Google’s Pet Memory Can Name the Wrong Cat—and That Makes the Feature Hard to Trust

Updated August 18, 2026 3:26 pm

Google’s new Pet Memory feature for Gemini for Home is meant to identify individual pets by name, but after two weeks of testing, it still could not reliably tell one cat from another. That matters because the feature is designed to power smarter camera alerts, home automations, and pet tracking in Nest cameras.

The problem is bigger than a funny mix-up. If an AI system cannot distinguish Boone from Smokey, then pet-specific notifications, feeding routines, and “where is my cat?” checks become unreliable in the situations where they are supposed to help most.

Google has spent years trying to make Nest cameras more useful than simple motion sensors. Now, with Gemini for Home, the company is pushing that idea further by promising not just object detection, but personalized recognition of the animals in your house. In theory, that could make a smart home feel less like a flood of generic alerts and more like a tailored assistant for pet owners. In practice, at least in one real-world test, the system struggled immediately and often.

The disappointing result is important because pet recognition is one of the clearest examples of where consumer AI is heading. Camera makers want to move beyond “something moved” and toward “your gray cat is by the feeder” or “the dog is on the couch again.” Google is trying to turn that future into a subscription feature, but the present version appears to be too fragile to rely on.

What Google’s Pet Memory is supposed to do

Pet Memory is a Gemini for Home feature that lets Nest cameras remember individual pets and attach names to what they see. Rather than simply reporting an animal in the frame, the system is supposed to identify which animal it is, then feed that information into alerts, summaries, and automations.

That sounds like a modest upgrade, but for households with multiple pets it could be a meaningful one. Knowing that a cat has come in through the door is useful; knowing which cat arrived is better. For owners trying to manage feeding, outdoor access, nighttime routines, or health issues, individualized detection could save time and reduce confusion.

The promise is also part of a broader shift in smart home software. Standard motion alerts are noisy and often useless. AI descriptions, by contrast, can tell users what triggered a clip without forcing them to open the app and scrub through footage. Google has been pushing this direction for some time, and Pet Memory is its most personalized version yet.

Why pet recognition matters

Pet recognition matters because a lot of the most useful home-camera behavior is pet-related. People use cameras to check whether animals are indoors before dark, see which pet is at the door, monitor feeders, watch for illness, and keep an eye on yards or livestock.

In homes with multiple animals, a generic “pet detected” alert is only a starting point. The moment a camera can distinguish between pets, it becomes more useful for routine management and potentially for automation.

• It can cut down on false alerts from the wrong animal.
• It can help owners check whether a specific pet has come home.
• It can support feeder routines tied to one pet instead of another.
• It can create smarter daily summaries for households with several animals.

How the feature performed in real-world testing

It did not perform well enough to trust. After the cats were entered into Google Home, the system repeatedly identified all of them as the first cat added to the database. In this test, that meant Boone, Osa, and Smokey were all commonly labeled as Smokey, even when the camera footage clearly showed different animals.

The failure was not just occasional. The issue persisted after repeated attempts to improve the setup with more detail about each cat’s appearance. Google Home would not accept extra descriptive information, and there was no obvious way to upload reference photos or correct the model when it made mistakes.

That limitation is crucial. A feature that promises named recognition but offers no meaningful training or correction path is difficult to improve from the user side. If the system guesses wrong, the user appears to have no practical way to teach it otherwise.

The cats involved in the test

The household in the test case includes three cats with distinct looks: Boone, a tuxedo cat with white paws; Smokey, a large gray-and-white cat; and Osa, a new tabby kitten. Despite those differences, Pet Memory kept collapsing them into one identity.

The result was not just a technical annoyance. It undermined the feature’s main value proposition: personal recognition.

The system kept assigning the wrong cat’s name to nearly every clip, making the feature unusable for the kind of pet tracking it is designed to support.

Why did the AI get it wrong?

Google has not publicly explained why the feature failed in this test, but the behavior suggests the product may be doing only shallow identification rather than building a robust visual profile for each animal. The setup process appears to ask for a name and breed, then tries to infer the rest on its own.

That approach may work when pets are highly distinct or when there are only two of them. It is much more likely to fail in a home with several similar-sized cats, changing lighting, fast movement, or partial views of faces and bodies. Cats also tend to be photographed from awkward angles, which makes any recognition system work harder.

Unlike a carefully trained computer vision model with labeled examples, the current Pet Memory workflow gives users very little control. That may make the feature easy to set up, but it also makes it hard to fix when recognition breaks down.

What’s missing from the setup

The most obvious missing pieces are a proper training step and a correction loop. There is no visible process for teaching the system with photos, and no reliable way to tell it when it has made a wrong match.

That is especially problematic for pet identity, where accuracy is the whole point. If a system cannot be retrained in the home environment, it may never get better on its own.

Can Pet Memory trigger automations?

Yes, in theory, but the automation tests were also undermined by the identification problem. Google Home now lets camera detections act as triggers, and its Help Me Create tool can generate automations from plain-language instructions.

In this case, the goal was to use a Nest camera to trigger an Aqara pet-feeding scene whenever a particular cat was spotted near the feeder. The idea was simple: if the camera sees Smokey, dispense Smokey’s food; if Boone appears, trigger Boone’s portion instead.

Google Home did create the automation from a text prompt. But because the camera could not distinguish the cats, the feeder was triggered every time either animal approached it. That quickly created an unintended result: too much food in the bowl by the end of the day.

There was also another practical limitation. The Google Home app did not let the user cap how many times the automation could run. That meant a repeated detection loop could keep firing without an obvious safeguard.

How does Google compare with other camera systems?

Google is not alone in adding AI to home cameras, but it may be the first to emphasize personalized pet identity. Other companies have leaned into AI-generated scene summaries, object labels, and animal detection, but not always at the level of named recognition for individual pets.

