Monsoon Research Lab

Monsoon Research Lab

Texas AI Search Visibility Index
Lubbock Pilot · September 2026

A clearer picture of local discovery.

Who appears when people ask Google and AI about local businesses? We pair search queries with AI questions, repeat them and publish what the recorded responses show.

Available research / September 2026

Lubbock: one city, six discovery surfaces

6,000 selected usable responses. Twenty business verticals. Five paired questions per vertical, repeated ten times on six configurations: Google Organic, Google Maps, ChatGPT Search, Gemini, Claude and Perplexity.

Across 20 Lubbock industries

Showing up on Google does not guarantee regular appearances in AI answers.

139 of 244 tracked business website groups, about 57%, appeared in Google's top ten at least once but were named in fewer than one in ten sampled answers on at least one AI platform.

The qualifying groups span all 20 industries studied. This is a finding across the full tracked cohort, not a single-business example.

This measures low name detection, not complete absence from AI. Website groups can include multiple locations or national brands. It is not a percentage of every business in Lubbock.

How we calculated the Google-to-AI gap

We included every tracked business website with at least one top-ten Google appearance in the pilot. A shared domain counts once. For each, we checked the five paired questions and ten repetitions on ChatGPT Search, Gemini, Claude and Perplexity in the industries where that website appeared on Google.

The threshold is fewer than 5 of 50 answers per platform. Eight website groups appear in two Google-visible industries, so their 100 answers are pooled and the equivalent threshold is fewer than 10 of 100. A single Google appearance qualifies a website for this comparison; it does not mean that website consistently ranks highly.

We reviewed shortened names, legal suffixes, punctuation, acronyms and scraped name formatting. Corrections reduced the initial 168 qualifying groups to 141. Two unresolved groups are withheld from the headline count, while remaining in the denominator. Known uncertain name matches are treated as possible detections before a group qualifies.

Website groups with fewer than one in ten detected-name answers
AI surfaceQualifying groups / 244
ChatGPT Search45 / 244
Gemini84 / 244
Claude72 / 244
Perplexity32 / 244

The platform counts overlap. A website group qualifying on multiple platforms counts only once in the 139. This does not measure recommendations, citations, customer traffic, business quality or algorithmic causes. An own-website link without a name is not counted as a name detection.

Download the calculation, cohort and observation IDs (JSON)

Exploratory follow-up completed September 11, 2026, using unchanged study v2 evidence. The below-10% threshold was chosen after collection. Repeated answers are not independent consumer samples. Automated checks and builder review are not independent human review; unknown aliases may still affect counts.

Original tables are unchanged; the original article and brief remain archived. The earlier conservative zero-detection calculation uses a different definition. It should not be compared as a change over time.

One business, different search surfaces

Gorilla Law Firm: 50 observations per surface

Google Organic50/50
ChatGPT Search1/50
Claude40/50
Perplexity20/50

Google: exact domain in top 10. AI: specified name in answer prose. Different measures, not recommendations.

This is a descriptive study of deliberately selected questions, not a statewide sample or a census of every local business. No overall business score or recommendation league table is released.

Why We Built It

Local discovery no longer happens through one interface. A business may appear prominently on Google yet rarely be named in AI-generated answers, or vice versa. Monsoon Research Lab measures those differences so businesses, journalists and residents can see what each system actually returns.

Learn how Monsoon approaches AI search optimization.

Research that can be checked

Every finding points to its evidence. Exact prompts, model labels where supplied, collection dates and downloadable counts accompany each edition. Original collection evidence is retained privately; public materials exclude full third-party answers and internal review records.

Repeat editions are planned every three to six months. New cities and statewide conclusions require new, explicitly defined collections. There are no statewide results in this edition.