Google has introduced Personal Intelligence, a new feature for the Gemini app that allows users to personalize Gemini by securely connecting select Google apps, including Gmail, Google Photos, YouTube, and Google Search. Rolling out as a beta in the U.S., the feature is designed to make Gemini more proactive and context-aware by allowing it to reason across a user’s own information — with explicit user control and privacy safeguards.

According to Google, Personal Intelligence is optional and turned off by default. Users choose which apps to connect and can disconnect them at any time. Once enabled, Gemini can retrieve and reason across information from connected apps to answer questions or assist with tasks more efficiently.

With Personal Intelligence, Google is allowing users to personalize the Gemini app by securely connecting select Google services, including Gmail, Google Photos, YouTube, and Google Search, through a single opt-in flow. Once enabled, Gemini can reference information from those connected apps to answer specific questions or assist with tasks that require personal context.

Google says the feature has two primary capabilities. First, Gemini can retrieve specific details from connected sources, such as information contained in an email or a photo. Second, it can reason across multiple sources, combining text, images, and video to generate more tailored responses. In many cases, these capabilities work together.

In one example provided by Google, a user needed new tires for a 2019 Honda minivan and realized, while waiting at a tire shop, that they didn’t know the vehicle’s tire size. After asking Gemini, the assistant not only retrieved the correct specifications but also suggested different tire options — including one for daily driving and another for all-weather conditions — drawing on context from family road trips identified in Google Photos. Gemini then surfaced ratings and pricing information for each option.

As the transaction continued, the user also needed the vehicle’s license plate number and trim details. Rather than searching manually or leaving the line, Gemini pulled the license plate number from a photo stored in Google Photos and confirmed the vehicle’s specific trim by referencing information in Gmail, completing the request within a single interaction. Google presents this scenario as an example of how Gemini can combine retrieval and reasoning across personal data sources to assist with time-sensitive, real-world tasks.

Google says Personal Intelligence can also support planning and recommendations by drawing on a user’s past activity and stored content when personalization is enabled. In one example, Gemini was used to help plan an upcoming spring break trip, analyzing prior travel details and family interests found in Gmail and Google Photos.

Rather than offering generic destination suggestions, Gemini avoided common tourist attractions and instead proposed a more tailored itinerary, including an overnight train journey and specific board games suited to the family’s preferences. Google presents this as an example of how Gemini can move beyond broad recommendations to deliver suggestions shaped by a user’s personal history — based only on the apps they choose to connect and when personalization is enabled.

Google emphasizes that Personal Intelligence is off by default and requires explicit user consent. Users choose which apps to connect, can disconnect them at any time, and can disable personalization for individual conversations or use temporary chats that do not reference connected data.

When enabled, Google says Gemini can access information from connected apps when a user asks it to help with a specific task. For example, Gemini may look up a detail from an email, reference a photo, or pull information stored in another Google app to answer a question, rather than requiring users to search across those apps themselves.

Google argues that this approach keeps personalization within its own systems, since the data already lives inside Google services like Gmail and Google Photos. As a result, users are not required to upload or transfer sensitive personal information to external AI platforms in order to receive personalized responses — a distinction Google describes as a key difference between Gemini and AI tools that rely on third-party data transfers.

To help users understand how responses are generated, Gemini will attempt to explain or reference the sources used from connected apps so you can verify it. Google says users can also regenerate responses without personalization or switch to temporary chats that do not reference connected data.

According to Google, Gemini does not directly train its models on personal content from Gmail or Google Photos. Instead, personal data is referenced to generate responses to specific requests. Model training relies on limited information, such as user prompts and Gemini’s responses, after steps are taken to filter or obfuscate personal data.

Google says it has implemented guardrails to avoid making proactive assumptions about sensitive topics, such as health information, unless a user explicitly asks. In addition, Google notes that users can review its privacy documentation and adjust settings, disconnect connected apps, or delete chat history at any time.

Google acknowledges that Personal Intelligence remains a beta feature and may produce inaccurate responses or what it calls “over-personalization,” where Gemini makes incorrect connections between unrelated information. To address this, Google encourages users to actively provide feedback during use.

Users can give a response a thumbs-down rating when something feels incorrect, misleading, or overly personalized. Gemini also allows users to correct assumptions in real time — for example, by clarifying preferences such as “Remember, I prefer window seats”.

Google notes that nuance can be difficult for AI systems, particularly when interpreting changing relationships or evolving interests. For example, a large number of photos taken at a golf course might lead Gemini to assume a user enjoys golf. In this example, that assumption is incorrect: the user does not enjoy golf but regularly attends because their son enjoys playing. Google says users can correct these kinds of mistaken inferences directly — for instance, by stating “I don’t like golf” — to help refine future interactions.

The beta rollout of Personal Intelligence is beginning in the U.S. for eligible Google AI Pro and AI Ultra subscribers. Once enabled, the feature works across web, Android, and iOS, and is compatible with all models available in the Gemini model picker.

At launch, Personal Intelligence is limited to personal Google accounts and is not available for Workspace, enterprise, or education users. Google says access will expand over time, including to additional countries and eventually the free tier. The feature is also expected to integrate with AI Mode in Search in the future.

Users may see an invitation on the Gemini home screen as access becomes available. If not, the feature can be enabled manually through Settings.

To turn it on, users can:

Google says users can disconnect apps, adjust settings, or delete chat history at any time, reinforcing that personalization remains fully optional and reversible.

Q: What is Gemini Personal Intelligence?
A: Personal Intelligence is a Gemini feature that allows users to personalize the assistant by securely connecting select Google apps. When enabled, Gemini can reference information from those connected apps to deliver more context-aware responses to specific requests.

Q: How does Google handle privacy and user control?
A: Google says Personal Intelligence is off by default and requires explicit user opt-in. Users choose which apps to connect, can disconnect them at any time, disable personalization for individual chats, or use temporary chats that do not reference connected data.

Q: Does Gemini train on personal emails or photos?
A: According to Google, Gemini does not directly train on content from Gmail or Google Photos. Personal data is referenced to generate responses, while model training relies on limited information such as user prompts and model outputs, after steps are taken to filter or obfuscate personal data.

Q: What are the current limitations?
A: Google acknowledges that the beta may produce inaccurate responses or “over-personalization,” where the system draws incorrect connections between unrelated information. Users are encouraged to provide feedback and correct Gemini in real time to help improve accuracy.

Personal Intelligence moves Gemini closer to being an assistant that can help with real-world friction, not just answer general questions. By allowing Gemini to reference a user’s own emails, photos, and search history — with explicit permission — Google is aiming to reduce the small but constant effort involved in everyday tasks like planning, remembering details, or locating information spread across apps.

For users, the value isn’t just personalization, but time saved and fewer interruptions. Instead of manually searching through inboxes, photo libraries, or past searches, Gemini can surface relevant details in the moment they’re needed. That convenience becomes more meaningful as digital lives grow more complex and fragmented across services.

At the same time, Google’s emphasis on opt-in controls, transparency, and user feedback highlights the tradeoffs that come with deeply personalized AI systems. The feature’s limitations — including inaccurate assumptions or over-personalization — underscore why human oversight still matters, especially when AI systems draw conclusions from personal data.

If Personal Intelligence performs as intended, it could make AI assistants feel more useful in daily life without requiring users to give up control over their data. Whether that balance holds will depend on accuracy, trust, and how well users can understand — and correct — what the system is doing.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.