How the census is measured
This page stands on its own, before a single number exists. It says exactly what the BreedQuest census counts, where it is skewed, what it will not claim, and the privacy line it will not cross — so that when the figures arrive with the 2027 launch, you already know what they are and are not.
What a scan is
Point the BreedQuest app at a dog and take a photo, and an on-device model returns its best guess at the breed, with a confidence score. Each of those results, stripped of the photo and of anything that ties it to a person, is one scan. A scan is a sighting, not a pet. The same dog scanned twice is two scans; a dog nobody ever scans is invisible to the census. That is the raw material — sightings of dogs, identified by a model, contributed by people using a phone. Everything the census reports is built from counting those, and the honest reading of any figure starts there. You can see the same identification step in the app, and try a lightweight version of it in the breed finder.
What gets counted
From each scan the census keeps a short, deliberately coarse record: the breed (or the blend of breeds) the model returned, how confident it was, the estimated adult size band that follows from the identified breed, a region no finer than a country — and only a city where enough separate scans exist to hide any one of them — and the local time-of-day bucket the scan fell in. That is the whole list.
What a scan never carries is as important as what it does. There is no photo kept against it, no GPS coordinate, no account or device identifier, and nothing that links two scans to the same person or the same dog. The record is designed so that it cannot be walked back to an individual, because the census only needs the totals.
Where the sample is skewed
The census counts the dogs BreedQuest users point a phone at, which is not the same as the dogs that exist. Three skews are worth stating plainly, because they shape every number that follows.
First, it is a self-selected sample. People who install a dog-scanning app lean younger, more urban, and more smartphone-comfortable than dog owners as a whole, and the app will catch on faster in some countries than others. Second, it is weighted by sightings, not by dogs. A dog walked twice a day in a busy park gets scanned far more than one that rarely leaves a garden, and a striking or unusual dog draws more scans than a plain one — so the census over-counts the visible and the photogenic. Third, the model has its own blind spots. Breeds it recognises confidently show up more than breeds it hedges on, and rare breeds sit on thin data where a handful of scans swings the picture.
None of this makes the numbers useless. It makes them numbers about scanned dogs, and every published figure is labelled that way rather than dressed up as a national count.
What the data does not say
The census is not a count of how many dogs live in a country. It is not a pedigree registry, and it does not measure which breeds are bred or sold. When a breed's share of scans rises, that means BreedQuest users scanned it more often — which can move with fashion, with where the app grew, or with the model improving, not only with any real change in the dog population. So the census publishes the share of scans and the direction it moves, and stops there. It will not print a headcount of a nation's dogs, because it cannot measure one. The one place mixes are genuinely well suited to this method — every scanned mix is its own combination — is covered on the mixed-breed hub, and a breed as common as the Labrador Retriever will still only ever be reported as a share, never as a number of animals.
Privacy: aggregates only, never a pet or a household
The census only ever reports groups. A figure is published only when enough separate scans sit behind it that no single dog, owner, or household can be identified from it; groups too small to hide an individual are withheld, not shown. No published number carries a name, an account, a home address, or a precise location, and there is no path from a census figure back to a specific pet. We do not sell the underlying scans and we do not share them. The fuller account of what BreedQuest stores and what it never does lives in the privacy policy; the census inherits that position and adds one rule of its own — aggregates only, always.
How to cite the census
Cite the report, its year, the dataset name, and the URL. Each chart in a report also carries its own citation line so a journalist can lift the figure without emailing first. The first edition is the 2027 report; update the year for each edition: