Google's Ad Policy Adds an AI Data-Sharing Exception

Google changed its privacy-policy language on October 16 to add an exception involving Gmail, Drive and Photos content. The policy still says Google does not show personalized ads based on content from those services, but now adds: “unless you choose to make data from these services available to other Google services that support personalized ads.” The new condition puts an important question inside a familiar convenience: what else happens when you connect a personal app to an AI feature?

The exception concerns a choice to share data across services. It does not mean every Gmail user has started getting ads selected from their inbox. Google says connections are opt-in and advertisers do not receive the underlying private content. Yet the wording leaves room for information from a connected app to influence ad personalization through another Google service. For someone deciding whether to connect Photos or Gmail to an assistant, that possible second use belongs in the decision alongside the promised better answers.

How an AI Convenience Reached the Ad Policy

In January, Google introduced Personal Intelligence as a way for Gemini to retrieve details from connected apps and reason across them. Google described linking Gmail, Photos, YouTube and Search, with connections off by default and users choosing which apps to turn on. Its examples included finding information in an email or using past trips and family interests to suggest travel plans. The feature launched as a US beta for eligible Google AI Pro and AI Ultra subscribers; that launch description does not establish who can use every connected-app feature today.

That original pitch explains why someone might grant access. Finding a booking in an inbox beats searching through confirmation emails while standing at a hotel desk. The same design also creates a boundary to examine: information kept in one app becomes useful to another service. By October, the reported ad-policy wording explicitly addressed what can happen when the receiving service supports personalized ads. A decision that feels like permission to answer a question can also be a decision about the information used around that answer.

The older ad-policy wording was simpler: “We don’t show you personalized ads based on your content from Drive, Gmail, or Photos.” The October revision kept that protection with the new conditional exception, as the before-and-after wording shows. The change is in the stated boundary for connected data. It is not a statement that Gmail, Drive and Photos have ceased to have any protection.

Personal Intelligence also reaches AI Mode in Search, where sponsored content may appear. Gemini’s app did not have ads at the time of the October account. That gives the exception a more concrete setting than the image of an advertiser browsing somebody’s mailbox: an AI search experience can use connected information to shape a response while also showing ads. Whether a given piece of connected information changes a given ad, and through what exact route, remains unclear. The policy describes a permitted use, not a count of affected ads.

Connecting apps is entirely opt-in, and ads remain relevant to factors such as the query, response context and a person’s interests. “Ads will not be based directly on your Gmail inbox or Photos library, and as always, sensitive information will not be used to tailor ads,” the company said in a statement obtained by Ars. A Google spokesperson also told Ars that the clarification did not signal a change in how Google handles data with advertising partners or mean it gives that data to third parties.

That is a consequential limit on the threat. An advertiser receiving an email or photo would face a different privacy question from Google using information inside its own services to decide what appears beside an AI answer. Google rules out the former in its explanation; the revised policy permits a conditional version of the latter. Its statement about ads not being based directly on an inbox or library does not spell out precisely how connected information might contribute indirectly. Readers should be able to evaluate that trade-off without being told either that their mail is for sale or that an opt-in connection affects only the AI reply they requested.

Three Uses of Data, Three Questions to Ask

Google’s January announcement made another privacy promise: Gemini does not train directly on a Gmail inbox or Google Photos library. It said connected material could be referenced to produce a reply, while limited information such as Gemini prompts and responses could be used to improve the model after steps to filter or obscure personal data. Those are Google’s stated practices, rather than a guarantee that every detail in a conversation stays out of training.

Put the January announcement beside the October policy change and three uses emerge. First, an assistant can retrieve connected information to answer a request. Second, Google describes limited training on prompts and responses rather than direct training on the inbox or photo library. Third, the ad-policy exception concerns data made available to another service that supports personalized ads. A promise about the second use does not settle the third. This is why asking only, “Does Gemini train on my emails?” misses a question for people who also use personalized Search.

The comparison also has a limit. Google’s training description concerns Gemini, while the ad exception is phrased around other Google services that support personalized ads. It would be a leap to treat them as two measurements of the same data flow. They are separate rules a person has to consider when deciding whether the convenience of a connected app is worth granting access.

What You Can Change

If you would rather keep those apps disconnected, Google says you can turn off Personal Intelligence connections in Gemini settings. For Search, the account-settings route runs through Data & privacy, Personalization settings, Manage Personalized Recommendations in Search Services, then Connected Content apps, where you can disconnect Google Workspace or Photos. Search Services Personalization also offers a switch for Personalized Recommendations. Settings labels can change, so check which connections your account actually shows before assuming that one switch governs every Google product.

There is a cost to disconnecting: a feature you wanted for finding bookings or drawing on photos may lose that context. Disabling recommendations is another choice with its own effect on personalization; it should not be mistaken for a documented, standalone switch that preserves every Personal Intelligence connection while excluding its data from ad use. If retaining those connections is important to you, the precise scope of that trade-off is the question to put to Google.

Google should make that scope plain at the moment someone connects an app: which receiving services can use the information for ads, what information can carry over and which control stops that use. Users can make a decision about a hotel booking, an AI answer and an ad. They should not have to infer all three from the word “connect.”