NameBeam · September 2026 corpus · dated receipts

There are two phone books now.

AI assistants and Google name different local businesses. We counted for 28 days across 22 local US markets. 84% of named businesses appear in only one universe.

AS ABOVE, SO BELOW. 3 6 9.
84%

live in a single universe

Of 272 named businesses in the 22 local markets (Perplexity's per-market top ten counted from its saved answer text, and Google's per-market most-cited list from the published reports, non-business phrases removed), 228 (83.8%) appear in only one of the two lists and 44 appear in both (counted from the saved answer text, Sept 1 to 28, 2026). Some of the most AI-dominant businesses in the corpus sit outside Google's top-10 entirely: Arrington Roofing in Dallas, named by Perplexity on 24 days, never appears in Google's most-cited list for Dallas. Others live in both books: Google also cited the sites of All Star Movers (Pleasanton, 10 days), Reliable Heating & Air (Atlanta, 10 days) and Gates Enterprises (Denver, its site gatesroof.com, 6 days).

Perplexity's side of this comparison is counted from the saved answer text (full business name, whole words, spelling variants merged; ties at tenth place broken by name). Google's side is read from the published Most cited tables, because Google results are not answer text, and a business that Google lists under a shorter name counts as unmatched (counted from the saved answer text, Sept 1 to 28, 2026).

So what: every SEO agency on earth is still optimizing for one phone book. The second phone book has different winners, and nobody is reading it. AI answers are 3 to 5 names, not 10 blue links, second place in AI is tenth place in Google.

Method note: the comparison uses published top-10 lists per market with case-insensitive name normalization, and even that understates the mess. Perplexity's own Denver top-10 lists both "Cenco Roofing" and "CENCO Roofing" (same business, counted twice), while Google sometimes cites domains (e.g. allstarmoverscal.com) rather than business names. Normalization debt lives inside our own lists too. The directional finding, two largely disjoint winner sets, is robust to that. Recomputed Oct 12, 2026: Perplexity's per-market top ten counted from its saved answer text, Google's list read from the published Most cited tables, non-business phrases removed: 272 distinct named businesses in the 22 local markets, 44 in both sets (228 of 272, 83.8% in one universe; counted from the saved answer text, Sept 1 to 28, 2026). Earlier cuts (282, 276 and 248 businesses) used the extracted name lists and are superseded. A few businesses appear on both lists under variant names (trade name vs. legal entity, or a shorter name in Google's table); the overlap count matches full names.
Corrected 2026-10-12: counts recomputed from the saved answer text. The versions published 2026-10-01 (276 businesses, 26 in both sets, 90.6% in one universe) and 2026-10-06 (248, 25, 89.9%) are superseded.
21.2

businesses per niche-market: the real engine ranking

Our own September recap ranked OpenAI (880) above Gemini (870) on raw totals. That was a coverage artifact, Pleasanton tracks 8 niches, Los Angeles tracks 1. Normalized per niche-market, the ranking reverses:

Gemini 21.2 · OpenAI 18.7 · Perplexity 17.8 · Google SERP 13.7 · Anthropic 1.5

So what: raw totals mislead; per niche-market counts depend on the method: counting every name, Gemini is highest; counting only names that are businesses, it is not (our recount of 2026-10-04, second check in progress). Gemini named a business on 26 September dates and on none after September 26; every Gemini run from September 27 to October 11 was quota-blocked or failed, so this describes September, not Gemini today. We are correcting our own published headline below, because the standard applies to us first.

A correction, published by us

Our September 2026 recap page ranked the engines OpenAI > Gemini on summed namings (880 vs 870). Re-analysis shows the gap is driven by uneven niche coverage across markets, not engine behavior. Which engine names the most per niche-market depends on the method: counting every name, Gemini is highest; counting only names that are businesses, it is not (our recount of 2026-10-04, second check in progress). The recap's engine section has been updated with the normalized figures. Every claim owes the future a receipt, including ours.

76

of Claude's 114 namings are nationwide digital products

Claude's 114 total namings include 76 nationwide digital products (AI tools, courses), verified from the US-nationwide data. The remaining 38 are short-term-rental listings and traditional local services; the published per-market JSONs don't separate that split, so we print what verifies. In September 1 to 28, 2026, Claude named no business on any answer day in 7 of the 9 local markets where it was asked about roofers, HVAC techs or lawyers. Denver and Liberty, MO were the exceptions.

So what: this isn't blindness, it's architecture. Claude answers from training memory, travel, software, and has no local-service knowledge to draw on. "AI visibility" is not one thing; it depends on how each engine knows what it knows.

35%

of top-ten slots changed between the first and second half of the month

In the 24 report markets, 68 of the 196 businesses in Perplexity's second-half top ten (Sept 15 to 28) were not in its first-half top ten (Sept 1 to 14). 67 of the 535 businesses Perplexity named in these markets held steady, meaning named on 15 or more days and on at least 70 percent of the days between their first and last naming. Late September saw fresh entrants: six Columbus tanning salons were first named in the final week (Sept 22 to 28), each on one or two days (counted from the saved answer text, Sept 1 to 28, 2026).

