Skip to main content

AI pet matching explained

How AI Pet Matching Works

AI-assisted pet matching compares available photographs and reported characteristics to identify meaningful similarities between a missing animal and a found animal. It can help organize possible connections, but it cannot independently prove identity, ownership, or a successful match.

Development status. KCAI’s reporting and matching workflows are being developed and tested. This page explains the intended process and the principles behind responsible AI-assisted matching; it does not mean every described tool is publicly operational today.

On this page: the six-step process · how photographs are compared · why human review is required · what is live today

The process

The six steps, and where KCAI actually is

Responsible matching is a sequence, not a single act of recognition. Each step narrows the field a little and each one can also rule a possibility out. The status tag on each step below is KCAI’s real position today, not a roadmap gloss.

  1. A lost-pet or found-pet report is submitted

    In controlled testing

    Everything downstream depends on this. An owner describes an animal they know well; a finder describes an animal they have just met and may not be able to approach. Those two accounts are rarely symmetrical, which is why the intake asks for observable detail rather than confident conclusions — what colour, what markings, what size, what it was wearing, where and when.

  2. Photographs and identifying details are organized

    In controlled testing

    Free text is turned into comparable fields. “Medium brown dog, white chest patch, floppy ears” becomes separate values for species, size, primary and secondary colour, pattern and ear carriage. This is unglamorous and it is where most of the value is: two reports written in completely different words cannot be compared until they are described on the same terms.

  3. Visual similarities are evaluated

    Under development

    Photographs are compared against photographs, and structured details against structured details. The output is a ranking of which pairs are worth a person’s attention — not a verdict. A high similarity score between two black-and-white shorthaired cats in the same suburb is a starting point, and no more than that.

  4. Location and timing are compared

    Under development

    Two reports that look alike still have to be physically possible. Dates have to run in the right order, and the distance between them has to be plausible for the animal in question. This step is as much about discarding as confirming.

  5. Potential matches receive human review

    Under development

    A person looks at the pair side by side, with everything both reporters supplied, and asks whether the resemblance holds up or falls apart on a detail the comparison could not weigh. Nothing reaches anyone because software ranked two records highly.

  6. Owners, finders, shelters, or rescues determine the appropriate next action

    Under development

    The decision belongs to the people involved. KCAI’s role ends at putting a well-evidenced possibility in front of them, along with what is known and what is not.

To be unambiguous about it

KCAI does not currently run steps three through six. There is no automated comparison producing possible matches on this platform today, and no review queue behind one. The first two steps exist as forms in controlled testing, which means a submission does not create a public case, does not start any matching, and does not notify anyone. The sequence above is the design being built toward and the standard it will be held to.

Inputs

What information can support a comparison

These are the categories that carry useful signal. They are grouped by the kind of question they answer, because a comparison is stronger when independent kinds of evidence agree than when one category agrees emphatically.

What kind of animal

  • Species — the first and hardest filter to get wrong.
  • Dog or cat — recorded separately because the two behave differently once loose, and that changes how far and how fast they move.
  • Approximate breed or physical type — treated as an impression, never as a fact. “Terrier-ish, wiry, about knee height” is more useful than a confident wrong breed name.
  • Approximate size — often described by comparison to a familiar object or another animal.
  • Body proportions — leg length against body length, chest depth, overall build.

What it looks like

  • Coat colour — primary and secondary, with the caveat that people name colours very differently under different light.
  • Coat pattern — tabby, brindle, patched, ticked, solid; where the boundaries fall matters more than the label.
  • Facial markings — a blaze, a mask, an asymmetric patch. Asymmetry is especially useful because it is hard to coincide.
  • Ear shape and position — erect, folded, cropped, one that sits differently from the other.
  • Muzzle shape — length and breadth, and any greying around it.
  • Tail characteristics — length, carriage, a kink, a curl, a docked tail, a distinctive tip.
  • Distinctive scars or features — a healed injury, a missing or cloudy eye, a lump, an unusual gait. Often the single most decisive category.

