What 860 AI buyer answers across 61 B2B software categories reveal about who gets recommended, who gets ignored, and where AI answers actually get their evidence.
Broadcastwell Research. Published July 2026. Data collected July 18 to 23, 2026.
Between July 18 and 23, 2026, Broadcastwell ran 10 standardized buyer questions through AI search with live web results for each of 85 B2B software companies across 61 categories, producing 860 scored answers and 5,160 traceable source citations.
On engines: this study holds one engine constant (Claude with live web search) across all 860 runs so results stay directly comparable across categories. The current Category Audit runs five engines (ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode) and compares them. The single engine constraint applies to this study only. The measurement method is published in full at github.com/Broadcastwell/audit-method. Every figure comes from the published method, v1.1.
One limitation matters up front: all 860 answers came from a single engine, Claude with live web search, run once per prompt. This is a snapshot of one system on the days it ran, not a claim about AI search in general. The fuller limitations are in the Methodology section below.
The results describe a market that has quietly reconcentrated. The median B2B company is named in just 20% of AI answers for its own category, and 35% of companies are never named at all. Meanwhile, in most categories a single brand captures 80% or more of all AI mentions. The average answer recommends only 2 vendors: the shortlist has shrunk from ten blue links to two names.
Median company rate: 2 of 10 answers across 85 companies; 30 of 85 companies were never named. The 80% leader figure is a median leader rate, equivalent to 8 of 10 answers across 85 company sweeps. Interval not reported in the published source.
The evidence behind those answers is not what most marketing teams assume. Among the 100 most cited domains, 77.4% of citations resolve to vendor authored content, companies writing the category's answer material themselves. Review platforms account for 10.2%, independent media 6.5%, and analyst firms 5.9%.
Observed citation shares among the top 100 domains: 1,454 vendor, 192 review, 123 media, 110 analyst and 0 community citations, out of 1,879 classified citations. Interval not reported in the published source.
The practical implication: in B2B software, AI visibility is not won on community platforms or earned media alone. It is won by publishing the reference content that AI systems retrieve, and by being present on the small set of review platforms they consistently trust.
Key Findings
Finding 01
35% of B2B companies are invisible in their own category.
Across 85 companies, the median company was named in 20% of the 10 buyer answers for its own category. 30 of 85 companies (35%) were named in zero answers. Only 31% of companies were named in at least half.
The median is 2 of 10 answers across 85 companies; observed proportions are 30 of 85 named zero times and 26 of 85 named in at least half. Bar ranges are visibility bin definitions. Interval not reported in the published source.
Named in 0% of answers
30 companies
Named in 1 to 20%
17 companies
Named in 21 to 49%
12 companies
Named in 50% or more
26 companies
Finding 02
AI answers are winner take most.
In each category we identified the single most mentioned brand across that category's answers. The median category leader appears in 80% of its category's answers. In 50 of 85 company sweeps (59%), the leader appears in 80% or more of answers. In only 3 of 85 does the leading brand appear in fewer than half.
Median leader rate: 8 of 10 answers across 85 company sweeps. The observed high visibility leader proportion is 50 of 85. Bar ranges are visibility bin definitions. Interval not reported in the published source.
Leader takes under 50%
3 companies
Leader takes 50 to 79%
32 companies
Leader takes 80% or more
50 companies
Finding 03
77% of AI evidence is vendor authored. Reddit: 0%.
We classified the 100 most cited domains (1,879 citations, 36% of all citations). 77.4% of those citations resolve to vendor owned domains: comparison posts, category guides and 'best X software' listicles written by the vendors themselves. Review platforms such as G2, Capterra, TrustRadius and SourceForge account for 10.2%; independent media 6.5%; analyst firms 5.9%, with gartner.com the single most cited third party domain in the study (110 citations). Community sources such as Reddit, Quora and Stack Overflow account for 0.0% of citations among the 100 most cited domains (1,879 citations, 36% of the total).
Observed citation shares among the top 100 domains: 1,454 vendor, 192 review, 123 media, 110 analyst and 0 community citations, out of 1,879 classified citations. Interval not reported in the published source. The top 100 share uses 1,879 of 5,160 total citations.
