AI citation is far more concentrated than most visibility strategies assume. The Machine Relations Index measures citation rates for 16,356 domains across six AI answer engines — ChatGPT, Claude, Gemini, Perplexity, Google AI Mode, and Google AI Overviews — over a 74-day observation window. Of those 16,356 domains, only 8 earn A-confidence status and 40 earn B-confidence. The remaining 98.2% are either C-confidence or still collecting enough evidence to be scored.
What the MRI Confidence Tiers Measure #
The MRI v2 methodology grades each domain's citation rate by the volume and consistency of evidence behind it. A domain's citation rate is the percentage of observed answer-engine runs in which it was cited. The confidence tier reflects how much evidence supports that rate:
- A confidence — Sustained citation rates backed by high observation volume across multiple engines and many distinct run dates. These domains are cited consistently enough that the rate is stable.
- B confidence — Meaningful citation rates with moderate evidence. The rate is directionally reliable but may shift as more observations accumulate.
- C confidence — Measurable citation activity with limited evidence depth. The rate exists but carries more uncertainty.
- Collecting — The domain has been cited, but observations have not yet crossed the evidence floor (at least 10 observed runs across at least 7 distinct dates per segment). No rate is published; the domain is accumulating data.
The evidence floor exists to prevent thin data from producing misleading rates. A domain cited twice in two days is not comparable to one cited 300 times across 57 days. The confidence tier separates the two.
The 8 Domains That Earn A-Confidence Status #
Out of 16,356 measured domains, exactly 8 sustain citation rates high enough and consistently enough to reach A confidence. They represent five different source roles:
| Rank | Domain | Citation Rate | Runs Cited | Engines | Days Cited / Observed | Source Role |
|---|---|---|---|---|---|---|
| 1 | 11.81% | 1,122 / 9,498 | 4 | 69 / 74 (93%) | Community / social | |
| 2 | YouTube | 9.00% | 855 / 9,498 | 6 | 36 / 74 (49%) | Platform / search |
| 3 | 7.99% | 759 / 9,498 | 6 | 67 / 74 (91%) | Community / social | |
| 4 | Medium | 6.97% | 662 / 9,498 | 6 | 72 / 74 (97%) | Editorial media |
| 5 | Gartner | 4.00% | 380 / 9,498 | 6 | 63 / 74 (85%) | Analyst research |
| 6 | Forbes | 3.92% | 372 / 9,498 | 6 | 60 / 74 (81%) | Editorial media |
| 7 | G2 | 3.62% | 344 / 9,498 | 6 | 57 / 74 (77%) | Market database |
| 8 | arXiv | 3.30% | 313 / 9,498 | 6 | 42 / 74 (57%) | Academic / government |
Several patterns stand out:
No vendor-owned domain reaches A confidence. The highest-ranked vendor site, Landbase, sits at B confidence with a 3.13% citation rate. Microsoft (2.47%), IBM (3.01%), and Palo Alto Networks (1.72%) are all B-tier. Traditional brand authority does not automatically translate to AI citation reliability.
Reddit's 11.81% rate is 3.3 times higher than G2's (3.62%) and 118 times the median domain's rate (0.01%). A study by GeoBuddy analyzing 86,000+ citations across four AI engines similarly found that only 16 domains earned citations from all four engines they measured — confirming the extreme top-heavy distribution the MRI data shows at larger scale.
Temporal consistency — the percentage of observation days on which a domain was cited — separates otherwise similar citation rates. Medium is cited on 97% of measured days despite ranking fourth by rate. YouTube is cited on only 49% of days despite ranking second. A high rate concentrated in bursts is less reliable than a moderate rate spread across most days.
The B-Confidence Tier: 40 Domains #
B-confidence domains have meaningful citation rates but less evidence depth than the A tier. The top 15:
| Rank | Domain | Citation Rate | Runs Cited | Engines | Source Role |
|---|---|---|---|---|---|
| 9 | Landbase | 3.13% | 297 | 6 | Vendor owned |
| 10 | NIH.gov | 3.07% | 292 | 6 | Academic / government |
| 11 | IBM | 3.01% | 286 | 6 | Vendor owned |
| 12 | Crunchbase | 2.85% | 271 | 6 | Market database |
| 13 | Microsoft | 2.47% | 235 | 6 | Vendor owned |
| 14 | Yahoo | 2.41% | 229 | 5 | Editorial media |
| 15 | TechRadar | 2.18% | 207 | 3 | Editorial media |
| 16 | Substack | 2.14% | 203 | 5 | Community / social |
| 17 | Grand View Research | 1.98% | 188 | 6 | Market database |
| 18 | Deloitte | 1.87% | 178 | 6 | Analyst research |
| 19 | Dev.to | 1.80% | 171 | 6 | Community / social |
| 20 | NerdWallet | 1.77% | 168 | 5 | Editorial media |
| 22 | Palo Alto Networks | 1.72% | 163 | 6 | Vendor owned |
| 23 | Y Combinator | 1.71% | 162 | 5 | Community / social |
| 24 | Fortune Business Insights | 1.70% | 161 | 6 | Market database |
The B tier is where source-role diversity becomes visible. Market databases place four domains (Crunchbase, Grand View Research, Fortune Business Insights, and Mordor Intelligence at rank 30 with 1.46%). Community platforms place three (Substack, Dev.to, Y Combinator). Vendor-owned sites finally appear but only with broad enterprise portfolios (IBM, Microsoft, Palo Alto Networks).