Circumstance

  • Collar, harness, or clothing — colour, type, tags, a bandana. Strong evidence when present, and absence proves nothing: collars come off.
  • Last known or found location — the area an animal went missing from, or was picked up in.
  • Reported dates and times — when it was last seen, and when it was found.

Circumstance is what turns a resemblance into a plausible story, and what most often exposes one that cannot be true.

No single characteristic should automatically establish identity. A matching coat colour, a matching breed guess, even a matching distinctive marking is one line of evidence among several. Categories can also be wrong in good faith — an owner and a finder can describe the same animal’s colour differently and both be doing their honest best. Agreement across independent categories is what carries weight.

Images

How photographs are compared

Computer-assisted image comparison measures how similar two pictures are. Broadly, software reduces each photograph to a set of numbers describing visible patterns and structure — where light and dark regions fall, the shape of edges and outlines, the arrangement of markings, proportions between features — and then measures how close two sets of those numbers are.

That is a genuinely useful thing to be able to do across thousands of reports. It is also narrower than it sounds: it tells you two photographs resemble each other. It does not tell you they are photographs of the same animal.

Conditions that weaken a comparison

  • Different angles — a profile and a head-on shot of one animal can measure as less similar than two profiles of different animals.
  • Lighting differences — indoor tungsten, overcast daylight and a camera flash can each render the same coat as a different colour.
  • Distance from the camera — a distant animal simply contains fewer pixels of usable detail.
  • Image resolution — a screenshot of a shared social post has usually been compressed several times over.
  • Motion blur — frightened animals move, and a blurred outline destroys exactly the edge detail a comparison relies on.
  • Partially obstructed features — grass, fencing, a doorway, another animal, or a hand holding the pet.

Ways the animal itself changes

  • Wet or dirty coats — a soaked coat can look darker, shorter and a different texture entirely.
  • Seasonal coat changes — many animals carry a visibly different coat in summer than in winter.
  • Grooming changes — a clipped coat can alter apparent colour, pattern and body shape at once.
  • Weight changes — an animal missing for weeks may have lost noticeable condition.
  • Growth in young animals — a kitten or puppy can change proportions substantially in a month, and markings shift as the coat comes in.

These are the reason a low similarity score is not evidence against a match, any more than a high one is evidence for it.

What we do not claim

KCAI is not described as facial recognition. That term carries specific technical meaning and specific expectations, and it will not be applied to this platform unless and until such functionality is built and technically verified.

We also make no claim about a particular model, algorithm, biometric system or proprietary method. Nothing of that kind is running on this platform today, so naming one would be an invention. The explanation above is how computer-assisted image comparison works in general, offered so you can judge what it is reasonable to expect from it.

Plausibility

Why location and timing matter

Appearance says a match is possible. Location and timing say whether it is plausible. A visually convincing pair can still be the wrong pair, and this is usually how you find out.

Signals that a similar animal is an unlikely match

  • The locations are implausibly far apart for the animal, the elapsed time and the terrain in between.
  • The report dates conflict — most plainly, an animal cannot be found before it went missing.
  • A verified sighting places the missing pet elsewhere at a time that cannot be reconciled with the found report.
  • Shelter intake timing does not align — an animal already in care on a given date cannot have been at large somewhere else on it.
  • Physical descriptions changed inconsistently — coats and condition change over time in predictable directions; a change in the wrong direction is a warning.

Why distance alone must not eliminate a possibility

Animals travel further than people expect, and they are also transported. A frightened dog can cover ground quickly. A cat can be shut in a vehicle, a van, or a garage and released somewhere else entirely. Well-meaning people move found animals — to a vet, to a relative, to a rescue in another town.

So distance is weighed, not used as a cut-off. KCAI does not publish a maximum travel distance or a geographic limit, because any such number would be invented and would quietly discard real matches.

The decisive step

Why human review is required

This is the part of the process that cannot be automated away, and the part that protects people from the worst thing this kind of system could do: telling someone their pet has been found when it has not.

A potential AI-assisted match is a lead for review — not proof that two reports describe the same animal.

