Google's Magic Pointer identifies interactive items on a Googlebook locally, but sends selected content to a cloud model when a user asks Gemini to act on it. The distinction defines when information visible on the screen leaves the laptop.
Android Authority reports that Alexander Kuscher, Google's senior director of product for laptops and tablets, explained the process on the Android Faithful podcast. Activating Magic Pointer causes the computer to recognize items such as an email, image or block of text and highlight them without first uploading the screen.
The cloud step begins with an explicit action
After the user chooses an item and asks Gemini to do something, the selected content is sent to Google's cloud model. Kuscher used checking an email for spam as an example: recognizing that the item is an email happens on the device, while analyzing its contents requires the online model.
Google says the AI cursor remains inactive until the user deliberately triggers it. That means Magic Pointer is not supposed to continuously transmit everything on screen merely because the feature exists.