Google Gemini struggles to identify individual pets

Google has introduced Pet Memory, a generative AI feature for Gemini for Home designed to help Nest cameras recognize individual pets by name. Available through the $20-a-month Google Home Advanced Plan, the feature attempts to move beyond basic animal detection. Instead of building a visual profile like Google's Familiar Faces facial recognition tool, Pet Memory works by matching user-provided text descriptions submitted via the Ask Home chatbot with AI-generated descriptions of camera footage.

However, hands-on testing reveals significant limitations when managing multiple animals of the same species. In a multi-cat household featuring pets named Boone, Smokey, and Osa, the system repeatedly misidentified all three felines as Smokey. According to Google Home product manager Rudra Bhatt, while the underlying Gemini model handles different species well, it "can struggle to distinguish between multiple pets of the same type." Furthermore, the feature is currently restricted to indoor cameras, preventing users from setting up outdoor monitoring or automated digital pet doors.

These identification errors disrupt connected smart home integrations. For instance, attempting to use Google's Help Me Create tool to trigger an Aqara Pet Feeder (previously paired with an Aqara G3 camera) resulted in overfeeding, as the Nest Cam could not differentiate which cat was at the bowl. Because the Google Home app lacks options to limit how frequently an automation runs, the feeder triggered continuously. Misidentifications also led to incorrect security summaries, such as falsely reporting a cat in the garage when the animal was actually in the laundry room.

For smart home practitioners and developers, these shortcomings highlight the gap between generative AI descriptions and robust computer vision. While Nest cameras have evolved from basic motion alerts to animal detection since the Nest Cam IQ in 2017, relying on text-based matching rather than dedicated visual training profiles remains highly unreliable. Until Google refines its multi-pet recognition capabilities, users seeking dependable tracking may find continuous video recording far more practical than automated AI summaries.