Most B2B companies are invisible to AI.
We measured exactly why.
Two waves of original research across ChatGPT, Perplexity, and Google AI Overviews. Wave 1 scored 20 companies on the structure AI actually rewards. Wave 2 measured four full industries and found 73% to 86% of vendors receive zero citations. Everything is public. No gate.
See what actually drives AI visibility.
Five Wave 1 findings and four industry snapshots. What AI rewards, what it ignores, and the patterns keeping vendors off the shortlist.
Read the findings ↓ Journalists & analystsCite the research. Interview the researcher.
Named patterns, exclusive datasets, quotable findings, and a 24-hour response commitment for anyone covering AI search.
Media resources ↓ Your companyFind out where you fall.
The same 7-layer analysis used in this research, run on your company. Written diagnostic in 48 hours.
Run the Snapshot ↓salesmate.io
monday.com
+ 5 others
Five findings from Wave 1.
The companies publishing the most content were cited the least.
Layer 4 · Citation InvisibilityAsked "What is the best approach to AI CRM?" Perplexity cited emailvendorselection.com, salesmate.io, monday.com, and five others. Zero citations of salesforce.com.
The $41B category creator of CRM is talked about by AI constantly. AI does not cite a single piece of Salesforce's owned content when answering the question Salesforce should own. This pattern held across every company in the study. Layer 4 (Training-Ready Content) failed universally. Regardless of company size, budget, or content volume.
The finding: Brand recognition and content authority are structurally disconnected in AI retrieval systems. Being known is not the same as being cited.
Revenue does not predict who AI calls an authority.
The Tier ParadoxVanta, a $2.45B funded challenger, scores 84 in AI authority. monday.com, $1.2B in annual revenue and publicly traded, scores 83. A Tier 3 company outranks a Tier 2 company.
The mechanism: Vanta owns "compliance automation" with no direct peer alternatives on any platform. monday.com competes in "work management" against Asana, Jira, and Smartsheet simultaneously. Narrow category ownership produces authority language. Broad category competition produces peer language.
The finding: Category ownership determines authority framing. Market position does not. This pattern confirmed across 4 of 5 industries.
Curious where your company falls?
Every company in this research was measured on the same 7-layer framework. The Snapshot runs the same analysis on yours. Written diagnostic in 48 hours.
Thought leadership volume produces less citation, not more.
The Deloitte ParadoxDeloitte Insights publishes thousands of thought leadership pieces annually. On every AI platform, zero results cite deloitte.com when answering questions about digital transformation or AI strategy. ChatGPT cites CIO.com, Gartner, PMI, and ISACA instead.
The mechanism: vendor-branded content is structurally treated as marketing by AI systems. Neutral, standards-aligned, third-party sources win the citation game regardless of content quality or volume.
The finding: Content volume is not content authority. Structure, neutrality, and standards-alignment drive AI citations. The failure is industry-wide and structural.
Visibility does not decline. It collapses.
The T4 CliffThe score degradation between tiers is not linear. Tally (Tier 4, SaaS) scores 57. Vanta (Tier 3, SaaS) scores 84. A 27-point cliff between adjacent tiers, concentrated in Layer 2 (Expertise: 1.0), Layer 4 (Content: 0.3), and Layer 6 (Trust: 1.7).
Semantic position determines it. A company either owns its category language or borrows it. Tally is algorithmically defined as "cost-effective Typeform alternative." Rho (also Tier 4, FinTech) scores 74 because it owns "all-in-one finance ops." Same tier. 17-point gap. One owns its category. One does not.
The finding: Derivative positioning collapses faster than owned positioning in AI retrieval. At every tier.
Three AI platforms produce three different verdicts on the same company.
Platform Behavior Is Systematic, Not RandomGemini calls Vanta "pioneering the category" and "the gold standard for early-stage startups." ChatGPT lists Vanta alongside 10 peers with zero authority language. Perplexity cites vanta.com twice as the primary example. Same company. Same week. Three different assessments.
