraints a person volunteers: budget, dates, who else is involved, what they have already rejected. That is better targeting data than anything a search box produces, and the assistant is the only place it exists.
Japanese law gives the ministry less room than the finding implies. The Act on the Protection of Personal Information governs how personal data is used and shared, with consent requirements and an opt-out mechanism for third-party provision, and the working group’s questions track those obligations. Whether a chat transcript is personal data used for advertising, and whether existing disclosures cover that use, is a question the report raises rather than settles.
The international context is moving in the same direction without converging. European rules on targeted advertising and platform data use, Korean amendments to privacy law and various national reviews have all begun asking what happens to a conversation once it is stored. The Japanese report is unusual in publishing a company-by-company comparison, which turns a general concern into a named list.
For the companies, the exposure is reputational first and legal second. A platform that discloses in a policy document that it reads chats for advertising has a defensible position in most jurisdictions. A platform whose users believe otherwise has a problem the moment a regulator writes it down, which is roughly what happened here.
What the ministry holds over the companies is the next monitoring round. Companies that adjust disclosures, add opt-outs or separate advertising from assistant data will have something to point to. Companies that do not will be named again in a document other regulators read, and a second mention is harder to describe as an oversight.
There is a version of this argument that ends in consent screens, and a version that ends in silence. Nothing in Japanese law currently requires a chat assistant to tell a user, at the moment of typing, that the message may be used to choose an advertisement the user will see somewhere else. The working group has now put that arrangement on the record without saying what should replace it.
The question left open is whether a chat window needs to say what it is. Meta does not show ads inside Meta AI, which is true and also the reason the arrangement is easy to miss. LINE Yahoo shows them, which is more honest and less comfortable. The ministry has established that both happen, and has not said which one it thinks is acceptable.
