What AI says when a buyer asks for a dog breeder: 735 answers, 27 states, kept word for word

The finding in one sentence

When a buyer asks AI where to find a puppy, the AI answers with registries, marketplaces and forums, not with breeders — and a small number of websites decide what it says.

The first run: 180 answers, 12 states

• 180 answers — 5 questions × 12 states × 3 engines, 60 per engine, collected on 2 September 2026. All 180 completed; none failed or was dropped.
• 3,884 citations across 919 domains, a mean of 21.6 sources per answer. Claude read the most before answering (34.4 sources per answer), then Perplexity (19.9), then ChatGPT (10.5).
• Ten domains carry 1,352 of the 3,884 citations — 34.8%. 161 of the 180 answers cite at least one of those ten.
• gooddog.com alone was cited 315 times, in 84 of the 180 answers — as many as the bottom 315 domains put together.
• 552 of the 919 domains appear in exactly one answer. The long tail is long and thin.
• The AKC was named in 146 to 148 of the 180 answers, depending on how strictly the name is matched — about four in five.
• 13 answers cited no sources at all — 12 from ChatGPT, 1 from Claude. They are kept and counted.

The ten most-cited websites, first run

Counted from the saved source lists. The first run asked about Golden Retrievers, Cavalier King Charles Spaniels and “any breed”, which is why two Golden and two Cavalier sites appear here.

Both runs together: 735 answers, 27 states, 17 breeds

• 735 answers: 180 from the first run and 555 from the extension, 245 per engine. All completed.
• 14,560 citations across 2,846 domains.
• The ten most-cited domains carry 4,648 of the 14,560 citations. gooddog.com alone: 1,671 citations in 445 of the 735 answers.
• 39 of the 735 answers came back with no source list — 37 from ChatGPT, 2 from Claude. Kept, not dropped.
• The extension added fourteen breeds — among them Bernedoodle, Cavapoo, German Shepherd, Goldendoodle, Labrador Retriever, Standard Poodle and Bernese Mountain Dog — and fifteen states, from Idaho to New York.

The ten most-cited websites, both runs

Correction, 8 October 2026. Our scanner saved each Perplexity source twice, once from each of the two source lists Perplexity returns, so every citation total on this page was too high. All totals are now recounted with each Perplexity source counted once. The “answers citing it” counts were not affected. Before the fix this page said 19,383 citations (now 14,560), 5,076 in the first run (now 3,884), and 571 for gooddog.com in the first run (now 315). bbb.org drops out of both top-ten tables, and Perplexity reads 19.9 sources per answer, not 39.7, so Claude reads the most.

What it means if you breed dogs

The AI is not reading breeder websites to decide who to recommend. It is reading a marketplace, a registry, a health database, a forum, and a few review sites, and it repeats what it finds there. If those pages say nothing legible about your program, you are not in the answer — however good your dogs are. That is the whole reason the Litter Launch spends its six weeks on the pages the AI actually quotes, and on making your own health-testing page something a machine can read and check.

One thing this page does not say: how many different kennels were named across all 735 answers. Nobody has counted it, so no figure is given. Counting the names inside one answer is easy; one Perplexity answer about Cavaliers in Texas names fourteen programs in a single sentence. Extrapolating that to the whole set is not something I will do.

My own kennel, run through the same questions

I did this as a breeder before I did it as a business. In the first run, my own program is named in 3 of the 180 answers, all of them Oregon. Six answers cite my pages — 10 citations across 8 URLs — in Oregon, California and Ohio. My Cavalier health-testing write-ups are being quoted to buyers 2,000 miles away while nothing recommends my kennel. The writing travels; the business does not. One cell of the grid, Golden Retrievers in Arizona, is written up answer by answer on its own page.

How it was measured

Three named engines, five fixed questions per state, every transcript saved with its source list, the empty runs kept, and nothing estimated. The full protocol — including why single-run numbers are not trusted and what the “40%” in most sales decks actually measured — is on the method page. The full write-up, with every count shown, is at Who decides which breeders an AI names? and the raw data is downloadable from the dataset page.

Questions about the study

How many answers are in the Breed Study?

735, in two runs: 180 answers across 12 states on 2 September 2026, and 555 across 27 states on 14 to 16 September 2026. Each run put the same fixed questions to Claude, ChatGPT and Perplexity, 245 answers per engine in total, and all 735 completed.

Which websites does AI rely on when asked about dog breeders?

A small group. In the first run, ten domains carried 1,352 of 3,884 citations and gooddog.com alone was cited 315 times in 84 of 180 answers. Across both runs the ten most-cited domains carried 4,648 of 14,560 citations, led by gooddog.com, akc.org, devotedtodog.com and ofa.org.

Does AI name individual breeders at all?

Sometimes, and the count varies by answer: one Perplexity answer about Cavaliers in Texas names fourteen programs in one sentence. Nobody has counted how many distinct kennels are named across all 735 answers, so this page does not give that number.

Can I see the raw data?

Yes. Every transcript and its source list is published as CSV on the dataset page, and the full write-up shows every count. Nothing on this page is estimated.

Was the study author’s own kennel in it?

Yes, and the result is published as found: named in 3 of the 180 first-run answers, all in Oregon, with six answers citing its pages, 10 citations across 8 URLs, in Oregon, California and Ohio.