There is no list of directories to get on: ChatGPT cited 1,092 different domains across 511 local-business answers, and 97% of them appeared in only one vertical

What we measured, and how

Our production corpus is 173 completed audits — 1,024 answers (64 med spas, 42 personal-injury law firms, 40 dentists, 27 home-services businesses; 42 US metros; audits run 2026-06-12 to 2026-07-14). Synthetic QA rows are excluded by an exact-name filter rather than a regex over the vertical field — a discipline we adopted after publishing a false finding in July and correcting it publicly. For this cut we took every completed answer, extracted every raw URL from the answer text, and collapsed each to its registrable domain. The counting unit is answers-citing-a-domain, not links: an answer counts once for a domain no matter how many times it links it, because five links to the same directory page is one source, not five.

One limit belongs at the top, not in the footnotes: this is a measurement of ChatGPT only. All 511 of our ChatGPT answers carried at least one link — 5,081 links in total, a mean of 9.9 links and 5.5 distinct domains per answer. Our 513 Perplexity answers carried 13 links between them, because we capture the model's text output through the API rather than the citation panel its own interface renders. That is a limitation of our instrument and says nothing about what Perplexity reads. Every number below is ChatGPT's behaviour.

Finding 1: 1,092 domains, and 61.6% of them cited exactly once

Across the 511 answers ChatGPT cited 1,092 distinct domains. 673 of them — 61.6% — appear in exactly one answer in the entire corpus. The concentration at the top is real but modest: the ten most-cited domains together account for 20.8% of all answer-domain pairs, and the top 25 for 30.3%. There is no small set of gatekeepers here. There is a steep head, a very long tail, and a lot of one-off local blogs, city magazines and forum threads that happened to be the best available source for one question on one day.

The single most-cited domain in the corpus is not a directory a business can "get listed on" in the usual sense — birdeye.com is a reputation-management platform, and the pages ChatGPT cites are the review pages it hosts for businesses that already pay for it. The second is the BBB. Both are, in effect, byproducts of software a business bought for another reason entirely.

Finding 2: only 12 of 1,092 domains span all four verticals

We tagged each cited domain with the verticals whose answers cited it. 1,057 of 1,092 domains (96.8%) appeared in exactly one vertical. Sixteen appeared in two, seven in three, and twelve in all four: bbb.org, birdeye.com, bestprosintown.com, threebestrated.com, chamberofcommerce.com, trustindex.io, trustanalytica.org, trustburn.com, reddit.com, mapquest.com, indeed.com, lantern.llc. Comparing each vertical's top 25 sources against every other vertical's top 25, the overlap is 4 to 7 domains out of 25.

Finding 3: the surface is small inside a vertical and unusable across them

We ran a greedy cover — repeatedly picking the domain that appears in the most answers not yet covered — to answer the practical question: if you could be present on N domains, how many of the engine's answers would you touch? Corpus-wide, four domains reach 52.6% of all 511 answers (birdeye.com, bestprosintown.com, medspascout.com, bbb.org) and ten reach 69.3%. But three of those four are vertical-specific, so the corpus-wide answer is the wrong one. Run the same cover inside each vertical and the practical list changes completely:

Note what these lists are not. Nine of the twelve names above are niche directories most agencies have never audited a client against — medspascout.com, auravenu.com, medicalspalocator.com, lawleaderboard.com, hvacservice.io, toprateddentist.com, dentalcountry.com, counselindex.com, injuryattorney.reviews. Yelp, Google Business Profile and Angi are not the story in this corpus; a set of small, vertical-specific aggregators is. And 11.5% of all 5,081 links pointed at the recommended business's own website — consistent with the 28% of answers that cite the business's own site at all, measured on the same corpus.

What this means if you sell AI visibility

• "Get listed on the AI directories" is not a deliverable until you name the vertical. The overlap between two verticals' top-25 source lists is 4–7 domains out of 25. A citation-surface audit built for a dental client is roughly 80% wrong for a med-spa client in the same city.
• The reachable surface per vertical is small — three to five domains for 49%–79% of answers. That is a finite, checkable work item for an agency, and it is a different work item per client category.
• Two of the top four sources are SaaS byproducts, not directories. Birdeye review pages and BBB profiles led the corpus. Whether a client's review platform publishes an indexable profile page turns out to be a citation-surface decision.
• One audit will show you mostly noise. 61.6% of cited domains appear in exactly one answer. A source list from a single run is not the citation surface; the repeated, multi-business version is.

Limits of this cut, stated up front

• ChatGPT only. Our Perplexity answers carry 13 links across 513 answers because of how we capture them — an instrument limit, not a finding. Gemini and Claude are in the paid audit but not in this cut.
• Four verticals, 42 metros, one month. Med spas are 191 of the 511 answers, so the med-spa directory names are the best-evidenced and the home-services list (81 answers) the weakest.
• A domain is not a listing. Being cited means the engine linked a page on that domain, sometimes a review page a business does not control and cannot buy.
• We sell audits. A finding that says "the citation surface is vertical-specific and needs measuring" is convenient for us. Read the method and the counts; discount the conclusion.

The takeaway

There is no directory list to get on. There is a different short list per vertical, sitting inside a tail of a thousand domains that each appear once — and in our corpus the top of every one of those lists is a page the business itself does not publish.

The free 60-second check runs a business's own buyer questions and shows what the engines cite when they answer. Agencies mapping a citation surface across a book of clients can run five at once with the $249 Agency 5-pack — 25 buyer questions × 4 engines × 3 samples per business, white-labeled, which is the per-client version of the corpus cut above. If your domain is one of the 1,092 and you want the exact answers that cited it, ask us and we will send them.