September 2026 / Open aggregate research materials

Follow the finding back to the evidence.

Read the research article online or download a PDF. The observation facts let you reproduce published totals without access to provider accounts.

Research articlePDF: updated local leads, supporting research and limitations Media briefPDF: two pages with key findings and contact details Study & observation factsJSON: the selected 6,000 observations and evidence ledger Observed visibility tablesJSON: all original name/domain counts, aliases and supporting observation IDs Finding countsCSV: numerators, denominators and descriptive overlap Edition change logJSON: prior and revised counts Public page inventoryJSON: every public HTML route and indexing decision Release checksumsSHA-256 hashes for the packaged release Updated local story factsJSON: reviewed counts and sector calculations Editorial revisionChange log, original document links and current document checksums Count verifierNode.js: recompute the downloaded study without network access Reproduction functionsDependency-free JavaScript used by the verifier

Markdown source files

Optional editable text for research and technical workflows. The HTML pages are the primary reading and citation versions.

September 11 editorial follow-up

Evidence behind the three local lead stories

About 57% (139 of 244) of tracked business website groups that appeared in Google's sampled top ten were rarely detected on at least one AI platform. This is low detection in sampled answers, not complete absence from AI.

Rare means fewer than 10% of sampled answers on at least one of four AI platforms: fewer than 5/50 for a single-vertical group, or 10/100 when two verticals are pooled. It does not mean absent.

The 244 tracked business website groups each appeared at least once in a sampled Google top ten. They are not a census of Lubbock businesses or 244 consistently high-ranking businesses. Names and domains are different measurements; neither is a recommendation or business-quality score.

The low-detection threshold was selected after collection. This is exploratory analysis, not a preregistered endpoint.
Reviewed name variants are included. Known uncertain matches count as possible detections; two identity-uncertain groups remain in the denominator but cannot qualify. Unknown aliases can still affect counts.
Sector counts use each vertical separately and should not be summed into a citywide total. Questions differ by vertical; ten repetitions are not ten independent consumer samples.

Download the updated story facts and sector calculations · Reviewed cohort and supporting observation IDs

The original study.json verifier covers the original evidence ledger. The later reviewed-name projection and editorial facts are separate, versioned follow-ups, not silently substituted values in that ledger.

Named assistant contrasts

CoNetrix was named in 50/50 sampled ChatGPT answers and 0/50 Claude answers. Gorilla Law Firm showed the opposite contrast: 1/50 on ChatGPT and 40/50 on Claude. These are examples of platform differences, not a ranking of the assistants.

Reviewed name detections / 50 sampled answers
Tracked website group	ChatGPT Search	Claude
conetrix.com	50/50	0/50
gorillalawfirm.com	1/50	40/50

These two selected examples have no flagged uncertain name matches in the reviewed counts. They do not establish typical performance for all businesses. See the original CoNetrix evidence and Gorilla Law Firm evidence.

Restaurant and dentist website groups

The pattern also appears in everyday local services: 14 of 21 tracked Google-visible restaurant website groups and 16 of 25 dentist groups were detected fewer than 5/50 times on at least one AI platform, even when known uncertain matches count as possible detections.

Tracked Google-visible website groups with fewer than 5/50 detections on at least one AI platform
Sampled vertical	Qualifying groups	Google-visible groups
Restaurants	14	21
Dentists	16	25

Each group must have at least one sampled Google top-ten appearance in that vertical. We use only that vertical's 50 answers per platform, not the pooled cross-vertical count. Shared domains count once within a vertical. These deliberately selected examples are not a census or a ranking of which industry performs worst.

Reproduce the counts

Download study.json, verify.mjs and reproduce-pilot.mjs into one folder. With Node.js 22 or later, run node verify.mjs. No dependency installation, account or network request is needed.

The verifier independently recomputes every numerator, denominator and overlap from the public projection. It does not independently establish the correctness of the name-match labels or semantic meaning of an answer.

Explore the visibility tables Inspect all observations Original edition & history

What is deliberately not public

Full third-party answers, raw provider JSON, contact details from provider records, credentials, provider task identifiers and internal review passages are excluded. The published name/domain tables describe observed matches; they do not certify branches or rank business quality.

Credit Monsoon Research Lab, Texas AI Search Visibility Index: Lubbock Edition, September 2026, descriptive study v2. Link to this edition and preserve the scope, denominator and limitations. Aggregate tables, original charts and methods may be reused with attribution. Source answers and third-party content are not licensed for redistribution.