Google's Pet Memory Draws a Blank on My Cats

Gemini for Home's promise that Nest cameras can identify individual cats by name falls short, as the system struggles to differentiate between pets like Boone and Smokey. This significant limitation renders pet-centric smart notifications and automations largely ineffective.

My smart home setup has proven invaluable for pet care, with security cameras being particularly useful for monitoring my diverse animal companions. However, the constant stream of generic notifications often buries crucial alerts. Consequently, when Google unveiled its new Pet Memory feature for Gemini for Home, I anticipated it would be the ideal solution to this challenge.

Pet Memory aims to educate your Google Home smart home about your specific pets, enabling connected Nest cameras to not just detect an animal, but to identify *which* animal. This individualized recognition is designed to allow your smart home to adapt its functions to your pets' unique needs.

My intention was to leverage this feature to remotely monitor my three cats from my home office, differentiate them, and automate their feeding based on which cat was at the feeder. Yet, after two weeks of thorough testing, Google's AI consistently failed to distinguish my felines.

Residing on a three-quarter-acre suburban lot, I've embraced a lively household of animals, including a flock of free-ranging chickens, a large Wirehaired Pointing Griffon named Gus, two indoor/outdoor cats, and a new kitten. (My husband has, regrettably, vetoed a goat and a horse.)

As a smart home reviewer for The Verge, I frequently test connected technology, and I've found pet monitoring around my home and yard to be a prime application for security cameras. They help protect my chickens from predators, my lunch from Gus, and serve as a digital pet door for the cats. A notification from a video doorbell or outdoor camera signaling an animal at the door often indicates a cat seeking entry.

While I don't need to name my chickens or distinguish my single dog, I envisioned three key benefits Pet Memory could offer for managing my three cats. First, it could identify which cat desired entry, reducing false alerts from a neighbor's cat or a chicken. Second, it could confirm all my pets were safely indoors before nightfall. And finally, it could enable personalized, automated feeding for each feline.

The ability to recognize individual pets marks a significant advancement in smart security camera technology. While some specialized pet cameras offer facial recognition, most standard security cameras historically only provided motion alerts. Recent strides in machine learning and generative AI are now delivering more descriptive notifications. Instead of a simple "motion alert," cameras are increasingly specifying what they observe, eliminating the need to open an app and wait for a video clip to load.

Nest has historically led in this area, particularly concerning animal detection. Its Nest Cam IQ, released in 2017, could differentiate people from pets, and Google later extended animal detection to other Nest cameras. More recently, Google introduced AI-powered text descriptions detailing camera observations.

Competitors like Amazon's Ring, Wyze, and Apple Home (via HomeKit Secure Video) have also incorporated AI-driven descriptions, often including the type of animal detected (though not always accurately). However, Google Home positioned itself as the first to offer personalized pet identification.

Nevertheless, the feature comes with notable limitations and costs. Pet Memory necessitates the Google Home Advanced Plan, priced at $20 monthly. Although integrated into Google Home, it currently functions exclusively with Nest cameras and, more restrictively, only with *indoor* Nest Cams. This rendered it useless for my intended "digital pet door" upgrade.

Undeterred, I strategically placed several indoor Nest Cams, hoping Pet Memory could effortlessly track which cats were inside, ensuring their safety before dark. My ideal scenario involved simply asking, "Hey Google, when did you last see Smokey in the house?"

Despite my efforts to describe my three cats to Google Home, the system consistently identified all of them as "Smokey," the first cat I registered. For instance, querying the Google Nest Hub about Smokey's last sighting would often display an adorable video of the new kitten playing in my daughter's room.

My testing suggests that Pet Memory does not construct individual visual profiles for each pet. There's no training phase, no option to upload pet photos, and no mechanism for correcting errors. Users input a name and breed, and the system attempts to infer the rest. When I tried providing more specific details—Boone is a tuxedo cat with white paws, Osa is a tabby kitten, and Smokey is a large gray-and-white cat—the system indicated it couldn't process the additional information and continued to label every cat as Smokey.

I reached out to Google for clarification on why the system appeared to be underperforming, but the company has yet to provide a detailed response.

Despite these setbacks, I persevered, attempting to integrate Pet Memory with my Aqara Pet Feeder for smarter dispensing. Previously, I had configured the feeder to activate when Aqara's G3 camera detected any animal, as it lacks species or individual identification capabilities.

With a Nest Cam, I aimed to calibrate the feeder to dispense precise food portions upon identifying a specific cat, thereby preventing overeating without relying on RF tags or other proprietary solutions smart feeders often employ (my cats refuse collars).

This endeavor also served as an opportunity to test Google's new capabilities: using camera detection events as automation triggers and the "Help Me Create" feature, which can generate automations from simple descriptions.

I instructed the Ask Home chatbot in the app: "Run my Aqara 'Feed Smokey' scene whenever the Nest Cam in the laundry room sees Smokey," and it successfully configured the automation. I then set up a separate one for Boone. However, I encountered several issues.

The primary and most significant problem remained the system's inability to distinguish Boone from Smokey, causing the feeder to activate whenever either cat approached. A secondary issue was the Google Home app's lack of control over automation frequency (a feature available in Aqara's native automation). Consequently, the automation triggered every time either cat visited the bowl, leading to an overflowing food bowl by day's end.

My hopes were also high for Google's Home Brief, intended to provide a daily summary of pet activity. After requesting more detailed pet activity in my Home Brief, it now largely reports "there were many pet interactions," which offers little useful insight.

In one instance, the system incorrectly reported Smokey eating in the garage—a concerning alert, as cats should not be in that area. However, upon reviewing the associated video clip, it was clearly Boone eating in the laundry room, as intended.

I subsequently checked the garage camera's footage for the day, confirming neither cat had been there. Frankly, misidentification is more detrimental than no identification at all. While mistaking a cat's name can be amusing, errors become serious when relying on the system for pet safety or location tracking.

Ultimately, recognizing individual pets proves far more challenging than simply detecting an animal. While this feature offers a glimpse into the future of AI-powered cameras, it is not yet reliable enough for everyday use. Though AI promises to streamline the tedious task of sifting through camera footage, until it can dependably identify specific events, continuous recording remains more valuable than unreliable AI summaries.

Continuous recording is a strength of Nest Cams, included in the same $20 monthly subscription. Whether monitoring an indoor kitten, a tropical aquarium, or a flock of chickens in the yard, reviewing a day's footage is invaluable for tasks like determining what made a cat ill or how a chicken disappeared (hawks often don't trigger motion alerts). When AI can accurately answer these questions, the investment might be worthwhile; in the meantime, Boone continues to enjoy his abundance of kibble.