The 98.2% That Are Still Collecting #
The distribution's most striking feature is its tail. Out of 16,356 measured domains:
| Confidence Tier | Domains | Share |
|---|---|---|
| A | 8 | 0.05% |
| B | 40 | 0.24% |
| C | 249 | 1.52% |
| Collecting | 16,059 | 98.18% |
The median domain citation rate across all 16,356 measured domains is 0.01% — cited in roughly 1 of every 10,000 observed answer-engine runs. The top 50 domains by citation rate all exceed 1.06%. The top 100 exceed 0.66%.
This distribution is steeper than traditional search. In organic search, rank positions 1 through 10 share the first page, and positions 11 through 20 still receive measurable traffic. In AI citation, the functional equivalent of "page one" is fewer than 50 domains out of more than 16,000, and the gap between position 1 (11.81%) and the median (0.01%) is more than three orders of magnitude.
Research by Jacques et al. examining 10,038 health citations from Claude found that established institutional sources accounted for 97.8% of all citations, with Mayo Clinic alone representing 24.7%. That level of concentration in a single vertical mirrors the cross-vertical pattern the MRI measures: AI engines select a small number of trusted sources and cite them repeatedly.
What Separates Domains That Get Scored from Domains That Don't #
The evidence floor — 10 observations across 7 distinct dates per measured segment — is not arbitrary. It represents the minimum threshold at which a citation rate becomes stable enough to report. Domains below it are not "uncited." They are cited too infrequently or too recently to produce a reliable rate.
Three structural factors separate scored domains from collecting ones:
Query coverage breadth. A-confidence domains are cited across multiple subject categories and question shapes (best-of lists, comparison queries, problem-first questions, news-driven topics). A domain cited for one narrow topic in one question format accumulates evidence slowly because each category-shape combination is scored independently.
Engine reach. Six of the eight A-confidence domains are cited by all six measured engines. The exceptions are Reddit (four engines — not cited by Google AI Mode or Google AI Overviews in measured segments) and YouTube (six engines but with temporal gaps). Domains cited by only one or two engines accumulate evidence in fewer segments.
Temporal spread. Being cited in bursts — 20 citations in one week, then silence — produces fewer distinct run dates than steady citation across the observation window. The evidence floor requires at least 7 distinct dates, which filters out domains whose citations cluster around specific events.
Implications for AI Visibility Strategy #
The confidence tier data reframes what AI visibility means in practice. Several conclusions follow from the distribution:
Most AI visibility work is competing for C-tier or collecting status. The 249 C-confidence domains represent the realistic ceiling for most optimization work. Moving from collecting to C — earning enough consistent citations to be scored — is the first meaningful milestone, not reaching A or B.
Source role matters more than SEO authority. The A-tier includes Reddit (a user-generated forum), arXiv (a preprint server), and Medium (an open publishing platform) alongside Gartner and Forbes. These domains do not rank because of backlink profiles or domain authority scores. They rank because AI engines select them for specific evidential functions.
Multi-engine consistency is the hardest bar. A domain can accumulate high citation volume on one engine (Gemini cites heavily, averaging 11 sources per response) and still sit at C confidence because it is invisible to the other five. The MRI's confidence system rewards breadth and temporal consistency, not volume spikes.
How This Connects to Machine Relations #
Machine Relations treats AI citation as a measurable, engineerable signal — not a black box. The confidence tier system makes the measurement itself transparent: instead of publishing a score and leaving practitioners to guess its reliability, the MRI publishes the evidence behind every rate and grades that evidence explicitly.
The practical implication for Machine Relations practitioners: the first goal is not to maximize citation rate. It is to cross the evidence floor and move from collecting to C — to be cited consistently enough, across enough engines and dates, that the rate becomes measurable. From there, the levers are query coverage breadth (appearing in more category-shape segments), engine reach (earning citations from engines that currently ignore the domain), and temporal consistency (sustaining citations over weeks, not days).
The 48 domains at A and B confidence did not achieve that status through content volume or optimization tricks. They achieved it by being the source AI engines repeatedly select when answering specific types of questions. That structural alignment is what Machine Relations measures and what practitioners should engineer toward.
FAQ #
How does the MRI calculate confidence tiers? #
The MRI grades each domain's citation rate by the volume and consistency of evidence behind it. A domain must cross an evidence floor of at least 10 observed answer-engine runs across at least 7 distinct dates before any rate is published. The tiers — A, B, C, and collecting — reflect how stable and well-evidenced that rate is, not the rate itself.
Why are so many domains still in the collecting tier? #
The 98.2% collecting rate reflects how selective AI citation actually is. Most domains have been observed in answer-engine runs but cited too infrequently or too recently to produce a stable rate. The evidence floor prevents thin data from being reported as settled scores.
Can a domain move from collecting to A confidence? #
In principle, yes, but the practical path is long. A domain needs to be cited consistently across multiple engines, subject categories, question shapes, and dates. The 8 current A-confidence domains have been cited on an average of 67% of observation days across the 74-day window. Achieving that level of temporal consistency requires structural alignment with how AI engines select sources, not short-term optimization campaigns.
Why are no vendor-owned domains at A confidence? #
No vendor-owned domain sustains the cross-engine, cross-category citation consistency required for A confidence. IBM, Microsoft, and Palo Alto Networks reach B tier because they are cited broadly across enterprise categories, but their citations are less consistent across all six engines than community platforms (Reddit, LinkedIn) or reference sources (Gartner, arXiv) that AI engines treat as neutral third-party evidence.