These are not random variations. ChatGPT (The Cautious Advisor) lists peers without ranking and cites frameworks over vendor content. Perplexity (The Transparent Auditor) shows citations, surfaces negative reviews, and is the hardest platform to survive with mixed trust signals. Gemini (The Confident Recommender) assigns authority language readily. With the least rigorous sourcing.
The finding: AI visibility is platform-specific. Most companies optimize for none of the three.
Wave 1 Executive Summary
Layer-by-layer scores for all 20 companies. Platform-by-platform behavior analysis. The five findings with supporting verbatims. 8 pages. Direct download, no form.
Visibility does not decline gradually. It collapses when category ownership breaks.
Score gradient across company tiers. Tally to Vanta: 27-point cliff in adjacent tiers. Ownership of semantic position determines the drop, at every company size.
What AI actually said. Unedited.
"Asked best approach to AI CRM: cited emailvendorselection.com, salesmate.io, monday.com, and 5 others. Zero citations of salesforce.com."
The $41B category creator. Not cited in its own domain question.
"Vanta: 'pioneering the category' and 'gold standard for early-stage startups.' monday.com: 'Best Known For: Visual Customization.'"
One gets a positioning claim. One gets a feature label. Same platform. Same response.
"Digital transformation strategy citations: CIO.com, Gartner, PMI, ISACA. Citations of deloitte.com: zero."
The most prolific thought leadership publisher in the study. Least cited.
"'The largest company in the world dedicated to industrial automation.' Allen-Bradley: 'a de facto standard in many North American factories.'"
T2 company. T1 authority language. Mechanism: owned niche, zero peer alternatives.
Wave 1 scored companies. Wave 2 measures categories.
Wave 2 asks the buyer's question: when someone asks AI for a shortlist, who gets named? Share of Model measures the percentage of AI responses in a query set that name a specific vendor. Above 5% composite makes a vendor a Category Winner. Below 1% is Category Absence: functionally invisible at the moment of shortlist. Across four industries, 73% to 86% of tested vendors sit below that line.
73% invisible. Eight vendors hold the entire category.
MSSP, MDR, and SOCaaS across ChatGPT, Perplexity, and Google AI Overviews. 15 buyer-intent queries per platform against roughly 3,000 active vendors. Eight capture the full Share of Model.
Read the findings → Industrial B2B · Published120 vendors. 84% invisible. The old brand names win.
Wave 2 findings across 12 industrial sub-categories, including the Zombie Rebrand pattern keeping rebranded vendors locked to their former identity.
Read the findings → Healthcare & Healthtech · Published86% invisible. The strictest trust thresholds in the study.
Wave 2 findings across 12 healthcare sub-categories. 42% of clinical queries refused, evidence routed through peer review, and Cerner still winning 44 months after becoming Oracle Health.
Read the findings → B2B Fintech · Published88 of 90 startups invisible. The incumbents split the map.
Wave 2 findings across 12 fintech sub-categories, 120 vendors. Pure-plays win developer-led categories, legacy incumbents win bank buyers, and the old brand name returns in 81% of rebrand queries.
Read the findings →Marketing technology and revenue operations vendors.
The most crowded category language in the study. Peer-language saturation at its worst.
In queueFractional executives and advisory firms.
Where the founder is the brand. Personal entity resolution decides the shortlist.
In queueWant your industry prioritized? Apply for the next snapshot → · Want your company's own numbers? Run the Snapshot →
The Index measures who AI names. The citation study measures who told it what to say.
On 20 August 2026 the study captured every citation behind 40 buyer-intent vendor-selection queries across ChatGPT, Perplexity, Gemini and Claude: 1,855 citations, 972 domains, every one classified by source type. A stratified random sample of the third-party layer was then audited by opening and reading the cited page on each domain. Three quarters of those pages sell into the category the page is ranking, and seven in ten of those disclose nothing. More than half the sample is a software vendor placing its own product in its own ranking.
