Data & Statistics
Machine Relations Research
Named-source data on AI search adoption, citation rates, and industry disruption — each stat attributed to its original study.
Published reports
Research studies, index analysis, research synthesis, reference material, and practitioner analysis from Machine Relations Research — each report classified by content type under the editorial standards.
- Study
How Many Sources Does a Typical AI Answer Actually Cite? Published Counts Range From 2 to 22
Fourteen public studies of how many sources an AI answer cites report figures from about 2 to nearly 22, and the spread is not mostly engine behavior. A worked audit of the Machine Relations Index's own public release shows the trap directly: a field named runs_cited_domains reads as a per-answer citation count but is a distinct-domain count for the whole stratum, understating true citation density by a factor of 3.2x portfolio-wide (2.16 against a correct 6.98). Corrected per-category, an AI answer in the median measured category cites 6.6 to 8.6 sources, not 2.
- Index Analysis
The Largest Source Class in the Head of the Index Is Companies' Own Websites
In the September 25, 2026 release of the Machine Relations Index, 29 of the 100 most-cited domains are vendor-owned sources, against 22 editorial publications, 16 market and company databases and one wire domain. Across the whole index a vendor-owned domain averages 13.9 domain-run citations; an editorial publication averages 10.8.
- Index Analysis
Machine Relations Index Report, September 23, 2026: Healthcare Services publishes as forecast, and 27 community-platform domains still out-lead 1,266 editorial publications
Second edition of the weekly Machine Relations Index Report. Release mri_score_v2.0+2026-09-23: 23,280 cited domains, 16,475 answer runs, 93 of 151 measurable segments published. Healthcare Services crosses the evidence floor as the prior edition forecast; segment leaders, engine differences, and movement since September 18.
- Study
A Placement Gets Cited When the Question Names What It's About, Not Because of the Outlet
Two runs, 12 and then 57 live earned-media placements, 216 and then 995 answers across six AI surfaces: the outlet a placement ran in barely moved its citation rate. Whether the buyer's question named the company or a term the article owns moved it from single digits to double digits. Company-naming questions: 6 of 7 placements cited. Generic category questions: 3 of 29.
- Study
How Many Domains Cover Half of AI Citations in Your Category? Between 49 and 247
The Machine Relations Index measures the domain-count concentration curve inside each of 20 subject categories. In the 16 that clear the evidence floor, the ten most-cited domains hold between 8.1% and 24.2% of that category's citations, the top 25 never reach 40%, no single domain reaches 4.1%, and covering half of a category's citations takes between 49 domains in family-software and 247 in enterprise-software. Two independent 2026 panels put the top ten at 11% to 13% and under 15%, inside the same band. The cross-topic claim that 15 or 30 domains own roughly two thirds of AI citations does not hold in any measured category.
- Study
Blocked and Cited Anyway: What a robots.txt Disallow Actually Does to AI Citation
Joining the 250 most-cited domains in the Machine Relations Index release of 2026-09-22 to their live robots.txt files, a full disallow on an engine's training crawler is a weak predictor of whether that engine cites the domain: 18 of the 26 head domains that fully disallow GPTBot were cited by ChatGPT across the 129-day window, and 15 of the 17 that fully disallow PerplexityBot were cited by Perplexity. The disallow that does track citation is the one on the search token: 3 of 10 for OAI-SearchBot and 6 of 18 for Google-Extended. Ten head domains, including the first and third ranked, serve a blanket Disallow for every crawler and are cited by at least three engines each.
- Study
Four Categories Where One Kind of Source Wins Every Buyer Question; Nine Where Nothing Does
Grouping the Machine Relations Index's classified citations by category and buyer-question shape, the leading source-role class (editorial publication, vendor-owned, academic and government, community and social, market and company database) changes with the question in 9 of the 13 categories that have full evidence across all six shapes. In the other four, one class wins every shape: editorial media in consumer-finance (39.8% to 73.9%) and family-software (54.1% to 79.3%), vendor-owned sources in cybersecurity (30.3% to 51.7%) and fintech (32.5% to 52.2%). A buyer's best-tools list and their head-to-head comparison draw from different source types in most categories; in these four they do not.
