Google has begun pushing Personal Intelligence into the car, a layer that lets Gemini answer spoken questions by drawing on data from Gmail, Google Calendar, Google Photos, Maps, YouTube and Search. The company announced the move on 20 October and described the rollout as global — a claim that holds only until you reach the footnote.
That footnote carves out the European Economic Area, Switzerland, the United Kingdom and Nigeria. Drivers in Spain, and across the entire EU, are excluded from the launch.
- What launched: Gemini in Android Auto can now retrieve details and cross-reference several Google services to answer questions hands-free.
- Who gets it: Personal Google accounts only, users aged 18 and over, in markets outside the EEA, Switzerland, the UK and Nigeria.
- How it works: Strictly opt-in from the Gemini settings, with per-app permissions that can be revoked at any time.
- The catch: Connected-app activity, including summaries and inferences, can feed Google’s model training, and part of it is reviewed by humans.
What actually changes behind the wheel
Gemini first reached Android Auto in November 2025 as the successor to Google Assistant, so the assistant itself is not new. What is new is memory: the ability to answer with the driver’s own data inside the vehicle.
Google frames this as two distinct capabilities. The first is retrieving a single concrete detail — an email, a photo, a place saved in Maps. The second is reasoning across several sources at once. The company’s own examples are more illustrative than any specification: asking for a boarding gate and whether a flight is on time, asking to be driven to a doctor’s appointment, asking for the car wash from a photo taken the previous week, or asking for directions to the café mentioned the day before.
The rollout began on 20 October on Android Auto. Vehicles with Google built in will follow “in the coming months”, with no firm date. Those cars carry a telling technical detail: they only share personalised answers while the phone is connected, which suggests the personal context travels through the handset rather than living in the vehicle. Google does not document what is processed on-device and what in the cloud — an information gap rather than a confirmed fact.
With the feature switched off, Gemini still handles navigation, music, calls and general queries, but nothing is tailored to the driver’s account.
Activation, requirements and the fine print
Personal Intelligence is opt-in and no app is connected until the user decides. It is enabled in the Gemini settings, where each service is chosen individually, and it can be turned off at any point. Users who already enabled it on their phone inherit it in the car automatically.
Google’s documented requirements are an Android phone running Android 11 or later with at least 2 GB of RAM, Gemini configured as the digital assistant, and a compatible vehicle. It is invoked with “Hey Google”, the on-screen microphone or a long press of the steering wheel’s voice button. Gemini Live starts with “Let’s talk Live” and stops with “Stop”, and can only be started or stopped by voice.
The exclusions matter as much as the inclusions: personal Google accounts only, so Workspace business and education accounts are out; users must be over 18; and the product remains in beta. On pricing, the picture is muddled. Personal Intelligence debuted on 20 January 2026 for Google AI Pro and Ultra subscribers in the United States, yet Google’s current documentation no longer lists a subscription requirement, and reporting since March points to a wider release to free accounts and additional markets without an official confirmation. One documented exception survives: Contacts access remains United States-only, English-only and tied to AI Ultra.
Privacy: training data and human review
This is the part the launch messaging barely touches. According to the feature’s own privacy page, Gemini activity — including data from connected apps — may be used to improve Google’s services, model training included, when activity saving is enabled.
Google is careful to say it does not train directly on Gmail, Drive, Contacts, Calendar, Wallet or the images and audio in Google Photos. It can, however, use summaries, excerpts, generated media and inferences derived from that material. The gap is narrower than it sounds: Google concedes that in some cases “the summary may simply be the email or the file”. A subset of processed connected-app data is also reviewed by people, including trained service providers; Photos files are not viewed except for abuse, harm or submitted feedback. The overall scope reaches beyond the apps named in the car announcement, extending to Workspace, search services, Contacts, Wallet, Photos, YouTube and location inferred from IP address, device and content.
One detail is rarely checked: disconnecting an app does not erase Gemini activity already recorded, and changes to an app’s data can take days to surface.
Google’s own warnings
The company’s help documentation is unusually blunt about the limits of the system.
“It is new and evolving, and it may hallucinate and present inaccurate information as fact,” Google states, adding that it “should not be relied upon for critical or safety-related information while driving.”
A second warning is more awkward still: anyone in the vehicle, children included, can use Gemini — which means younger passengers “may hear content that may be inappropriate for their age”. An assistant that reads email aloud with an audience on board is a scenario nobody has fully designed for.
The word “globally” is the weakest link in the announcement. Excluding the entire European Union is not a global launch with minor exceptions, and while Google offers no explanation, the pattern repeats with every feature that crosses data between services. For drivers in Madrid or Barcelona, the news is simply that this exists and is not for them yet. The product direction is sound — routing to a medical appointment without reciting an address solves a genuine problem without taking eyes off the road. But the dashboard assistant now holds the inbox, the calendar and the photo library, with human review on part of it and inferences feeding training. The car has become a data-collection surface, and whoever controls the assistant controls the dashboard.