What the AI Species Claims Actually Mean

The species count is the headline feature of this entire category. It is printed on every box, it leads every listing, and it is the number buyers use to rank one feeder against another. It is also, as far as we can establish, close to meaningless as a purchasing signal - and working out why is the most useful thing we can do for someone shopping here.

To evaluate a claim of 10,000 species you need to know how many species there are. Two authoritative checklists answer that, and both were checked directly for this page rather than taken from a secondary source.

• The IOC World Bird List, version 15.2 (13 April 2026) recognizes 11,227 species, of which 164 are extinct. That leaves roughly 11,063 living bird species on Earth.
• The ABA Checklist, version 8.19 (January 2026) covers 1,161 species for the entire ABA Area - the whole of North America north of Mexico, including Alaska, Canada and adjacent waters.

It would be easy and wrong to call these claims fraudulent. They are probably approximately true in a narrow technical sense: image classifiers are routinely trained on very large public datasets of labeled bird photographs, and a model can carry an enormous label set. Claiming 10,000 classes is not implausible.

The problem is that the claim answers a question no buyer has. Nobody's feeder is visited by 10,000 species. Nobody's feeder is visited by 1,161 species. The relevant number is the couple of dozen granivores that use feeders in your region - and every app in this category covers those, including the one that publishes no count at all.

So the species count does not discriminate between products. Two feeders claiming 10,000 and 6,000 species are, for your yard, making identical claims. Using that number to choose between them is using a differentiator that does not differentiate.

There is a reason to be actively skeptical of the biggest numbers, and it is not marketing cynicism - it is how classifiers behave.

A classifier trained to choose between 10,000 classes has 10,000 opportunities to be wrong on every image. A classifier restricted to the 1,161 species of North America has far fewer, and one restricted to the species plausible at a feeder in your state has fewer still. Adding classes that cannot possibly occur in your yard adds confusion without adding capability.

This is why serious identification tools use location and date as inputs. Cornell's Merlin is the obvious reference point: it narrows its candidates by where you are and what time of year it is, which is exactly the constraint that makes identification tractable. A feeder camera advertising a large global label set is advertising the opposite of that discipline. It is not evidence the product is bad - but the big number is not the reassurance it appears to be, and it may be pointing the other way.

Birdbuddy publishes no species count at all, in a category where the number is the headline feature. Two readings are available and published material does not settle between them.

The generous one: the company understands the number is meaningless for a backyard and declines to compete on it - consistent with the fact that it is also the only maker publishing a sensor dimension, which is a harder and more useful number. The cynical one: its identification is weaker and silence is strategic.

We cannot determine which, and we are not going to imply we can. What we will say is that publishing a difficult, checkable specification while declining to publish an easy, unfalsifiable one is a pattern that generally belongs to companies expecting to be checked. The full review is here.

1. Ignore the species count when comparing feeders. It does not discriminate. Two products claiming 6,000 and 10,000 are making the same promise about your yard.
2. Do not pay extra for a bigger number. You are paying for coverage of birds that live on other continents.
3. Treat confident identifications of rarities with suspicion. If the app names something implausible for your region, it is far more likely to be wrong than you are to have a vagrant at your feeder.
4. Be more skeptical with hummingbirds. Similar female species are hard for expert human birders using field marks a fixed camera often cannot see - covered in the hummingbird hub.
5. Cross-check anything interesting with Merlin. It is free, it uses your location and date, and it is made by the Cornell Lab of Ornithology.

None of this means the feature is worthless - it is the reason most people buy these, and it works. Its real value is not identifying rarities; it is telling a beginner that the small brown bird is a house finch rather than a sparrow, reliably, thirty times a week, until they stop needing to be told. That is a good product doing a useful thing.

It is simply not a thing the difference between 6,000 and 10,000 claimed species has any bearing on. Choose your feeder on optics, power, Wi-Fi band and three-year cost instead - those differences are real.