Google's Pet Memory AI Can't Tell One Cat From Another
A two-week real-world test found Google's new feature for identifying individual pets on Nest cameras keeps calling every cat by the same name, with consequences that go beyond the comical.
Key points
- Google's Pet Memory feature, part of the Google Home Advanced Plan at $20 a month, promises to teach Nest cameras to recognise individual pets by name.
- After two weeks of testing with three cats, the system labelled every animal as Smokey, the first cat entered into the system.
- The feature works only on indoor Nest cameras, not outdoor ones, and offers no photo-upload tool or way to correct misidentifications.
- Misidentification triggered a pet feeder for the wrong cat repeatedly, overflowing the food bowl within a single day.
- Google had not responded to questions about the failures at the time of writing.
Google recently launched Pet Memory, a feature for its Google Home smart home platform that is supposed to teach connected Nest cameras to tell your pets apart by name. Instead of a vague "motion detected" alert, your camera would say "Boone is at the door" or "Smokey hasn't come inside yet." A reviewer at The Verge ran the experiment with three cats, a dog and a kitten, over two weeks.
The short verdict: it doesn't work.
What is Pet Memory supposed to do?
Pet Memory lets you type your pet's name and breed into the Google Home app's chat assistant. Gemini for Home, Google's AI built into the smart home platform, saves that information, and cameras are then meant to cross-reference what they see against those profiles to send personalised alerts.
Most security cameras still only detect motion. Smarter ones from Ring, Wyze and Apple's HomeKit system can now say whether they saw a person or an animal. Google claims to go further and name the animal. Access costs $20 a month on the top-tier Google Home Advanced Plan, and the feature runs only on indoor Nest cameras.
Does it actually recognise individual pets?
No, not reliably. The three cats had clearly different appearances: Boone, a tuxedo cat with white paws; Osa, a tabby kitten; Smokey, a large grey-and-white cat. Every sighting, regardless of which cat walked past the lens, was labelled Smokey.
There's no setup process where you upload photos. No correction tool exists. You type a name and a breed, and the system tries to work the rest out alone. When the tester added more detailed descriptions, the app said it couldn't process the extra information.
Why that matters in practice:
| Problem | Real-world result |
|---|---|
| Every cat called Smokey | No way to know which cat was home |
| Wrong cat triggered feeder | Food bowl overflowed in one day |
| "Smokey in garage" alert | Footage showed Boone in the laundry room |
| Daily Home Brief summary | Reported "many pet interactions" with no useful detail |
One alert said Smokey was eating in the garage, which raised genuine concern. Checking the footage showed Boone in the laundry room, exactly where he should be, and the garage camera confirmed no cat had been there all day.
We covered a similar failure in August, when Whisker's Litter-Robot 5 Pro also couldn't distinguish between two visually distinct cats after six months of testing. Pet facial recognition is proving harder to ship reliably than the product announcements suggest.
What does this mean for pet owners thinking of signing up?
Skip Pet Memory for now if accurate pet identification is the goal. A wrong name on a notification is one thing. Trusting the system to confirm all your animals are safely inside before dark is another, and this feature isn't ready for that kind of responsibility.
The plain-English camera descriptions that Gemini for Home produces are genuinely better than a simple motion alert and worth having on their own merits. But personalised pet identification, the headline feature, needs significant work before it earns a place in anyone's daily routine.
Google has not explained why the system behaves this way or when a fix might arrive.