Vendor authored content
77.4%
Observed citation share: 1,454 of 1,879 classified citations. Interval not reported in the published source.
Review platforms
10.2%
Observed citation share: 192 of 1,879 classified citations. Interval not reported in the published source.
Independent media
6.5%
Observed citation share: 123 of 1,879 classified citations. Interval not reported in the published source.
Analyst firms
5.9%
Observed citation share: 110 of 1,879 classified citations. Interval not reported in the published source.
Community (Reddit, Quora, Stack Overflow)
0.0%
Observed citation share: 0 of 1,879 classified citations. Interval not reported in the published source.
Finding 04
There is no single source of truth. The evidence layer is fragmented.
The 5,160 citations spread across 1,753 unique domains. The top 10 domains account for 609 of 5,160 citations (11.8%), and the top 100 for 1,879 of 5,160 (36.4%). The recorded citations span a long tail of category specific pages.
Observed citation shares: top 10 domains, 609 of 5,160 citations; top 100, 1,879 of 5,160. Interval not reported in the published source.
12%
share of citations held by the top 10 domains
Observed citation share: 609 of 5,160 citations. Interval not reported in the published source.
1,753
unique domains cited
Finding 05
Compare question shapes under a fixed method.
The study separates best-of questions, direct comparisons and use case questions. Read the published dataset and method with each question set when comparing counts.
The study records 860 answers across 85 company sweeps.
Use each question set's recorded answer base when comparing naming and citation counts.
Finding 06
Being cited is not the same as being recommended.
In 97 of 860 answers, a company's domain was cited without the brand being named in the recommendation. Naming and domain citation are separate observed outcomes.
Observed count: 97 of 860 answers cited the company domain without naming the brand. Interval not reported in the published source.
97
of 860 answers cited a domain without naming the brand
Volume II extends this work by classifying every absence: The Absence Ladder.
Segment breakdown
Data table: Segment, Companies, Answers, Avg named, Avg cited, Named in zero answers
Segment
Companies
Answers
Avg named
Avg cited
Named in zero answers
Product & Dev Tools
21
210
47.1%
33.8%
2
Sales & Marketing Tech
21
220
16.7%
22.4%
10
Ecommerce & Site Search
12
120
18.3%
15.8%
5
HR & Operations
10
100
26.0%
24.0%
4
Customer Experience
9
90
28.9%
22.2%
4
Data & Analytics
9
90
16.7%
12.2%
5
Vertical & Niche SaaS
3
30
36.7%
40.0%
0
All segments
85
860
27.5%
24.0%
30
Avg named and Avg cited are means of per company rates, not pooled answer proportions. Company and answer bases are shown in each row; all segments cover 85 companies and 860 answers. Interval not reported in the published source.
Average share of a company's ten category answers in which it was named or its domain cited. Vertical & Niche SaaS covers three companies, so treat that row as directional. Sales & Marketing Tech shows 220 answers because one company in that segment was swept twice.
The Citation Index
Most named brands across all 860 answers
Data table: #, Brand, Mentions, Share
#
Brand
Mentions
Share
1
6sense
123
14.3%
2
Demandbase
101
11.7%
3
Algolia
63
7.3%
4
Gong
37
4.3%
5
Bloomreach
30
3.5%
6
Zendesk
29
3.4%
7
Terminus
27
3.1%
8
Constructor
27
3.1%
9
Clari
26
3.0%
10
Freshdesk
21
2.4%
Observed answer shares. Each row gives the numerator in Mentions; all shares use 860 scored answers. Interval not reported in the published source.
Most cited third party domains
Data table: Domain, Citations, Type
Domain
Citations
Type
gartner.com
110
Analyst
g2.com
57
Reviews
peoplemanagingpeople.com
31
Media
sourceforge.net
30
Reviews
trustradius.com
24
Reviews
thedigitalprojectmanager.com
22
Media
netguru.com
21
Media
thecmo.com
21
Media
capterra.com
20
Reviews
Full top 100 domain table available in the dataset downloads below.
Methodology
Three horizontal rows. Each shows a point estimate as a filled square with a line running through it that marks the range consistent with the observed answers.