Amazon Ring, Wyze, and Apple Home, through HomeKit Secure Video, have all moved toward more descriptive alerts. Those systems can sometimes identify an animal as a cat or dog, though they are not always correct. Google’s approach goes one step further by trying to distinguish one pet from another.

That extra step is what makes the feature more ambitious—and more vulnerable. Generic animal detection is useful even when it is imperfect. Personal recognition only helps if it is consistently right.

Where Nest has been ahead before

Nest has historically been strong in camera intelligence. The Nest Cam IQ, launched in 2017, could distinguish between people and pets. Google later added broader animal detection and then AI-driven text descriptions of camera events.

Pet Memory is a continuation of that line, but with a more ambitious goal. Instead of saying there is a pet, the system aims to say which pet it is. That makes the feature feel like a logical next step, but also a much harder technical problem.

What the subscription costs

Pet Memory is tied to Google’s highest Google Home subscription tier, the Advanced Plan, which costs $20 per month. That price matters because it places the feature in a premium category rather than a basic smart-home add-on.

It also currently works only with Nest cameras, and even then only with indoor models. That means people hoping to use it for outdoor pet monitoring, driveway watching, or broader yard surveillance are out of luck for now.

Those restrictions significantly narrow the feature’s potential audience. Households that might benefit most from pet identification often rely on cameras in multiple locations, including porches, yards, garages, and doorways. Limiting the feature to indoor Nest cameras reduces its usefulness immediately.

Key limitations at a glance

• Requires Google Home Advanced Plan at $20 per month.
• Works only with Nest cameras.
• Currently limited to indoor Nest Cam models.
• Offers no clear photo-based training flow.
• Provides no obvious manual correction tool for mistaken identities.

Why false identification can be worse than no identification

False identification can be worse than no identification because users may act on bad information. A wrong label is not merely a cosmetic error if it affects feeding, safety checks, or health monitoring.

In this test, one camera summary claimed a cat was in the garage, which would normally be a concern. But the footage later showed the wrong cat had been identified, and the animal was actually in the laundry room. That sort of error turns a helpful alert into a misleading one.

When camera footage is used for a vague overview, mistakes are annoying. When it is used to decide whether a pet is inside before dark or whether a feeder should run, mistakes become operational problems.

The real issue is not that the AI missed a cute detail. It is that an inaccurate pet identity system can send owners chasing problems that do not exist.

How useful is Home Brief?

Home Brief is only modestly useful in its current form. The feature is intended to provide a daily summary of activity detected by connected devices, including pet-related events. But in testing, it tended to reduce all the animal movement in the home to a vague statement that there had been “many pet interactions.”

That is too broad to help much. For a busy pet household, the whole point of an end-of-day summary is to quickly reveal who came and went, what happened, and whether any unusual event needs attention.

Instead, the summary was too generalized to be a strong decision-making tool. It may be better than checking each clip manually, but only barely.

What this says about AI in the smart home

This is another reminder that consumer AI is often more impressive in demonstrations than in the messiness of an actual home. A living room, laundry room, porch, feeder, hallway, and garage create a far harder environment than a controlled test lab.

Smart-home AI is being sold as a convenience layer, but the cost of being wrong is different depending on the task. Misreading a scene summary is one thing; misidentifying a pet tied to a feeding routine is another. The more a system is used for real household decisions, the more accuracy matters.

That does not mean the category is doomed. In fact, the progress is obvious. Security cameras are much more descriptive than they were a few years ago, and generative AI is clearly improving how much context they can provide. But Pet Memory shows the gap between “better than motion detection” and “reliable enough to trust.”

What still works well without pet identity

Even without personalized recognition, Nest cameras can still be valuable. Continuous recording remains one of the most useful features for pet owners because it lets them scrub back through a day’s events and reconstruct what happened.

That is particularly helpful if a cat is sick, a bird goes missing, or an animal gets into something it should not have eaten. For those kinds of questions, full video archives are still more reliable than AI-generated summaries.

In other words, the cameras do their best work when they document events instead of trying to interpret them too aggressively.

Why the feature still feels early

Pet Memory feels early because it lacks the controls and robustness that would make it trustworthy in a complex home. The vision is compelling, but the execution currently depends too much on guesswork.

Google has not shown a system that users can actively teach, tune, or audit. That is a major weakness for a feature marketed around unique identities. Without correction tools, the AI’s mistakes are more likely to become repeated mistakes.

The irony is that the most reliable answer for many owners may still be to keep using the cameras the old-fashioned way: review the footage yourself, compare appearances manually, and trust the recording more than the label.

Timeline of Google’s pet and camera AI push

Google’s pet recognition effort did not appear overnight. It is the latest step in a longer evolution of Nest camera intelligence that has gradually moved from basic motion detection to more descriptive AI summaries.

What comes next?

What comes next depends on whether Google can improve identification accuracy and give users more control. If the company adds training photos, correction options, or stronger safeguards for automations, Pet Memory could become genuinely useful.

For now, though, the feature looks more like a promising beta than a finished product. The concept is easy to understand and easy to want. The reality, at least in this household test, was that the AI still could not keep the cats straight.

That leaves Pet Memory in an awkward position: smart enough to sound impressive, not smart enough to depend on. For pet owners, dependability is the whole point.

Bottom line

Google’s Pet Memory is an ambitious attempt to make Nest cameras more personal, but its current performance shows how hard individualized pet recognition really is. When the system cannot reliably tell multiple cats apart, its alerts, summaries, and feeding automations become too error-prone to trust.

AI-powered home cameras are clearly moving in the right direction. The problem is that this specific feature may have arrived before it was ready.