Two of Three Gap-Free AI Engines Keep Exclusive Sources at Volume
The Machine Relations Index's raw exclusivity share for all six AI answer engines, then the same number conditioned on citation volume. Among the three engines the Index collected without a gap, ChatGPT and Perplexity keep an elevated exclusive-source share past 5 citations and Claude collapses to 1.1%. The three gapped surfaces cannot be ranked on this measure.
The number
Of the 23,872 domains the Machine Relations Index has observed any of the six major AI answer engines cite in its 2026-05-10 to 2026-09-25 window (release mri_score_v2.0+2026-09-25+04fcb7fb8f29), 16,244 — 68% — are cited by exactly one engine. Split that by engine and the unconditioned exclusivity share looks like a real ranking:
| Engine | Domains it cites | Cited by no other engine | Exclusive share | Collection |
|---|---|---|---|---|
| Perplexity | 10,667 | 5,007 | 46.9% | Gap-free |
| Google AI Mode | 7,458 | 3,359 | 45.0% | Gapped |
| ChatGPT | 5,355 | 3,008 | 56.2% | Gap-free |
| Gemini | 7,276 | 2,913 | 40.0% | Gapped |
| Claude | 5,564 | 1,596 | 28.7% | Gap-free |
| Google AI Overviews | 2,304 | 361 | 15.7% | Gapped |
That table is the wrong artifact to act on, for two separate reasons. The first is population thinness: 97.8% of the 23,872 domains carry the Index's collecting confidence grade — under 10 observed runs across 7 distinct dates — and a domain the Index has seen cited exactly once can, by construction, only ever look exclusive to whichever single engine happened to cite it. A raw exclusivity share computed over a population that thin describes the tail, not the engines.
The second is that an exclusivity count is a claim about absence, and three of these six engines have dated collection gaps. See the correction below: every exclusive count on this page is an upper bound, and the three gapped surfaces cannot be ranked on this measure at all.
Correction, September 26, 2026 — exclusivity is a claim about absence, and three arms have collection gaps
Three of the six engines the Index monitors have dated collection gaps inside this window: Google AI Overviews recorded zero citations on any monitored query between August 15 and September 24, 2026 (41 of 132 days), Google AI Mode between September 14 and September 24 (11 days), and Gemini on 22 intermittent days — May 11 to 15, then scattered single days from August 11 to September 16. The two Google lanes share one cause, a defect in the Index's Google collection path that graded a response as a success whether or not it carried content, fixed September 25, 2026. Gemini's intermittency is a separate pattern and its cause is not established. ChatGPT, Claude and Perplexity collected without a gap; one Claude date was flagged by the same scan and carries exactly one answer run, a scan artifact rather than a gap.
An exclusivity count is unlike a presence count, because it depends on every other engine's absence being real. A domain counted here as exclusive to ChatGPT may have been cited by Gemini on a blind day. So:
- Every exclusive count on this page is an upper bound. The true exclusive set for each engine is at most the number shown and may be smaller.
- The three gapped surfaces cannot be ranked on this measure in either direction. Their own citations are undercounted, which thins both their exclusive sets and their 5+-citation populations, and their absences from other engines' domains are not established. A near-zero exclusive share for Gemini, Google AI Mode or Google AI Overviews is not a finding.
- The comparison among ChatGPT, Claude and Perplexity holds. All three collected without a gap, so their shares are measured against the same evidence and against each other on equal terms.
What is withdrawn. The earlier title and description of this page said "only two of six" engines still look narrow, and the closing paragraph said Claude, Gemini, Google AI Mode and Google AI Overviews all fail to show the pattern. The claim about Claude stands — Claude is gap-free and collapses to 1.1%. The claim about the three gapped surfaces is withdrawn: it placed gap-free arms above gapped arms on a measure whose bound runs the wrong way. The finding is restated over the three arms the Index collected without a gap.
What does not change. The page's actual argument is untouched and, if anything, sharpened: a raw exclusivity share is a data-maturity artifact, not an engine trait. Conditioning on citation volume collapses it. The correction adds a second way the raw number misleads.