Reddit, the source everyone is chasing, held 10 citations of 1,855, and appeared zero times on Perplexity and zero times on Claude. Sources with an editorial or analyst function held 12.2%. This is Trust Seed Substitution measured at market scale: the evidence layer AI substitutes for your content is mostly your competitors' marketing.
Updated 26 August 2026. The provenance audit behind the figures first published on 21 August was re-drawn, because the row-level data from the first pass was not retained. The study page carries the full comparison. See what changed →
What held. What deepened. What broke.
Layer 4 failed universally, twice.
In Wave 1, Salesforce and Deloitte earned zero citations on their own domain questions. In Wave 2, 100% of tested fintech vendors fail training-ready content, and so do 100% of emerging healthcare startups. The Volume Paradox replicated in every vertical: more unstructured content produces less machine visibility.
Category ownership held too. Wave 1's Vanta effect returned at category scale as the vertical specificity lift: specialized vendors outpace generalists by 28% in industrial and 31% in fintech.
The 27-point cliff has a floor. Wave 2 found it.
Wave 1 measured a 27-point score cliff between Tier 3 and Tier 4. Wave 2 measured what that cliff looks like at category scale: startup Category Absence reaches 92% in healthcare and 97.78% in fintech. In fintech, 88 of 90 tested startups register on no platform at all. Funding does nothing to offset it.
Three patterns surfaced only at category scale.
The Zombie Rebrand: AI returns the old brand name in 78% to 81% of rebrand queries, years after the change. Cerner still wins 44 months after becoming Oracle Health. Bill.com still wins 91% of queries after becoming BILL.
Trust Seed Substitution: when brand content fails extraction, AI routes evidence elsewhere. Clinical queries route to peer review and FDA databases. Fintech queries route to FinCEN, the CFPB, and trade press. Administrative queries route to G2 and Reddit. Which surface receives the substitution is query-specific, and the difference is large: the August citation study found Reddit at 0.54% of citations in vendor-selection queries, and zero on Perplexity and Claude. Most vendors invest on none of the surfaces that decide their shortlist.
Category Confusion: 64% to 71% of returned vendors operate in a category adjacent to the one queried. The buyer receives a confident shortlist built on the wrong category and cannot see the error.
The YMYL hypothesis split down the middle.
Healthcare confirmed it with the strictest trust behavior in the study: 86% Category Absence and 42% of clinical queries refused outright. Fintech contradicted it: 74.17% absence across 120 vendors, below industrial's 84%. Regulation raises the evidence bar for a citation. It does not automatically deepen invisibility. Each vertical earns its own trust threshold, and the snapshots document each one.
The dashboards stopped explaining reality. So I built a new measurement.
Maria Dykstra diagnoses why B2B companies are invisible to AI systems. Before this research: 13 years at Microsoft as Senior Product Planning Manager, building global ad systems that drove $2B in revenue across 36 markets and served a billion ads a month. Then 13 years running the digital agency TreDigital.
The pattern that started the Index: companies ranking on page one of Google never appearing in a ChatGPT answer. Human visibility and machine visibility had split into two separate systems, and every existing metric measured only the first one. Nobody was measuring the second.
The Algorithmic Authority Stack became the diagnostic: seven layers, from market identity to visibility measurement, tested against live platform behavior. The Index became the benchmark: Share of Model, measured company by company, then industry by industry. The book on this research, The Invisible Audience: Why AI Can't Find You (and How to Fix It), publishes October 2026.
This research is built to be cited.
Industry snapshots publish throughout 2026, each with a documented dataset, named failure patterns, and platform verbatims. If you cover AI search, B2B marketing, or how retrieval systems reshape buying, the underlying data is available to you.
Open to: co-authored industry reports, exclusive dataset access ahead of publication, podcast discussions, conference presentations, expert commentary on breaking AI search news, and AI visibility analysis for your audience or membership.