A comparison can only weigh what was submitted, and it weighs everything with the same flat confidence. A reviewer can tell that one photograph is unusable, that a finder was describing an animal they only saw at dusk from a distance, or that a single asymmetric scar settles the question on its own. What a reviewer works through:

The evidence itself

  • Image quality — is there enough usable detail here to conclude anything?
  • Conflicting characteristics — one clear, well-described difference outweighs several vague similarities.
  • Whether distinctive markings truly align — not just present in both, but the same shape, in the same place, on the same side.
  • Location and timing — does the sequence of events actually work?

The animal’s record

  • Sex and approximate age — a frequent and decisive mismatch, and frequently misreported in good faith.
  • Microchip information — where a chip has been scanned, it outranks everything else on this page.
  • Veterinary or adoption records — dates, procedures, previously documented markings.
  • Shelter intake information — when and where an animal entered care, and in what condition.

What the people know

  • Owner knowledge — owners know things no field captures: a habit, a reaction to a name, an old injury, exactly where a patch sits.
  • Finder observations — behaviour, condition, whether the animal was approachable, what it did next.

Two questions to the right people routinely resolve a pair that no amount of image comparison could settle.

Capabilities, in general terms

What AI-assisted comparison can help with

Stated as what this class of tool is capable of, not as a description of KCAI’s live functionality. None of the following is publicly operational on this platform today.

Organize large numbers of reports

Volume is the problem people cannot solve by reading. Structure is what makes a large set searchable at all.

Surface visually similar animals

Bring plausible-looking pairs to the surface instead of relying on someone happening to scroll past the right post.

Identify details people may overlook

A faint pattern boundary or a subtle proportion difference can be measured consistently even when it is easy to miss by eye.

Compare photographs consistently

The same criteria applied the same way to every pair, without fatigue and without the tenth comparison getting less care than the first.

Prioritize reports for human review

Ordering matters when there is more to check than time to check it. The point is to spend human attention where it is most likely to pay off.

Connect information submitted by different sources

An owner, a finder, a neighbour who saw something and a shelter may each hold one piece. Holding them in one structure is what lets the pieces meet.

Hard limits

What AI cannot determine

These are not gaps waiting to be closed by a better model. They are things outside what image and attribute comparison can establish at all.

Identity and ownership

  • Prove ownership. Ownership is a human and sometimes legal question, evidenced by records and history — never by resemblance.
  • Confirm identity with certainty. Two animals of the same breed, colour and pattern can be visually indistinguishable in the photographs available.
  • Read a microchip from a photograph. A chip is read by a scanner in physical contact with the animal. No image contains that number.
  • Resolve legal ownership disputes. Entirely outside what this platform does or should attempt.

Judgement and verification

  • Determine exact breed reliably from appearance alone. Visual breed identification is unreliable even among experienced professionals.
  • Know whether reported information is accurate. A comparison cannot tell an honest mistake from a fact.
  • Determine intent from a photograph. Nothing about circumstance, motive or how an animal came to be somewhere is visible in an image.
  • Replace a physical examination or replace a microchip scan. Both remain the reliable ways to establish which animal this actually is.
  • Guarantee reunification. Many pets are found through neighbours, shelters and persistent local searching. Keep doing all of it.

Read the full page on what AI can and cannot determine →

Complementary, not competing

Where AI comparison sits among the other identification tools

AI-assisted comparison is a way of finding candidates. The tools below are ways of establishing facts. They answer different questions, and the useful arrangement is for the first to lead into the second — never to substitute for it.

Microchip scanning

The strongest single identifier there is, and the one thing on this list that can settle the question outright. If a found animal has not been scanned, that is the first thing to arrange.

Identification tags

Immediate and requires no equipment. Also the easiest to lose, so their absence means nothing at all.

Veterinary records

Dates, procedures, chip numbers and documented markings, from a source with no stake in the outcome.

Adoption records

Establish history and often carry the original intake photographs, which can be compared against a current report.

Current pet photographs

Recent, clear, well-lit pictures from several angles are worth more than any single perfect portrait. Take them before you need them.

Distinctive marking photographs

A close, deliberate photograph of a scar, an asymmetric patch or an unusual ear is often the piece of evidence that resolves an otherwise ambiguous pair.

Shelter intake records

Place an animal at a known location on a known date — which both supports and rules out possibilities.

