Gemini Citation Change Log: At Least 452 of the Top 500 Domains, Recorded Six Sources Deep

This log is a dated record of what Gemini cites and what changes, read from the Machine Relations Index and AuthorityTech's own daily panel. Entry one: Gemini reaches at least 452 of the 500 most-cited domains in AI answers and cites the single most-cited source in at least 86 of the 93 published question segments. Both are floors rather than measurements, because the Index's own Gemini collection has dated blind days inside this window — see the correction of September 26, 2026 below, which withdraws this entry's original ranking of Gemini against Perplexity, Claude and the two Google surfaces and states what survives. The panel that tracks it day to day was recording a mean of 6 sources per answer against the roughly 10.8 the engine actually returns, disclosed here before any churn number built on it.

Every entry states the measurement, the release or window it came from, the primary source, and what a publisher can check. Entries are appended; nothing is rewritten. Entry one draws on the Machine Relations Index, which observes six engines — Perplexity, ChatGPT, Claude, Gemini, Google AI Mode and Google AI Overviews — against a fixed basket of buying and research questions and publishes a machine-readable release daily. Entry one also covers a methodological finding about AuthorityTech's own daily panel that has to be on the record before this log uses that panel for anything: the panel's citation log keeps only the first six sources per answer, and after Perplexity, Gemini is the engine it caught worst.

Correction, September 26, 2026 — this log's subject is itself a gapped arm

The September 25 correction below named two engines with collection gaps. There are three. Gemini — the engine this log is about — has 22 blind days inside the scoring window (May 11–15, then scattered single days from August 11 to September 16), carrying 442 eligible successful runs that produced no citations. The shape is why it went unnoticed for nine sweeps: intermittent single days rather than a continuous outage, so anyone hunting an outage found none. Three arms collected without a gap — ChatGPT, Claude and Perplexity — and three are gapped: Gemini, Google AI Mode and Google AI Overviews. (An earlier form of this note, posted the same day, named only ChatGPT and Perplexity. Claude's single flagged date carried one answer run and was an artifact of a zero-citation scan with no run floor, not an outage.)

Every Gemini figure on this page is a floor, not a measurement. A presence count only accumulates, so a blind day can subtract entries from Gemini's count and never add them: the true numbers are at least these and may be higher.

What is withdrawn. This entry's headline verdict — that Gemini is the second-widest engine, the closest follower to Perplexity, second on head overlap and segment leaders — is withdrawn. A floor cannot be placed below an exactly measured engine, because the floor can rise past it. So Gemini against Perplexity, Gemini against Claude's head overlap, and Gemini against either Google surface are all unresolved, and the last of those is a floor compared against a floor in both directions.

What survives, and it is most of the page's usefulness. A floor sitting above an exact count is still a real gap, because it can only widen. Gemini's floor of 452 top-500 domains sits above ChatGPT's exact 361; its 86 segment leaders sit above Claude's exact 72 and ChatGPT's 66; its 15 community platforms sit above Claude's 10 and ChatGPT's 8; its 7,118 distinct domains sit above Claude's 5,326 and ChatGPT's 5,054. Gemini is above ChatGPT on every composition measure this log tracks, and above Claude on all but head overlap. That is the finding a publisher can act on.

Correction, September 25, 2026

Every ranking on this page that places an engine above Google AI Mode or Google AI Overviews is withdrawn. The Index's own collection recorded zero citations from Google AI Overviews on any monitored query between August 15 and September 24, 2026 — 41 of the last 132 days — and zero from Google AI Mode between September 14 and September 24. The cause was a defect in the Index's Google collection path, fixed September 25, 2026: the prior collector graded a response as a success whether or not it carried content.

The direction is the whole of this correction. A presence count and a distinct-domain count only accumulate, so missing collection days can remove entries from a Google surface's count but never add them. Every Google AI Mode and Google AI Overviews figure on this page is therefore a floor: the true count is at least that and may be higher. So a claim that another engine sits below a Google surface still holds, because the gap can only widen that distance — while a claim that another engine sits above one does not, because the floor can rise past it. This correction originally said rankings among ChatGPT, Claude, Gemini and Perplexity were unaffected because all four collected continuously. That sentence was wrong about Gemini and is superseded by the September 26 correction above.

