SMB AI visibility study, October 2026 - the data Quaryan, https://quaryan.com/blog/ai-visibility-study-october-2026/ How often ChatGPT, Gemini, Claude and Perplexity name small local businesses in six Swiss cities, checked against a fixed panel. The report (PDF) beside these files says how the study works. Comma-separated, UTF-8, one header row. Shares are fractions (0.091 = 9.1%). Cities and trades by their study ids (zurich, basel, bern, geneva, lausanne, lugano; bakery, dentist, hair-salon, plumber, veterinarian, auto-repair); engines CHATGPT, GEMINI, CLAUDE, PERPLEXITY. No file holds a business's Google reviews or rating: Google's terms let those be kept only for a while, so only results computed from them are published. The panel's list of businesses contains information from OpenStreetMap, (c) OpenStreetMap contributors, and is made available under the Open Database License (ODbL) 1.0. Cite as: Quaryan (quaryan.com), SMB AI visibility study, October 2026, https://quaryan.com/blog/ai-visibility-study-october-2026/ If you use or quote the data, please link to the study: https://quaryan.com/blog/ai-visibility-study-october-2026/ smb-ai-visibility-2026-10-engines.csv 4 rows. The headline per engine: answers, coverage (the share of the panel in its city and trade one answer names, averaged over each question's three askings, the questions and the businesses), coverage counting only names in the text or only cited sites, reach (the share named at least once), names per answer, the panel's share of all names, and concentration (HHI, top-3 share). Claude's coverage_cited is near zero because its answers cite almost no pages. Columns: engine, answered, failed, searched_share, coverage, coverage_in_text, coverage_cited, reach, mentions_per_answer, panel_share, hhi, top3_share smb-ai-visibility-2026-10-cells.csv 36 rows. Coverage per city and trade, per engine and the mean over the four. Columns: city, trade, panel_businesses, coverage_chatgpt, coverage_gemini, coverage_claude, coverage_perplexity, coverage_mean, reach_mean smb-ai-visibility-2026-10-intents.csv 16 rows. Coverage per engine and kind of question: RECOMMENDATION (who is the best), PROBLEM (a problem to solve), ATTRIBUTE (a specific need), PRICE (a fair price). Columns: engine, question, answered, coverage, coverage_in_text, coverage_cited, reach smb-ai-visibility-2026-10-panel-businesses.csv 276 rows. Every panel business: how it came into the panel (drawn at random from OpenStreetMap, an OpenStreetMap place whose website was looked up, or found by hand where OpenStreetMap ran out), its coverage per engine (the share of the answers about its city and trade that named it), how many engines named it at least once (0-4), and the best place it was named at. Contains information from OpenStreetMap, (c) OpenStreetMap contributors; this file is made available under the Open Database License (ODbL) 1.0, https://opendatacommons.org/licenses/odbl/1-0/. Columns: city, trade, name, website, in_panel_by, coverage_chatgpt, coverage_gemini, coverage_claude, coverage_perplexity, coverage_mean, engines_naming, best_rank smb-ai-visibility-2026-10-engines-per-business.csv 5 rows. Panel businesses by how many of the four engines named them at least once. Columns: engines_naming, businesses, share smb-ai-visibility-2026-10-chance.csv 36 rows. Coverage against chance per city and trade: the businesses one answer names (services, institutions and platforms left out), the market (the larger of the places OpenStreetMap has for the trade in the city and the different businesses the answers have named there in every run so far — both undercount), chance = names per answer / market, and coverage / chance. Columns: city, trade, names_per_answer, osm_places, named_ever, market, coverage, chance, ratio smb-ai-visibility-2026-10-named-by-kind.csv 100 rows. Every business the answers named, one per answer that named it, by kind: panel, other independent, chain, institution, service. Columns: engine, trade, kind, mentions smb-ai-visibility-2026-10-site-types.csv 28 rows. The pages the answers cite, by kind of site; for an engine whose answers cite almost nothing (Claude), the pages its searches returned (basis = retrieved). share_of_answers is the share of the engine's answers with at least one page of that kind. Columns: engine, basis, site_type, pages, share_of_pages, answers, share_of_answers smb-ai-visibility-2026-10-what-goes-with-being-named.csv 8 rows. What the named businesses have in common, within the same city and trade, for the panel and the OpenStreetMap independents not drawn into it together. rate_ratio: how many times as often the top half for the feature (a yes/no feature: with vs without) was named at least once (or in six answers or more) as the bottom half; p from 5,000 shuffles of the labels within each city and trade. odds_ratio_reviews_held_constant: conditional logistic regression with the number of Google reviews held constant (for log_reviews itself: per standard deviation, about four times as many reviews). Correlation in one month's answers, not cause. Columns: feature, businesses, named, rate_ratio_named, p, rate_ratio_named_6_answers, p_6_answers, odds_ratio_reviews_held_constant, p_reviews_held_constant smb-ai-visibility-2026-10-where-names-come-from.csv 4 rows. Each time an answer named a panel business: was its own website among the pages the engine cited (Claude: retrieved), another page that lists it, both, or neither. Shares of the mentions. Columns: question, mentions, own_site_only, own_site_and_other_page, other_page_only, no_page_read smb-ai-visibility-2026-10-engine-agreement.csv 7 rows. Panel businesses by the number of engines that named them: observed, and expected if each engine chose independently given the city and trade and the engine, then also the features named. Columns: model, engines_0, engines_1, engines_2, engines_3, engines_4 smb-ai-visibility-2026-10-tests.csv 370 rows. Every feature tested, in every sample (panel, random draw only, without institutions, not drawn, pooled) and for being named at least once (named) and in six answers or more (named6): shares or medians for named and others, the rate ratio (rr), the permutation p and the Benjamini-Hochberg q. Columns: sample, label, group, feature, outcome, cells, n, named, median_named, median_others, mean_pct_named, rr, p, share_named_with, share_others_with, rate_with, rate_without, q smb-ai-visibility-2026-10-tests-reviews-held-constant.csv 358 rows. Every feature again, in a conditional logistic regression with the number of Google reviews held constant: odds ratio, 95% interval, p and q. Columns: sample, outcome, feature, label, group, n, named, odds_ratio, lo, hi, p, per, reviews_or, q