How Furriq Identifies Cat Breeds: Our Full Methodology

Most cat breed identifiers are a black box. Upload a photo, get a breed name, no explanation. That hides the real story: a photo can only show what is visible, and visible features are clues, not ancestry.

Furriq works the other way. Every report is framed as visible-trait clues with an honest uncertainty note, because that is all a photo can support. This page shows you exactly how that works: what the AI checks, how confidence is expressed, where the method breaks down, and what Furriq will never claim. Read it before you trust a result. Then test it yourself with the free cat breed scanner.

What the AI actually checks

Furriq's model reads seven families of visible features from your photo:

| Marker family | What the AI looks for | | ----------------------- | --------------------------------------------------------- | | Coat pattern | Tabby, pointed, ticked, spotted, marbled, solid, bicolor | | Color | Distribution, points, white markings, blue-gray, odd eyes | | Hair length and texture | Short, long, semi-long, double coat, wavy, hairless | | Face and head shape | Round, wedge, square muzzle, profile, stop | | Ears | Size, set, tufts, folded | | Body type | Size, build, leg length, tail carriage | | Photo quality | Brightness, sharpness, contrast, resolution, framing |

That adds up to 40+ individual markers across the seven families. Each marker is checked against 61 breed profiles, and every profile carries its own key markers, look-alike notes, and mix-identification tips. The scan itself takes seconds.

One more rule sits inside the model's system prompt, and it is worth quoting: "Focus on visible traits." The model is also told, in plain words, not to claim DNA proof or guaranteed breed identification, and to default to "domestic mix" unless visual pedigree clues are strong. Those two constraints shape every report you get.

How a scan runs

Three steps, and the first one never touches a server.

1. Your browser checks the photo first. Furriq measures brightness, sharpness, contrast, resolution, and aspect ratio locally, on your device. Each problem costs points: a dark photo loses 24, a soft or blurry image loses 26, a photo smaller than 700px on its shortest side loses 18, low contrast loses 10, and an extreme crop loses 12. The result is a 0-100 score. 78 or above is good, 55 to 77 is usable, below 55 needs work, and the photo gets specific fix-it advice.
2. The AI reads the visible traits. The model examines the photo against the seven marker families and returns a fixed structure: a summary, a list of visible traits, the likely breed clues, an uncertainty note, and retry tips.
3. You get a clue report. The summary states which breeds your cat resembles and whether a domestic mix is more likely. The traits and clues explain why. The uncertainty note says what the photo could not show.

The breed part of the report is never a single percentage presented as fact. It is written as clues with reasoning, and the uncertainty field is required, not an afterthought.

How confidence is expressed

Confidence shows up in three places, and none of them overclaim.

• Clue language. Results say your cat resembles a breed, or carries markers seen in it. They never say your cat is that breed.
• An explicit uncertainty note. Every report includes one. The default, when the model does not supply better wording, is: "This is a visual estimate from a photo, not DNA or pedigree proof."
• Domestic mix as the baseline. Most pet cats have no breed ancestry. Roughly 95% of cats are domestic shorthairs or domestic longhairs. The model is instructed to start from that assumption and only move away from it when the visual evidence is strong.

That is the honest version of confidence. A confident guess with no caveats is worse than a cautious answer with reasons, because the first one sends you down the wrong research path.

Where the method fails

Photo-based identification has real limits. Furriq will tell you about them, and here they are in full:

• Kittens. Faces, ears, and coats change a lot in the first year. A 10-week-old kitten can look like almost anything, and the scan will say so.
• Mixed breeds. A cat with three breeds in its background blends the markers. The scan may surface a breed that contributed one visible feature, like ear tufts or a coat color, without meaning that breed dominates the mix.
• Breeds that look alike. Maine Coon and Norwegian Forest Cat are separate breeds separated by small details: muzzle shape, profile, coat layering. Russian Blue, Chartreux, and Korat are even closer. When the differences are subtle, the report says the match is uncertain rather than picking one.
• Bad photos. Dark rooms, blur, tiny images, heavy filters, and extreme crops all hide markers. The local quality check catches most of these before the scan, and tells you what to fix.
• Wrong formats. Furriq accepts JPEG, PNG, and WebP up to 8MB. iPhone's default HEIC format is not supported in most browsers, so convert it first. The tool tells you exactly that.
• Rare breeds. Furriq covers 61 breed profiles with 121 side-by-side comparisons. That is a lot, but it is not every breed, and mixes can produce looks no profile predicts.
• More than one cat. A photo with two cats confuses any scanner. One cat per photo.

None of these are excuses. They are the reason the report is a hypothesis you can test, not a verdict you have to accept.

Furriq is not a DNA test

This is the most important sentence on this page: Furriq does not prove breed. It never has, and the model is explicitly forbidden from claiming otherwise.

| | Furriq photo scan | DNA test | | ------------- | ---------------------------- | --------------------------------- | | What it reads | Visible traits from a photo | Genetic markers from a cheek swab | | Result | Breed clues with uncertainty | Breed percentage breakdown | | Time | Seconds | 2-4 weeks | | Cost | Free | $80-150 | | Needs | A clear photo, no sign-up | A mailed swab and a lab | | Proves | What your cat looks like | Part of what your cat is |

Both tools have a job. The photo scan is for curiosity: fast, free, and good at pointing you toward breeds worth researching. A DNA test is for paperwork, health screening, or the rare case where an exact breed percentage matters. Furriq says plainly which one it is, and it is the first one.

What happens to your photo

Furriq does not store uploaded photos. Image retention defaults to none, which means the photo is used for the scan and not kept. There is no account needed to scan, no photo gallery of your pets sitting on a server, and no sign-up wall between you and the result.

The privacy story starts even earlier. The brightness, sharpness, and contrast check runs in your browser before anything leaves your device, so a bad photo never gets transmitted at all. The only things sent are the image and a few local diagnostics, and only when the photo clears the quality gate.

How we keep the tool honest

Honesty here is engineered, not hoped for. The system prompt forbids DNA claims and guaranteed identification. The output schema forces an uncertainty field. The retry tips remind users that coat pattern labels are clues, not proof of ancestry, and that pedigree or DNA evidence is the only route to a confident breed claim.

There is also a feedback loop. Every report page has a feedback form, and the methodology gets updated when the model's behavior or the breed data changes. This page is part of the product, not marketing for it.

Try it

You now know exactly how the tool thinks. Upload a clear, well-lit photo of your cat and see the method in action: try the free cat breed scanner. It takes seconds, needs no account, and the report will tell you what it sees, what it suspects, and where it is not sure. That last part is the feature.