Entry 1 — September 22, 2026: the baseline

Release mri_score_v2.0, contract machine_relations_index_public_view_v2.0, generated September 22, 2026. Window May 10 to September 22, 129 days, 16,307 observed answer runs, 34,720 cited-domain observations, 22,828 cited domains. A subject category paired with a question type forms a stratum, and a stratum publishes a rate only after clearing an evidence floor of 10 observed runs across 7 distinct dates; 93 of the 151 measurable strata are published today.

The Index records, for each domain, which engines cited it at least once in the window. It does not publish a per-engine citation rate, so everything below is presence and counts, not per-engine rates — the same boundary this log's companion ChatGPT, Perplexity and Claude entries run under.

Head overlap: a floor of 452, above ChatGPT's exact 361

Engine	Top 25	Top 100	Top 250	Top 500
Perplexity	25	96	244	490
Claude	23	94	237	463
Gemini	23	89	227	452
Google AI Mode	24	91	227	444
Google AI Overviews	24	89	203	378
ChatGPT	22	88	203	361

Gemini is recorded as missing 48 of the top 500 most-cited domains in the market — 11 of the top 100, 23 of the top 250. Read those as ceilings: Gemini's blind days can only manufacture an absence, never hide a presence, so it misses at most 48. Only ChatGPT's and Perplexity's rows in this table are exact. Gemini's floor of 452 sits above ChatGPT's exact 361, and that distance can only widen. It sits below Perplexity's exact 490 and Claude's 463, but a floor below an exact count settles nothing — a fully collected window could carry Gemini past either. Gemini against Google AI Mode at 444 and Google AI Overviews at 378 is a floor against a floor, and is not a ranking in either direction.

Segment leaders: at least 86 of 93, above Claude and ChatGPT

For each of the 93 published strata, take the domain with the highest citation rate in that stratum — the source the engines cite most when buyers ask that category's question — then ask which engines cite that domain anywhere in the window.

Engine	Cites the leading source, of 93 segments
Perplexity	93
Gemini	86
Google AI Mode	81
Google AI Overviews	79
Claude	72
ChatGPT	66

Gemini is recorded as missing the segment leader in at most 7 of 93 published categories. Combined with the head-overlap table, what both published composition facts support is the same narrower claim: Gemini's floor sits above ChatGPT on both measures and above Claude on this one, by margins wide enough that no amount of recovered collection closes them. What they no longer support is a place for Gemini in a six-engine order. Perplexity's 93 is exact and Gemini's 86 is a floor beneath it, which resolves nothing; the two Google surfaces are floors on both sides of it.

Domain count, stated without an exclusivity claim

Gemini cites at least 7,118 of the 22,828 domains observed in the window — more than Claude's exact 5,326 and ChatGPT's exact 5,054, and both of those gaps can only widen. It is not ranked against Perplexity's 9,986: a floor below an exact count does not establish an order. Google AI Mode's 7,441 and Google AI Overviews' 2,279 are floors too, so Gemini is not ranked against either in either direction. This log's ChatGPT and Perplexity entries each carry a floor-adjusted exclusive-domain count for their own engine, computed against the specific release each entry used. That figure is not restated here for Gemini: the floor-adjusted count has to be computed against this same September 22 release to be comparable to the numbers above, and that computation has not been done. An unconditioned exclusive-share number would repeat the mistake this log's ChatGPT entry already corrected once — most of an unconditioned count's weight sits on domains cited a single time — so this entry states domain count and stops there rather than publish a number this log would have to correct later.

Community platforms: a floor of 15 of 27, above Claude and ChatGPT

Twenty-seven classified domains in the Index are community and social platforms — a small set where engines diverge most.

Engine	Community platforms cited (of 27)
Google AI Mode	20
Perplexity	18
Gemini	15
Claude	10
Google AI Overviews	10
ChatGPT	8

Gemini's floor of 15 sits above Claude's exact 10 and ChatGPT's exact 8, the same direction the two composition tables above support. Google AI Mode's 20 and Google AI Overviews' 10 are floors, and Gemini is not ranked against either; nor is it ranked against Perplexity's exact 18, which sits above Gemini's floor and therefore settles nothing.

