Google is expanding its Personal Intelligence feature for Google’s Gemini to users in India. The feature, first introduced in the US earlier this year, allows Gemini to connect with services such as Gmail, Google Photos, YouTube and Search, with the aim to deliver more context-aware and personalised outputs. Users can choose which apps to link, and the system remains switched off by default.

Personal Intelligence is designed to pull together fragmented information spread across different Google services. Rather than relying solely on prompts, the system can reference emails, images, videos and search activity to generate responses.

For instance, a user planning a trip could ask Gemini for their itinerary, and the assistant would compile booking confirmations from Gmail, surface relevant screenshots from Photos, and even suggest places based on previously watched YouTube videos. The idea is to reduce the need to manually sift through multiple apps.

Google says the feature works by combining two capabilities: reasoning across different types of data and retrieving specific details when required. In practice, this means the assistant attempts to understand context while also pinpointing exact information from a user’s digital footprint.

There is enough privacy guardrails as AI systems handle personal data. Privacy is a central element of the feature. Users must actively enable app connections and can disconnect them at any time.

The company also states that data from Gmail or Photos is not directly used to train its AI models. Instead, the system accesses this information only to respond to specific queries. Training is limited to anonymised interactions, such as prompts and responses, after filtering or obscuring personal details.

Gemini also attempts to indicate the sources it uses for its answers, allowing users to verify information. For conversations that require less personalisation, users can opt for temporary chats or regenerate responses without connected data.

There are also safeguards around sensitive topics. The assistant is designed to avoid making assumptions about areas such as health unless explicitly asked.

Despite the emphasis on control, Google acknowledges that the system is still in a beta phase and may produce errors. These can include inaccurate responses or what it describes as “over-personalisation”, where the assistant draws incorrect connections between unrelated pieces of data.

Another challenge is nuance. For example, repeated photos of a user at a particular location may lead the system to infer preferences that are not accurate. Users can correct such assumptions, but the reliance on interpretation highlights the complexity of contextual AI.

The company says these limitations remain an area of active research.

The feature is being rolled out to eligible subscribers of Google’s AI plans, including AI Plus, Pro and Ultra tiers, with availability for free users expected in the coming weeks. It will work across web, Android and iOS platforms.