Do AI assistants know American local directories, and the 2,866 citations that answer it

Partly, and the part they know is narrower than a local marketing plan assumes. We counted it. Across 64 blind readings of ChatGPT and Gemini on American buying questions on 22 October 2026, the two engines produced 2,866 citation URLs across 1,143 distinct hosts. Two agency directories appeared near the top of that list, clutch.co in 11 readings and goodfirms.co in 6. One local review platform appeared further down. The consumer directories that dominate American local marketing conversations did not appear in the top tier at all. So a directory listing is worth real money when the directory is one the engines read, and close to nothing in this queue when it is not, and the only way to know which is which is to count.

What was measured

The bank was 198 American buying questions in 33 batches, put to the ChatGPT and Gemini product surfaces through a data provider. The United States was named inside the text of every question, because the provider accepts gl, hl and location parameters and ignores all three, so the prompt is the only market lock that holds. Batch 33 names us in the question and is therefore a ceiling measurement, excluded from every figure here. That leaves 64 blind readings, 32 from ChatGPT and 32 from Gemini. ChatGPT ran a live web search on all 32 of its readings and Gemini on 16 of its 32.

The hosts, counted, in order

Host	Readings citing it, of 64	Kind
developers.google.com	17	Platform documentation
support.google.com	15	Platform help
searchengineland.com	13	Trade press
blog.google	11	Platform announcements
clutch.co	11	Agency directory
semrush.com	10	Vendor property
ftc.gov	9	Regulator
blog.hubspot.com	9	Vendor property
business.google.com	7	Platform product
developers.google.cn	7	Platform documentation
searchenginejournal.com	7	Trade press
brightlocal.com	7	Vendor property
consumer.ftc.gov	6	Regulator
goodfirms.co	6	Agency directory

One host in the raw data, ccn.web.tr at 8 readings, is the data provider's own fetch infrastructure rather than a publisher. It is excluded from this table, it is never presented as a source, and it is never merged into another host's count. A reader auditing this table against a raw export will find it there and should treat it the same way.

What the count establishes, and what it does not

• It establishes that two agency directories are read. clutch.co and goodfirms.co together appeared in 17 readings of 64. For a supplier of marketing services, a complete and accurate profile on those two is a cheap input into answers buyers see.
• It establishes that platform documentation dominates. Google's own four properties in the table appeared in 50 readings between them, which is more than any other category.
• It establishes that the regulator is read. ftc.gov and consumer.ftc.gov appeared in 15 readings between them, which makes accuracy about FTC positions a visibility asset as well as a safety one.
• It does not establish that other directories are ignored. This bank asked marketing and supplier questions. A bank about restaurants, trades or retail would plausibly surface consumer review platforms that this one did not, and that is a different measurement we have not taken.

The city level counterpart, reported separately

A separate reading on 15 and 16 October 2026 asked six questions per city across New York, Los Angeles, Chicago and Houston, two engines, 12 readings each. Forty three of those 48 readings returned a named source list. The leading source differed in every city: localseo.nyc in New York with 3 citations, agencies.semrush.com in Los Angeles with 4 and in Chicago with 3, and houdigitalmarketing.com in Houston with 3. Two cities answered with a city specific domain and two with a national platform's agency directory.

These are two separate measurements on two separate banks and two separate dates, and the figures must not be combined into a single table or a single percentage. The national pass says which hosts the engines reach for across the country. The city pass says that the leading source changes per metro. Together they say that directory value is both host specific and city specific, which is a stronger conclusion than either reading alone.

What a listing is therefore worth

• On a directory the engines read, a listing is worth an afternoon per quarter. Complete profile, service description and price position matching your own site, and a recheck every three months.
• On a directory the engines do not read, treat it as a referral channel on its own merits. It may still send business. It is not doing work in this queue.
• On a national directory with city pages, the listing can be a local asset. Two of our four metros answered with exactly that.
• Verify what each directory says about you, not just that you are listed. A profile with a stale service area or an old phone number is worse than an absent one, because it gets quoted.

How to count your own

• Pick twelve questions your American buyers would ask, with the United States or the city named in each.
• Put each to both engines and save the full source list for every answer.
• Tally the hosts across all the answers. Rank by the number of readings that cited them rather than by raw URL count, because one answer can cite a host several times.
• Separate the infrastructure entries from the publishers before you draw a conclusion.
• Repeat per city if you sell in more than one metro, because on our evidence no two of four cities shared a leading source.

The New York reading is the clearest case of a city answering with its own sources and is written up in two of the four winning sources are the city's own. If your category is local service and the directory question is the main one, the review and citation habits of that market are counted in checked against the comparison table BrightLocal publishes.

Questions about directories and American assistant answers

Which directories do the engines actually read?

In this bank, clutch.co in 11 of 64 readings and goodfirms.co in 6. A different bank of questions, in a consumer category, would plausibly surface different platforms, and that would be a separate measurement.

Should I pay for a premium directory listing?

Judge it on referrals first. In this queue the value comes from the profile being complete and accurate on a host the engines read, which is usually available without the upgrade.

Why does Google documentation dominate the citations?

Because it is authoritative, specific, dated and freely readable, which is the combination an engine can quote with confidence. That is a lesson about the shape of a document rather than about Google.

Can I use these counts as a national benchmark?

Use them as a method and a starting hypothesis. The denominator is 64 readings on one bank on one date, which is a baseline rather than a national statistic.

What about the host you excluded?

ccn.web.tr is the data provider's own fetch infrastructure rather than a publisher, and it appeared in 8 readings. Presenting it as a source or merging it into another host would overstate a publisher that does not exist.

Counting the citation list rather than guessing which listings matter

The whole directory question resolves to a count, and the count is specific to the questions your buyers ask and the market they ask in. AI Knows Us reads ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews for those questions, records every cited host per answer, reports each engine separately with its denominator, and keeps the raw answer text so a tally can be audited and infrastructure entries separated out. In the United States that is the Growth plan at 299 dollars a month with 100 buyer questions watched and 1,600 credits, read on aiknowsus.com/pricing and aiknowsus.com/llms.txt on 22 October 2026.