Bio, headshots, logos, research summaries, and speaking topics: press kit · speaking
- Podcast interviews
- Conference keynotes
- Industry commentary
- Guest articles
- Research partnerships
- Expert quotes
Average response time: 24 hours
"Revenue predicts almost nothing about AI visibility."
Maria Dykstra · The Algorithmic Authority Index, Wave 1"Visibility does not decline. It collapses."
Maria Dykstra · The Algorithmic Authority Index, Wave 1"Buyers do not arrive at companies AI cannot place into the shortlist."
Maria Dykstra · The Algorithmic Authority Index"Most B2B companies have an SEO strategy. Almost none have an AI retrieval strategy."
Maria Dykstra · The Algorithmic Authority Index, Wave 2"Three quarters of the 'independent' sources AI cites in B2B vendor research sell into the category they are ranking. More than half are vendors ranking their own product."
Maria Dykstra · Who Ranks the Rankings, August 2026Dykstra, M. (2026). The Algorithmic Authority Index, Waves 1–2.
mariadykstra.com/index/ · Versions: AAI-W1 (March 2026), AAI-W2 (July 2026)
Dykstra, M. (2026). AI Citation Provenance Study: Who Ranks the Rankings.
mariadykstra.com/index/ai-citation-sources/ · Captured 20 Aug 2026, audit re-drawn 26 Aug 2026
Licensed CC BY-NC 4.0. Findings and named patterns are free to reference with attribution. For interviews, dataset access, or fact-checking, contact Maria directly. Response within 24 hours.
The Index shows what fails. The Snapshot shows what is failing for you.
The AI Visibility Snapshot tests your company against the same 7-layer Algorithmic Authority Stack used in the Index. You receive a written diagnostic within 48 hours: the layer breaking first, the structural reason, and the competitors holding the slots you should hold. The $497 credits in full toward the Audit if you upgrade within 30 days.
Run the Snapshot · $497 →No call. Two-field intake. Written delivery in 48 hours.
Wave 1. Each company was tested using standardized prompts designed to trigger retrieval across all 7 layers of the Algorithmic Authority Stack. Tests conducted across ChatGPT (GPT-4o), Perplexity (default model), and Gemini 1.5 Pro during a controlled testing window. Each company scored 0 to 3 per layer per platform, converted to a 0 to 100 composite. 420 individual assessments across the Wave 1 sample.
Wave 2. The same framework, adapted to category-level measurement. The industrial, healthcare, and fintech deep-dives each ran 25 buyer-intent queries per sub-category across 12 sub-categories and 3 platforms, producing 900 documented sessions per industry. The cybersecurity snapshot used a different design: 15 buyer-intent queries per platform against a vendor universe of roughly 3,000 active vendors (Momentum Cyber Market Review 2025). Share of Model aggregates Mention Share, Citation Share, and Recommendation Share per vendor. Category Winners score above 5% composite. Category Absence sits below 1%. Full methodology on each snapshot page.
Citation study, August 2026. A separate design from both waves. 40 buyer-intent vendor-selection queries across four B2B categories, run once per platform on ChatGPT, Perplexity, Gemini and Claude on 20 August 2026, capturing every cited source: 1,855 citations across 972 domains. A stratified random sample of 59 third-party commercial domains was audited by retrieving and reading the cited page on each, against a written rubric covering named authorship, identifiable legal entity, commercial interest in the category being ranked, and disclosure. 17 sampled domains turned out to be a named vendor's own marketing page and were excluded; 2 could not be fetched; 40 remained. Single day, single run per query, and the four platform arms ran under different conditions, so shares are comparable and raw counts are not. Full method, limitations and correction log on the study page.
Gemini 1.5 Pro was tested as the model backing Google AI Overviews at the time of Wave 1. AI systems update continuously. Each wave represents a controlled snapshot of platform behavior, not a permanent ranking.