GEO and AI search visibility statistics, 2026
Statistics on how AI answer engines name and cite B2B software vendors, drawn from published research. Broadcastwell provides AI search visibility services and publishes this research. The research sources and DOIs are linked below. Each statistic carries the available basis and any limits on verification. Engines measured: one engine held constant, Claude with live web search, for Volumes I and II; four engines, ChatGPT, Claude, Perplexity and Google AI Overviews, for Volume III. Last updated: 5 September 2026.
How often companies appear
860 AI answers were scored across 85 B2B software companies in 61 software categories. Ten buyer questions per category, asked of one engine held constant, Claude with live web search, in July 2026.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
27.5% was the mean per-company named rate. Mean of per-company rates across 85 companies and 860 observed answers. Interval not reported in the published source.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
The median company was named in 20% of the ten answers for its own category. Median across 85 companies, equivalent to 2 of 10 category answers. Interval not reported in the published source. Do not conflate it with the 27.5% mean above.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
30 of 85 companies, 35%, were never named in a single answer for their own category. Observed company proportion: 30 of 85 companies. Interval not reported in the published source.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
The average AI answer named 2.05 vendors, with a median of 2. Shortlists are short. Two or three names take the whole category conversation.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
24.0% was the mean per-company citation rate. Mean of per-company rates across 85 companies and 860 observed answers. Interval not reported in the published source. A citation records that the domain appeared among the answer sources.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
What engines cite
5,160 source citations were traced across 1,753 unique domains. The published dataset records source citations returned with the answers and their domains. A citation record does not establish an engine visit.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
77.4% of citations among the top 100 domains resolve to vendor-owned sites. Citation share: 1,454 of 1,879 citations among the 100 most-cited domains. Interval not reported in the published source.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
Review platforms account for 10.2% of citations. Citation share: 192 of 1,879 citations among the 100 most-cited domains. Interval not reported in the published source.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
0.0% of citations among the top 100 domains came from community sources. Citation share: 0 of 1,879 citations among the 100 most-cited domains. Interval not reported in the published source.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
47% of answers that named a company also cited that company website. Conditional naming/citation proportion within the 860-answer study. The exact joint count, denominator and interval are not reported in the published tables.
The 2026 State of GEO, Volume I. DOI: 10.5281/zenodo.21537014
The absence ladder
616 absences were recorded and classified by question shape. The unit of study is the absence: an answer where a measured vendor could have appeared and did not.
The Absence Ladder, State of GEO Volume II. DOI: 10.5281/zenodo.21586091
Question shape and visibility tier are associated: chi-square 61.8, df 15, p 1.3e-7. Which questions a vendor goes missing from depends on how visible that vendor already is. The association is far beyond chance.
The Absence Ladder, State of GEO Volume II. DOI: 10.5281/zenodo.21586091
Cramer's V, the effect size, is 0.18. A real but modest association, reported alongside the significance test.
The Absence Ladder, State of GEO Volume II. DOI: 10.5281/zenodo.21586091
35% of challenger vendors were named zero times in their own category. Across 60 categories. Observed company proportion: 30 of 85 companies. Interval not reported in the published source.
The Absence Ladder, State of GEO Volume II. DOI: 10.5281/zenodo.21586091
The median challenger was named in 2 of 10 category answers. Half of all challengers clear at most two of their ten buyer answers.
The Absence Ladder, State of GEO Volume II. DOI: 10.5281/zenodo.21586091
Leader strength does not predict challenger visibility: Spearman rho -0.051, p 0.64, n 85. A null result, published in full. Dominant leaders are not what keeps challengers out of AI answers.
The Absence Ladder, State of GEO Volume II. DOI: 10.5281/zenodo.21586091
Nine companies were named zero times while their own domain was cited as a source. A vendor domain can appear in the source list while its name is absent from the answer. Being cited is not being named.
The Absence Ladder, State of GEO Volume II. DOI: 10.5281/zenodo.21586091
Volume III, cross-engine divergence (2026)
Google AI Overviews agrees with itself 0.499 and with other engines 0.240, a gap of +0.258 (95% CI 0.178 to 0.338) over 62 repeat pairs.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
Claude agrees with itself 0.442 and with other engines 0.277, a gap of +0.165 (95% CI 0.082 to 0.242) over 74 repeat pairs.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
Once list length is controlled, Claude's gap falls to +0.075 under truncation and +0.002 under an overlap coefficient, both not significant. Google AI Overviews holds at +0.167 and +0.193.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
Mean pairwise Jaccard of the vendor set across all engine pairs: 0.313 (95% CI 0.295 to 0.330), 590 pair observations.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
63.3% of distinct vendor mentions came from exactly one engine; 17.5% from all three (142 questions, three engines). Vendor-mention shares: 1,416 of 2,238 from one engine and 392 of 2,238 from all three, across 142 questions. Interval not reported in the published source.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
Google returned an AI Overview for 92.1% of 280 B2B software buyer questions. Observed question proportion: 258 of 280 questions. Interval not reported in the published source.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
Vendors named per answer: Google AI Overviews 4.8, Claude 8.8, Perplexity 10.7, ChatGPT 14.7. The ChatGPT figure rests on 18 answers only and is indicative.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
14 of 32 companies testable on more than one engine were visible on some engines and invisible on others.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
853 scored answers, 40 categories, collected 4 August 2026.
The 2026 State of Generative Engine Optimization, v3.0. DOI: 10.5281/zenodo.21789120
The full Volume III analysis is at broadcastwell.com/engine-divergence.
Publisher note
Broadcastwell publishes these statistics and provides AI search visibility services. The dated research datasets retain their original observations, methods and limitations. Current commercial evidence is documented separately through selected client work, transparent pricing and clearly labelled illustrative examples.
Cite this page
Any figure above can be lifted whole with its study and DOI. A suggested citation for this page:
Broadcastwell (2026). GEO and AI Search Visibility Statistics 2026, from Open Datasets. broadcastwell.com/geo-statistics. Last updated 5 September 2026.
For the underlying studies: Broadcastwell (2026). The 2026 State of GEO, Volume I. Zenodo. doi.org/10.5281/zenodo.21537014 and Broadcastwell (2026). The Absence Ladder, State of GEO Volume II. Zenodo. doi.org/10.5281/zenodo.21586091
Method and limits
Volume I scored 860 AI answers, ten buyer questions per category, across 85 B2B software companies in 61 categories in July 2026, and traced 5,160 source citations across 1,753 unique domains. Volume II analyzed the same 85 companies across 60 categories and recorded 616 absences classified by question shape. Volume III put 280 B2B software buyer questions across 40 categories to four production AI search products on 4 August 2026 and re-ran 25 of them three times to establish a within-engine baseline. None of the three studies measures product quality. They measure how often AI engines name and cite vendors, in one window, with one question set per category: one engine held constant for Volumes I and II, four engines for Volume III. The datasets are public. Published open access under CC BY 4.0. Not peer reviewed. The linked research records provide the available datasets and methods. Where an exact base or a calculation is not available in the published tables, that limitation is stated beside the statistic.
Last updated: 5 September 2026.
See where you stand in your own category
See which buyer questions leave you out, the sources visible in those answers, and the three changes worth testing first. The Category Audit covers ten questions; the Diagnostic extends to 35 questions and 525 scheduled observed answers, adaptive runs extra. AI answers vary, so repeated observations and disagreements are reported.
Start with the $490 Category AuditResults appear in your account; a sign-in link arrives by email. Credits in full against the Diagnostic within 30 days.
Get the Diagnostic, $990Results appear in your account; a sign-in link arrives by email. Credits in full against the program.