Google's new Pet Memory feature, part of Gemini for Home, cannot reliably tell three cats apart. In a two-week test on Nest cameras across a three-quarter-acre suburban home, the system labeled every cat as Smokey — the first pet entered into the profile — despite being fed distinct names, breeds, and coat descriptions for Boone, Osa, and Smokey. The feature is gated behind the top-tier Google Home Advanced Plan at $20 a month and works only on indoor Nest Cams.
Pet Memory is Google's pitch for turning generic animal-detection alerts into per-pet identification, letting Nest cameras tell owners which specific pet triggered a notification. The company positions it as the next step beyond the animal-versus-person distinction Nest cameras have offered since the Nest Cam IQ launched in 2017.
“Pet Memory promises to teach your Google Home smart home who your pets are, allowing connected Nest cameras to go beyond simply telling you they saw an animal to which animal they saw.”— Jennifer Pattison Tuohy, Senior Reviewer, Smart Home
Setup runs through the Ask Home chatbot in the Google Home app. Users type a pet's name and breed, and Gemini for Home saves the profile to memory. There is no training step, no photo upload, and no correction workflow when the model gets it wrong. Attempts to add descriptive detail — a tuxedo cat with white paws, a tabby kitten, a large gray-and-white cat — were rejected by the system, which said it could not process the extra information.
Key facts
- 01Google's Pet Memory launched as part of Gemini for Home, gated behind the $20-a-month Google Home Advanced Plan.
- 02After two weeks of testing on three cats, the system labeled every cat as Smokey, the first pet entered.
- 03Pet Memory works only on indoor Nest Cams, ruling out digital-pet-door use cases at exterior doors.
- 04The feature accepts a name and breed but rejects added visual details and offers no photo upload or correction workflow.
- 05Misidentification broke a Help Me Create automation tied to an Aqara Pet Feeder, overflowing the bowl by end of day.
The practical failures cascade from the identification problem. Asking the Google Nest Hub when Smokey was last seen surfaced a video of Osa, the kitten, playing in another room. Google Home Brief, which summarizes the day's device activity, defaulted to generic lines like "there were many pet interactions." In one case it reported Smokey eating in the garage — a false alert that on review turned out to be Boone eating in the laundry room, with no cat in the garage at any point that day.
Automations built on top of Pet Memory inherit the same flaw. A Help Me Create automation set to run an Aqara Pet Feeder scene whenever the laundry-room Nest Cam saw Smokey triggered for both cats indiscriminately. A second automation for Boone did the same. Google Home also does not let users cap how many times an automation fires per day, which the third-party Aqara app does allow. The end result across one day of testing was an overflowing food bowl.
Pet Memory does not appear to build a per-animal visual profile from user input. The behavior across the two-week test suggests the system latches onto the first pet name registered and applies it broadly to any cat the camera sees, rather than distinguishing coat pattern, size, or markings. Google was asked for detail on how the feature is supposed to work and had not responded at the time of the review.
Competing camera platforms have added AI-generated descriptions of what a camera sees, including Amazon Ring, Wyze, and Apple Home via HomeKit Secure Video. None of those systems currently claim per-pet identification. Google Home is the first mainstream smart-home platform to sell the feature — which makes the gap between the pitch and the delivered behavior more conspicuous, particularly at a $20-a-month price point that sits above most competing camera subscriptions.
“Frankly, misidentification is worse than doing nothing at all.”— Jennifer Pattison Tuohy, Senior Reviewer, Smart Home
The core Nest camera experience remains intact. Gemini for Home's descriptive alerts still work — a chicken at the back door, a cat on the porch, a dog on the counter — and scrollable footage is available for review. The feature that broke is the one Google added on top: the promise that the camera knows which specific pet it is looking at. For households with a single pet, the distinction may not matter. For multi-pet homes, which are the ones Pet Memory is nominally designed to serve, the feature does not currently function as advertised.
Per-pet computer vision is a solvable problem — dedicated pet cameras have shipped facial-recognition features for years, and the underlying model work is not novel. What Google shipped reads more like a name-tag layer bolted onto generic animal detection than a trained per-pet classifier. Fixing it likely requires a photo-upload enrollment flow and a real training step, neither of which exists in the product today.
The broader read for Gemini for Home: shipping AI-labeled features at a premium subscription tier raises the bar for reliability, and a $20-a-month plan that misidentifies household pets on day one is a hard sell. Google can patch the feature — and probably will — but the incident is a reminder that generative-AI marketing is now running well ahead of what generic vision models can reliably do in a consumer smart-home context. Anthropic-style enrollment and correction workflows are the missing piece, and whichever camera vendor ships them first will own the multi-pet market.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