Owner and finder communication

The most underrated tool here. A short, specific exchange between the two people involved settles more cases than any technology on this page.

After a possible similarity

What responsible next steps look like

Once a possible similarity is identified, the order of operations matters. Verify before anyone is told anything, and involve the right people at the right point.

  1. Review all available photographs

    Every image from both reports, not the two that were compared. A weaker photograph often contains the detail the better one cropped out.

  2. Compare distinctive features

    Work through the specifics side by side — markings, scars, ear carriage, tail. Look actively for a difference, not only for agreement.

  3. Confirm report dates and locations

    Establish that the timeline is physically possible before anyone is contacted.

  4. Contact the appropriate reporting parties through protected procedures

    Reach people without exposing either party’s private details to the other, and without publishing anything that was given in confidence.

  5. Request additional evidence where necessary

    A specific ask — a photograph of the left flank, of the underside of the tail, of a scar — is usually faster and more decisive than more discussion.

  6. Arrange a safe verification process

    Meet in a public or supervised place, or at a veterinary practice or shelter. Personal safety and the animal’s welfare both come before speed.

  7. Scan for a microchip when appropriate

    Where a chip exists this is the step that answers the question. Veterinary practices, shelters and animal control can scan.

  8. Involve shelters, rescues, animal control, veterinarians, or law enforcement when circumstances require it

    An injured animal, a welfare concern, a disputed claim or a situation that feels unsafe should be handed to people with the standing and the equipment to deal with it.

Not automated

KCAI has no automated notification workflow. Nothing on this platform currently contacts an owner, a finder, a shelter or a rescue on its own, and no such message has ever been sent. The steps above describe how a possible match should be handled by people; they are not a description of software that runs today.

Responsible handling

Privacy and data handling in matching

Matching needs some information to be visible and other information to stay private. Those are two different categories, and treating them as one is how reunification efforts end up exposing the people they were meant to help.

Pet details and owner details are handled differently

A description of an animal is meant to be recognised by strangers. A person’s name, telephone number and home address are not. They are held as separate categories rather than one record with parts hidden.

Exact contact and location details are not automatically public

Publishing a precise address alongside a description of what is at it creates a risk that has nothing to do with finding a pet. A public area description is a general one.

Share only what matching actually needs

Enough to recognise the animal, and enough to be reached. Extra personal detail does not improve a comparison and cannot be un-shared later.

Sensitive information requires controlled access

Information that could identify or endanger a person is kept apart from the information that helps identify an animal, and reached by fewer people.

A possible match does not justify exposing private data

A promising lead is a reason to contact someone carefully. It is not authorisation to hand one person’s contact details to another, or to publish them.

Read the detail

The published privacy policy sets out what is collected, what is shown, what is never shown, and the choices available to you.

Privacy and pet-data protection →

Verified status

Where KC AI Pet Match actually stands today

Everything below is the current, verified state of this platform. It is on this page so that nothing above can be mistaken for a description of working software.

Live now

  • The public homepage.
  • This educational resource layer, including the pages on process, limitations, privacy and shelter participation.
  • Published privacy and terms information.

You can read and rely on all of it.

In controlled testing

Not accepting live reports

  • The lost-pet and found-pet report forms.

Submitting one does not create a public case, does not produce a public listing, does not start any matching, and does not notify anyone.

Under development

Not available

  • Automated comparison of reports and photographs.
  • The human review queue behind possible matches.
  • A public, searchable list of reports.
  • Community sighting reports and mapped location information.
  • Shelter and rescue participation.

There is no automated public matching on KC AI Pet Match today. No possible match has been generated for anyone, and no automated notification has been sent. Public workflows will be activated only after they have been securely tested and verified — and each one will be described plainly here before it is switched on.

If your pet is missing right now, do not wait for this platform. Contact KC Pet Search & Rescue, your local shelter, and animal control. If you have found an injured animal, contact a veterinarian or animal control first.

Understand the KCAI Matching Process

The honest version of how this works, what it cannot do, and how your information would be handled. All three are published and readable now.

Free to read. No account is required, and no reporting workflow is live yet.