The 160 are Perplexity top-ten businesses with no full-name match in Google's Most cited tables. A business that Google lists under a shorter name, such as Reliable Air for Reliable Heating & Air, counts as unmatched, and Google's published tables are truncated, so 160 is a ceiling, not a floor. The table shows the 25 most-named of the 160 (counted from the saved answer text, Sept 1 to 28, 2026).

So what: AI visibility rotates weekly. A monthly check-up may be the wrong cadence, by the time a monthly report arrives, the winners have changed. This one is an open product question, stated as such.

Where Gemini did not answer

Gemini posts a high per niche-market count on all names, then, in September 1 to 28, 2026, Gemini returned no answered run for the Los Angeles AI-agency or Maryland consultancy questions, and named one San Francisco agency on its one answered day. OpenAI named an agency on 19 of 28 Los Angeles answer days, 4 of 28 Maryland and 11 of 28 San Francisco. We don't know why yet. The query-level data to answer it lives in the raw corpus, and we're going after it.

Gemini returned no answered run for these questions in the window, so a zero here means no answer, not a blackout.

The other direction: 160 businesses AI names that Google's tables do not list

The gap runs both ways. 160 of 208 businesses in Perplexity’s per-market top ten (24 report markets, counted from the saved answer text, Sept 1 to 28, 2026) have no match in Google’s “Most cited” tables, including a most-named roofer (Arrington Roofing, Dallas, 24 of the 25 Perplexity answer days). AI’s darlings, invisible to the old phone book. Recomputed 2026-10-12 against the public Google tables; the Invisible 100 carries the reverse list.

The earlier figure of 68 of 218 used a turnover definition that cannot be reproduced from the page text, so this is a new, stated definition: a top-ten slot in the second half of the month held by a business that was not in the first-half top ten. The share and the steady count are not comparable with the earlier figures.
BusinessMarketNichePerplexity days
Doug BuenzPleasanton, CAreal estate agents25
Arrington RoofingDallas, TXresidential roofing24
Gates EnterprisesDenver, COresidential roofing24
Big OPleasanton, CAtire shops24
J. Michael HostermanPleasanton, CApersonal injury law24
SpeeDee Oil ChangePleasanton, CAoil change shops24
Tire ChoicePleasanton, CAoil change shops24
NuLight ConsultingKansas City, MOhealth insurance brokerage23
Anderson CollaborativeMarylandmarketing and brand strategy consultancy23
Howard AirPhoenix, AZhvac repair and replacement23
Right at HomePleasanton, CAhome care agencies23
Senior HelpersPleasanton, CAhome care agencies23
John Foy & AssociatesAtlanta, GApersonal injury attorneys22
Botanical TansCharlotte, NCtanning salons22
Salvi, Schostok & PritchardChicago, ILpersonal injury attorneys22
Grossman Law OfficesDallas, TXpersonal injury attorneys22
Royal Air HoustonHouston, TXhvac repair and replacement22
Tucker's 1 Handyman ServiceJoplin, MOresidential remodeling and repair22
FD Insurance AgencyKansas City, MOhealth insurance brokerage22
Eagle Moving & Storage, IncPleasanton, CAmoving companies22
ATA Martial Arts St. PetersSt. Peters, MOmartial arts for kids22
Sun Studio TanColumbus, OHtanning salons21
Thrive Internet Marketing AgencyLos Angeles, CAAI marketing agency21
Van Sant LawAtlanta, GApersonal injury attorneys20
Smith & HasslerHouston, TXpersonal injury attorneys20
Perplexity: 208 top-ten rows across 24 report markets, counted from the saved answer text (Sept 1 to 28, 2026). Google: 130 rows read from the published Most cited tables, which are truncated. Names are matched in full, and 160 is the count with no match.

Methodology

Runs on September 1 to 28, 2026. Not every engine answered with names every day: Perplexity named businesses on 25 September dates, OpenAI on 28, Gemini on 26 and Anthropic on 28 (text rule, Sept 1 to 28). 22 local US markets, 33 niches. Four AI assistants (ChatGPT, Claude, Perplexity, Gemini) plus Google SERP as baseline. Same buyer-intent questions per market, every answer dated. Counts are distinct businesses named per engine per market from the published per-market JSONs; per-business Perplexity day counts from the saved answer text and Google day counts from the published top-10 tables. Engine coverage figures are distinct-businesses-per-market from engine_coverage. Turnover figures compare the first and second half of the month in the saved answer text. Everything here is computed from the September corpus; the 24 per-market reports are public and every figure above can be checked against them. Headline figures are rounded, and the unrounded values are printed alongside.

All 24 per-market reports: markets.namebeam.ai/reports/

Run the free check Get the Invisible 100 See the leaderboard