Useful AI Assistants Need Personal Context

Smartphones already hold much of our daily information. They contain messages, photos, contacts, locations, schedules, and apps. An AI assistant can become useful when it finds the right personal information without making the user copy it into a chat. Apple’s new Siri AI began rolling out in beta in English on September 14, 2026. Its results show that access to context and AI capability must work together.

The author tested old Siri, Siri AI, and ChatGPT after a weekend in New York City. Siri AI found three breakfast places recommended by the author’s brother Emmett. It also understood a follow-up question about whether the author visited them. ChatGPT could search connected work accounts, but those sources did not contain the needed messages. More connections only help when they contain information related to the task.

The assistant must explain what it could and could not search. Siri AI missed a plan sent through Signal because it did not have access by default. It found a related message from another person but did not clearly explain the gap. A better answer would name the sources it searched and ask the user to provide information from another app. Otherwise, “I could not find it” may sound like “it does not exist.”

Understanding intent is also important. Siri AI found the author’s trip photos when the request did not say “my photos.” But it confused Dimes the restaurant with Dimes Square. It should have asked a short question before opening directions. It later forgot the confirmed starting address and opened Google Maps from the current location. Details accepted in one step must remain correct in the final app.

Designers should test the complete task, not only whether an assistant opens the right app. Users need to inspect and correct important actions before they happen. Examples can teach new abilities, but they must match the data the assistant can reach. Progress labels should explain useful facts, not only show activity. A useful assistant retrieves relevant context, understands the request, shows its limits, and preserves intent through the final result.