aigrow-gg/ai-visibility-index

AI Visibility Index, October 2026 edition

Which businesses ChatGPT, Claude, Gemini and Grok name when a buyer asks them for a recommendation. The same buyer questions were asked about 44 kinds of business in 108 places across 20 countries, in the language of each place, and every name was counted.

• Publisher: AIGROW, https://aigrow.gg
• Edition: 2026-10, published on 6 October 2026. A new edition is published every month.
• Licence: Creative Commons Attribution 4.0 International (CC BY 4.0)
• Browse every ranking: https://aigrow.gg/index
• Full method: https://aigrow.gg/index/method
• Findings: https://aigrow.gg/research/ai-visibility-report-2026
• Same file on Zenodo (DOI 10.5281/zenodo.23188442) and Kaggle: https://www.kaggle.com/datasets/aigrow/ai-visibility-index
• Earlier editions: ai-visibility-index-2026-09.csv in this repository (config 2026-09), with the same columns

What was asked

• Kinds of business. 25 local trades (dentists, plumbers, lawyers and so on) asked about a city, 10 software and online service categories (CRM, website builders and so on) asked about a country, and, new in this edition, 9 consumer brand categories (banks, mobile carriers, supermarkets and so on) asked about a country.
• Places. 88 cities and 20 countries: Australia, Austria, Belgium, Brazil, Canada, Denmark, France, Germany, Ireland, Italy, Mexico, the Netherlands, Norway, Poland, Portugal, Spain, Sweden, Switzerland, the United Kingdom and the United States.
• Pages. 2,580 pages, each one kind of business in one place.
• Questions. Four buyer questions per page, written in the language of the place rather than translated word for word: English, German, French, Italian, Spanish, Portuguese, Dutch, Swedish, Danish, Norwegian and Polish. Each ranking page on aigrow.gg prints its four questions.
• Runs. Each question was put to each assistant with no web search, no location and no chat history, using the system prompt of the AIGROW free scan.
• Assistants. ChatGPT and Gemini on all 2,580 pages, three times each. Claude on all 2,580 pages: once on 1,755 of them and three times on the other 825. Where it was asked once, its column rests on 4 answers instead of 12. Grok on 104 pages only (1,248 calls); on the other pages its column is 0 because it was not asked.
• Web search. The first question was also asked once with web search switched on: ChatGPT on 2,580 pages and Claude on 680 pages (3,260 calls). Those answers fill web_share and never count toward share.
• Volume. 80,082 calls without web search, 80,059 answered.

Files

ai-visibility-index-2026-10.csv: 38,191 rows, one for each business named in two or more answers on a page. Names given only once are counted in the page totals but not listed. UTF-8, comma separated, header row.

Columns

Column	Meaning
edition	Edition month, YYYY-MM
category	Kind of business, as a slug (dentists, crm, banks)
place	City or country, as a slug (milan, spain)
country	ISO 3166-1 alpha-2 code of the place
indexable	yes when the page met the three ranking conditions: at least 24 answers, fewer than 20% failed calls, at least five businesses named in two or more answers. 2,580 of 2,580 pages did
business	The name as the assistants wrote it, after spellings of one business were merged (see below)
mentions	Answers on the page that named the business. A business named twice in one answer counts once
answered_probes	Answers on the page without web search, across all assistants
share	mentions divided by answered_probes
ci95_low, ci95_high	Wilson 95% confidence interval of share
share_chatgpt, share_claude, share_gemini, share_grok	Share within that assistant's own answers on the page
questions_named	How many of the four questions produced the name at least once
avg_rank	Average position in the answers that named it, 1 being first
web_share	Share in the answers with web search switched on, on the same page. Filled on every row in this edition

An answer that names no business still counts as an answer, so it lowers every share on its page.

How names are merged

Names are compared with accents, capitals, legal suffixes, a leading "the" and notes in brackets removed. A language model then groups the spellings on each page that are one business, such as the branches of a chain, and is told to keep names apart when unsure. A kept list of manual overrides wins over the model. Nothing fuzzier is applied: folding two real businesses into one would be a worse error than listing one business twice.

Figures for the whole edition

Averages over the 2,580 ranked pages.

• On average 55.7% of the names an assistant gave came back when the same question was asked again.
• On average two assistants agreed on 20.6% of the names they gave for the same page.
• The most named business on a page appeared, on average, in 59.0% of that page's answers.

Limits

• It is not a rating of the businesses, a local search result or a forecast. It records what these models said in October 2026 to questions asked without a location or a chat history. A real conversation can go differently, and the next model version will.
• Names are printed as the assistants wrote them. Nobody checked that a business exists, is open or is any good, and an assistant can name one that does not exist.
• With a few dozen answers per page the intervals are wide. Two businesses whose intervals overlap should be read as level.
• Editions can differ in coverage. Compare a page across months only where the same assistants were asked as often.

Corrections

If a name is wrong, is not a business, or is yours and you want it removed, write through https://aigrow.gg/contact?topic=index. Corrections apply from the next edition.

How to cite

AIGROW (2026). AI Visibility Index, October 2026 edition [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23188442

To cite every edition, use the concept DOI, which always resolves to the latest one: https://doi.org/10.5281/zenodo.23082266

@dataset{aigrow_ai_visibility_index_2026_10,
  author    = {{AIGROW}},
  title     = {AI Visibility Index, October 2026 edition},
  year      = {2026},
  publisher = {Zenodo},
  version   = {2026-10},
  doi       = {10.5281/zenodo.23188442},
  url       = {https://aigrow.gg/index}
}

Reuse under CC BY 4.0 with a link to https://aigrow.gg/index or to the ranking page the data came from.