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.
-
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.
-
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.
-
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.
-
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.
-
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.
-
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.
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.
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.
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:
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.
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.
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.
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.
-
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.
-
Compare distinctive features
Work through the specifics side by side — markings, scars, ear carriage, tail. Look actively for a difference, not only for agreement.
-
Confirm report dates and locations
Establish that the timeline is physically possible before anyone is contacted.
-
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.
-
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.
-
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.
-
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.
-
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.
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.
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.