01Every rate is reported with its numerator and denominator
02Every rate carries a 95 percent interval
03The interval describes this run, not a future outcome
A point estimate on its own hides the spread. Every rate carries its numerator, denominator and interval.
Engine. This edition records Claude answers with web search enabled, together with the sources those answers cited at the recorded observation times. This edition holds the engine constant across all 860 runs so results stay directly comparable across the categories in this published edition. The Category Audit now runs five engines (ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode) and compares them; the single engine constraint applies to this study only. The Q4 2026 edition will extend the multi engine comparison to the full research sample.
Prompts. For each company we generated 10 standardized buyer questions in its exact category: a mix of best of, direct comparison, and use case questions, phrased the way a real buyer types them.
Scoring. Each answer was scored on two binary outcomes: whether the measured company was named, and whether its domain appeared among the cited sources. Every cited URL was logged, producing 5,160 citations across 1,753 domains. One company was swept twice, returning 20 answers rather than 10, which is why 85 companies produce 860 scored answers. Volume II's per company analysis uses a single score per company and therefore reports 850.
Sample. 85 B2B software companies across 61 categories, skewing toward challenger brands (the companies most likely to need this measurement). Category leaders enter the data through mentions, not through selection.
Classification. The 100 most cited domains (36% of all citations) were classified by hand into vendor authored, review platform, independent media, analyst, and community. Separately, all 5,160 cited URLs were matched as strings against reddit.com, quora.com and stackoverflow.com; none appeared in any answer. The community figure therefore covers the full citation set, not only the classified sample.The classified domain proportion is 1,879 of 5,160 citations. The full corpus community check found 0 of 5,160 cited URLs. Interval not reported in the published source.
Limitations. Single engine, single run per prompt. AI answers vary run to run: repeated runs of the same question can return different leaders, so all figures are point in time estimates, not fixed rankings. The sample skews toward challengers, which lowers average named rates relative to a random sample of all vendors. Percentages are rounded.
Dataset Downloads
most_named_brands_top50.csv
Every brand named 2+ times across the 860 answers, with mention counts.
Company identities in the underlying sweep data are withheld. Aggregates and brand level mentions of widely known market leaders are published; per company results are anonymized.
Suggested Citation
Broadcastwell Research (2026). The 2026 State of Generative Engine Optimization, v1.0. https://doi.org/10.5281/zenodo.21537014
@report{broadcastwell2026geo,
title = {The 2026 State of Generative Engine Optimization},
author = {{Broadcastwell Research}},
year = {2026},
version = {1.0},
doi = {10.5281/zenodo.21537014},
url = {https://broadcastwell.com/state-of-geo}
}
Published open access under CC BY 4.0 with a DOI. This study has not been peer reviewed. The full dataset is public so anyone can recompute every number in it.
The cited Volume I research and datasets are published open access under CC BY 4.0 and have not been peer reviewed.
Frequently Asked Questions
What is Generative Engine Optimization (GEO)?
GEO is the practice of making a brand visible and citable in AI answers (ChatGPT, Perplexity, Claude, Google AI Overviews) the way SEO made brands visible in search results. This report measures the current state of that visibility in B2B software.
How were the 860 answers generated?
10 standardized buyer questions per category, run through a frontier AI model with live web search between July 18 and 23, 2026, then scored for brand naming and domain citation.
Can we reuse these findings?
Yes, with attribution. Use the suggested citation on this page. The downloadable datasets are provided for research and editorial use.
Will this be updated?
Yes. The State of GEO is an annual flagship report with a multi engine edition planned for Q4 2026.
One buyer question. One completed repair.
We build one source-backed page or fix for $490. Choose Live or Handoff before payment. Live requires access and approval before payment. The deadline is 48 elapsed hours after recorded Ready.
If we miss the deliverable or deadline, we initiate a full $490 refund within five business days. You keep the work. You may also request a full refund within 30 days of delivery.
We build one source-backed page or fix for $490. Choose Live or Handoff before payment. Live requires access and approval before payment. The deadline is 48 elapsed hours after recorded Ready.
If we miss the deliverable or deadline, we initiate a full $490 refund within five business days. You keep the work. You may also request a full refund within 30 days of delivery.