The one robots.txt token that governs eligibility, and what it does not cover

Unlike ChatGPT, Perplexity or Claude, Gemini has no dedicated fetch-time citation crawler for a publisher to name in robots.txt. Google's common-crawlers documentation, last updated July 14, 2026, lists eleven Google crawler and fetcher tokens; the only one that names Gemini is Google-Extended, which the same page states "doesn't have a separate HTTP request user agent string" — crawling happens under Google's existing user-agent strings, and the robots.txt token exists purely "in a control capacity." A publisher cannot see it arrive in server logs because it never does.

Google's documentation states plainly that Google-Extended governs two things together, not one: whether crawled content may be used to train future Gemini models, and whether it may be used for grounding — "providing content from the Google Search index to the model at prompt time" — in Gemini Apps and in Grounding with Google Search on Vertex AI. Grounding is the mechanism that produces a citation. A publisher-facing explainer of the token puts the practical stakes plainly: "the cost of disallowing it is losing grounding citations in Gemini rather than losing rankings," and the same source notes there is no way in robots.txt to allow grounding while blocking training, or the reverse — one token, one directive, both uses. A second explainer reaches the same reading of Google's own page: there is no Gemini-named crawler to permit, only a training-and-grounding control token that a publisher either allows or disallows in full. Separately, a robots.txt reference for AI crawlers notes that disallowing Google-Extended does not remove a site from Google AI Overviews, which run on the regular Googlebot index rather than this token — a distinction worth stating because the two Google surfaces are easy to conflate and this log tracks Gemini, not AI Overviews.

This is a structural difference from the other engines in this series, not a finding about any specific domain's absence. Where the ChatGPT and Perplexity entries could name individual blocked or permitted domains and check each one's robots.txt, Gemini's single, log-invisible token means that check cannot be run the same way, and this entry does not attempt it.

Grounding also runs on content Google reaches through means other than crawling. Reuters reported in February 2024 that Google struck a roughly $60-million-a-year licensing agreement giving it access to Reddit's Data API, separate from and in addition to whatever Google-Extended permits. Community-platform reach in the table above is not explained by crawler permission alone.

What the daily panel was actually recording

AuthorityTech runs its own tracked panel alongside the Index: 35 fixed queries, five retrieval surfaces, once a day, the same instrument behind this log's ChatGPT and Perplexity entries and the site's 89-day per-engine churn measurement. That page documents that the daily log stores the first six sources returned by each engine, a cap that affects engines unequally. After Perplexity, it affects Gemini most of the five.

Measured against the stored per-run answer artifacts, which preserve each provider's complete returned source list — available from September 1, 2026 onward, not for the full panel history — Gemini returns a mean of 10.8 source URLs resolving to 8.8 distinct domains per answer. The daily log recorded a maximum of six. Gemini's answers exceeded six sources on 175 of 193 runs in that window — 90.7% truncated — against Perplexity's 350 of 350 (100%), Claude's 16%, ChatGPT's 12%, and Google AI Mode's 0%.

The practical consequence, recomputed rather than assumed small: day-over-day domain churn on Gemini's complete source lists for the September 1–22 window is 42.8%, against 46.0% on the panel's six-source cap — a difference of 3.2 percentage points, in the direction the six-slot depth predicts, with the cross-engine churn ranking unchanged. The full recomputation table lives on the churn page linked above; this log will not restate the panel's older 89-day retention, drop-out, and core-domain figures as if they described the complete list, because the complete-list artifacts do not reach back past September 1 and the churn page says so explicitly. Any figure in this log drawn from the panel is labelled by which list produced it.

Where other measurements land, and why they stay off this log

Third-party trackers publish their own Gemini citation-depth numbers, on their own prompt baskets, and they do not agree with each other or with this log's 10.8. Profound reports 6.6 sources per answer against Google AI Mode's 15.2. GetMint's three-engine study of 1,152 answers reports Gemini citing 46 instances and 10 unique domains per answer, driven heavily by Reddit and YouTube. SoRank's citation study puts Gemini at 11.1 sources per answer, the highest of five models it tracked, rising from 4.1 to 7.2 over its measurement period. A 1,250-prompt Indonesian-market study found Gemini's Search grounding fires on only 34.2% of consumer prompts at all, and cites 7.2 sources per answer when it does. A fifth source states a 4-to-8 range and separately confirms the Google-Extended grounding dependency above.