- Study
Most of What AI Agents Ask For and Cannot Find Is Not Content
Across a six-site sample over 28 days, AI user agents requested 59 URLs that did not exist. Twenty-six of them were credential, configuration or remote-execution paths, and those 26 took 226 of the 314 requests. Ranked by demand, the first request for a page a reader could want sat 24th. An AI-crawler 404 list is not a content backlog until it is filtered.
- Study
Recognition Is Not Recall: 85 Days of One Brand's Model Memory Against Its Retrieval Citations
Across 85 measured days and 2,175 prompt-days, a language model answering with retrieval disabled mentioned our brand zero times on prompts that did not already name it. Over the same window the same prompts on five retrieval engines moved our share of returned citation URLs from 22.7 percent to 30.7 percent. Retrieval standing and model memory are separate assets, and one of ours went to zero while the other rose.
- Index Analysis
Only 508 of 22,377 AI-Cited Domains Are Cited Often Enough to Score
In the September 21, 2026 Machine Relations Index, 22,377 domains were cited by AI answer engines and 508 were cited often enough to carry a confidence grade. Those 508 hold 34.66% of all citation. Which source class you belong to changes your odds of ever being scoreable by twelve times.
- Index Analysis
Do AI Citation Leaderboards Change Between Daily Releases? What the Machine Relations Index Re-Measures
The Machine Relations Index compared two of its releases three days apart, segment by segment. All 84 comparable published segments were numerically identical, because published buyer-decision segments are fixed seven-date samples while news segments are observed continuously. What that means for anyone buying a daily AI visibility dashboard.
- Reference
How the Machine Relations Index Measures AI Citations: Methodology and Update Log
The Machine Relations Index methodology and update standard: how the Index measures AI-engine citation, which figures are direct observations and which are class-level, the five rules the Index applies to every published figure, and a dated log of each methodology update.
AI Search Adoption
How quickly answer engines are replacing traditional search as the first step in commercial discovery.
810M
ChatGPT monthly active users (2026)
OpenAI investor presentations
750M
Google Gemini monthly active users
Alphabet earnings calls
9.7×
AI search traffic year-over-year growth
Similarweb, Cloudflare Radar
−50%
Gartner-predicted traditional search decline by 2028
Gartner Predicts 2024
Fresh Evidence
Citation & Earned Media
What independent studies show about where AI systems source their citations.
82–89%
of AI citations come from earned media, not brand-owned content
Muck Rack Generative Pulse (Dec 2025); Fullintel + UConn IPRRC (Feb 2026)
200×
faster AI visibility for brands publishing 12+ citation-optimized pieces per month
AuthorityTech client data, 2024–2025
34%
of AI citations in a given category go to a single publisher
AuthorityTech citation analysis, 2025
Industry Disruption
Where legacy PR firms are losing revenue and where the money is moving as AI reshapes discovery.
−8.1%
Edelman US revenue decline (2025) — third consecutive year
PRWeek
4–5%
Publicis growth driven by AI/data capabilities
Publicis investor presentations
76%
of PR professionals now using AI tools daily
PRCA AI Adoption Survey 2025
83%
of Google queries are now zero-click
SparkToro Zero-Click Study
Fresh Evidence
Key insight
Every independent study on the topic lands on the same number: 82–89% of what AI systems cite is earned media. Not press releases. Not brand blogs. Earned coverage in publications the model already trusts.
ChatGPT now handles over 2 billion queries daily (OpenAI, 2025). Gartner projects traditional search will decline 25% by 2026. The brands that secure citation positions in this window will be difficult to displace — because AI citation compounds the way domain authority once did.