The same question, conditioned
Re-run the same six exclusivity shares restricted to domains with at least 5 observed citations — enough that "no other engine has cited this" is a pattern rather than a coin flip — and the ranking above mostly disappears:
| Engine | Domains at 5+ citations | Still exclusive | Exclusive share | Reading |
|---|---|---|---|---|
| ChatGPT | 1,817 | 147 | 8.1% | Upper bound, gap-free |
| Perplexity | 4,247 | 326 | 7.7% | Upper bound, gap-free |
| Claude | 2,919 | 33 | 1.1% | Upper bound, gap-free |
| Gemini | 2,980 | 23 | 0.8% | Not rankable — gapped |
| Google AI Mode | 2,706 | 17 | 0.6% | Not rankable — gapped |
| Google AI Overviews | 1,470 | 0 | 0.0% | Not rankable — gapped |
At 10+ citations, every engine's exclusive share falls under 1.2%. The domains carrying real evidence — reddit.com, youtube.com, linkedin.com, forbes.com, nih.gov, gartner.com and arxiv.org — are cited by four to six engines apiece; the Index's confidence-A and -B tier is almost entirely shared, not exclusive, source territory.
Genuine narrow-aim sourcing, once noise is removed, survives on two of the three arms the Index collected without a gap. ChatGPT and Perplexity each still keep a small but real set of sources — seeklogo.com, praligns.com and mwm.ai for ChatGPT; supplementscored.com, cbssports.com and legalsportsreport.com for Perplexity — that no other tracked engine was recorded citing across the window, at citation counts (12 to 21 runs) too high to be single-observation noise. Claude, on the same gap-free footing, does not: 33 domains at 1.1%. That three-way comparison is the one this release can make. Where Gemini, Google AI Mode and Google AI Overviews sit is not something this release can settle.
What this changes about reading a "cited by only one engine" result
A brand or vendor page that shows up in the Index cited by a single engine is, on this evidence, in one of two very different situations, and the difference is citation volume, not which engine did the citing. Below 5 observed citations, single-engine presence is what the Index's own evidence floor predicts by default — most of the graph is still collecting, and a coin landing once tells you nothing about the coin.
At 5 or more citations, single-engine presence is a real, engine-specific pattern for ChatGPT and Perplexity, and close to nonexistent for Claude. So a domain that clears 5+ citations and shows up only on Claude is worth checking for a measurement or indexing problem before it is read as an engine preference. For the three gapped surfaces the same check is worth running for a different reason: on this release, a domain that looks exclusive to Gemini, Google AI Mode or Google AI Overviews, or exclusive to one engine because those three show no citation, may be an artifact of the collection gap rather than a pattern. Resolve any single domain against the Index's own record at machinerelations.ai/index/domains/<domain> before acting on it.
Method
Cited-domain records read live from the Index's public JSON dataset, contract_version machine_relations_index_public_view_v2.0, generated 2026-09-25, window 2026-05-10 through 2026-09-25 (132 days observed), 23,872 cited domains, release id mri_score_v2.0+2026-09-25+04fcb7fb8f29. Exclusivity share for each engine is (domains where engine_breadth.engine_count == 1 and the sole engine is E) / (domains where E appears in engine_breadth.engines), computed first unconditioned and then restricted to domains with runs_cited >= 5 and separately runs_cited >= 10. runs_cited and engine_breadth are read from each domain's mri_score_v2.overall block. The six engines are ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews, and Perplexity. Because engine_breadth records which engines were observed citing a domain, every exclusivity figure above is an upper bound, and figures for the three engines with collection gaps in this window are not comparable against the three without; see the correction above.
FAQ
Why does ChatGPT's unconditioned exclusivity share (56.2%) look close to a prior report of 55.5%? Both numbers come from the same instrument reading ChatGPT's overall citation set against the other five engines at different points as the Index's dataset has grown; neither was conditioned on citation volume, so both describe the same single-citation-dominated tail rather than a stable engine trait.
Does this mean cross-engine "citation overlap" studies published elsewhere are wrong? Not necessarily — it means any exclusivity or overlap statistic computed over a full domain population needs to state what share of that population sits at the evidence floor, and whether every engine in the comparison was collected without a gap, before the number is read as an engine behavior.
Which engines can be compared on this measure? The three the Index collected without a gap in this window: ChatGPT, Claude and Perplexity. Gemini, Google AI Mode and Google AI Overviews are tracked and their raw figures are shown, but their collection gaps mean their exclusivity cannot be ranked against the other three in either direction on this release.