Six figures — 6.6, 7.2, 10.8, 10, 11.1, and a 4-to-8 range — from six instruments measuring six different prompt baskets over six different windows. None of them is wrong on its own terms and none of them is comparable to this log's 10.8 without knowing the query basket behind it. This log's number comes from a fixed panel of 35 buying and research queries measured daily since September 1, checked against the complete per-run artifact rather than a display-layer or third-party scrape; it is stated as what it is, not as an industry consensus figure, because no such consensus exists in the public data this entry could locate.

How to read this log

Each future entry compares a release or a panel window against the one before it and reports what moved. Four numbers are the ones to watch:

• Head overlap. At least 227 of the top 250, at least 452 of the top 500.
• Segment leaders cited. At least 86 of 93.
• Community platforms. At least 15 of 27.
• Panel-recorded source depth. 8.8 distinct domains per answer on the complete list, against a six-item cap in the daily log; watch for the cap narrowing as the answer-evidence store's history lengthens past September 1.

A change in any of them is an entry. A release with none is recorded as no change.

What a publisher should take from entry one

Gemini is wider than ChatGPT on every composition measure this log tracks — head overlap, segment leaders, community reach, distinct domains — and wider than Claude on three of the four. Those are the claims a floor can carry, and they are the ones that matter to a brand deciding where to spend: treating Gemini as an afterthought behind ChatGPT is not supported by anything in this release.

Where Gemini sits in a six-engine order is a question this entry cannot answer and no longer pretends to. Three of the six arms — Gemini, Google AI Mode, Google AI Overviews — have dated collection gaps in this window, so any rank that places one of them against another, or places one of them below an exactly measured engine, is an artifact of the instrument rather than a property of the engine.

The measurement finding is about recording depth rather than the engine. A dashboard, a competitor tracker, or an internal panel that caps a citation array at six entries per answer will report close to the true count for Google AI Mode, understate ChatGPT and Claude modestly, and understate Gemini by nearly half — 8.8 true domains against a 6-item cap, truncated on 9 of every 10 runs. AuthorityTech's panel quantifies the effect by checking the complete artifact against the six-item log, which is how the depth figures above were measured.

Watch list for entry 2

• Whether Gemini's head-overlap or segment-leader numbers move at the next Index release — the release updates daily, so entry 2 will report on genuine movement, not a fixed weekly cadence.
• Whether a floor-adjusted exclusive-domain count for Gemini gets computed against a specific release, making that comparison publishable without the caveat in entry one.
• Whether the complete per-run artifact history extends further back than September 1, letting the panel restate churn over a longer window than 22 days.
• Whether any additional top-500 domain moves into or out of Gemini's cited set.

Frequently asked questions

Is Gemini the second-strongest engine on every measure in this series? This entry originally said it was the nearest follower to Perplexity, and that claim is withdrawn — see the September 26, 2026 correction. Gemini has dated collection gaps in this window, so every Gemini figure here is a floor, and a floor cannot be ranked below an exactly measured engine. What the data supports is that Gemini is wider than ChatGPT on all four composition measures and wider than Claude on three, with each of those distances able only to grow. Its position relative to Perplexity, to Claude's head overlap, and to either Google surface is unresolved.

Why does this entry not publish an exclusive-domain-share number for Gemini the way the ChatGPT and Perplexity entries did for their own engines? Because that number has to be computed against the same release as the rest of this entry's figures to be comparable, and that computation was not available when this entry was written. Publishing an unconditioned exclusive-share figure now would repeat the error this log's ChatGPT entry already corrected once — the raw count is dominated by domains cited only once — so this entry states total domain count and leaves the floor-adjusted figure for a later entry.

Is the six-source cap a property of Gemini or of the measurement? A property of AuthorityTech's daily panel, not of the engine and not of the Machine Relations Index — the Index used for entry one's composition figures is not affected by this cap. It is stated here, before this log uses the daily panel for anything, so that every churn or stability number in a future entry is read at the right depth.