Most pages cited in Google's AI Overviews do not rank in the organic top 10 for the query that produced them. Ahrefs checked 4 million AI Overview URLs across 863,000 SERPs in March 2026 and found 38% of citations came from top-10 organic results. That share measures rank position, and no public dataset breaks AI citations out by domain size; what it does show is citation coming loose from the ranked set, which is the opening a smaller site works in. The practical answer to whether a small site can out-cite a bigger one is yes, on specific questions, in specific engines, and never as a permanent property of the domain.
Topical authority is a description of a condition, not a score anyone computes. It means a site covers one question space densely enough, and with enough retrievable proof, that engines keep finding a usable passage there. Domain authority is a different object: a third-party estimate of backlink strength, produced by tool vendors who do not run the engines. The stakes are current: G2 found AI chatbots and review sites are now the two biggest influences on B2B shortlists, and 51% of software buyers now start research with an AI chatbot more often than with Google.
What is topical authority, and does any engine compute it?
No engine computes it. Google publishes no topical authority signal, and its documentation for AI Overviews and AI Mode states there are no additional requirements or special optimizations to appear, no new machine readable files or markup, and that pages must be indexed and snippet-eligible. The prerequisites are mechanical; the differentiator is what your page can prove about the question.
Domain authority scores a whole site rather than a subject, and the engines do not consume it. Google's helpful content guidance asks about the page in front of the reader, and its advice for succeeding in AI search repeats the instruction: answer the question, keep the page fetchable, keep it eligible to be quoted.
The useful reframe is scope. Domain authority is a property of a site. Topical ownership is a property of a question space that several sites compete inside, one question at a time. You can own the answers about HIPAA-compliant patient messaging workflows while owning nothing about healthcare software generally.
Does AI retrieval reward a bigger domain?
Sometimes, and how much depends on which index the engine borrows from. Seer's February 2025 analysis found 87% of SearchGPT citations matched Bing's top 20 organic results, most of them inside the top 10, across over 500 citations from 100 queries, so an engine leaning on a ranked index inherits whatever that index already rewards, including the accumulated link equity of incumbents. Google's AI Overviews point outside the top 10 far more often, per the Ahrefs figure above. The two readings use different depth thresholds and sit 13 months apart, so they bound the gap rather than measure it.
The second reason size matters less than it feels is decomposition. Google describes a query fan-out technique that breaks one question into several searches, and ChatGPT search issues one or more targeted queries of its own. Your competition is rarely the category head term. It is whichever sub-question the fan-out generated, and sub-questions are where thin coverage on a large domain loses.
Retrievability gates all of it: Vercel's crawler research with MERJ found that none of the OpenAI, Anthropic and Perplexity crawlers render JavaScript, and that the OpenAI and Anthropic crawlers fetch JavaScript files without executing them, so client-rendered content is an empty shell to all three, while Google's Gemini renders through Googlebot's infrastructure. Cloudflare's agent-readiness analysis of the 200,000 most visited domains found the agent-facing layer barely built, with 4% of sites declaring AI usage preferences in robots.txt. GPTBot's share of requests across the 30-plus crawlers Cloudflare tracks rose from 2.2% to 7.7% between May 2024 and May 2025.
Can a small site out-cite a bigger domain on the same question?
Yes, and the claim only survives if you measure it as a rate. No study in the public record measures win rates by domain size, so treat this as a reasoned inference to verify on your own questions rather than a measured result. SparkToro's 2,961 runs across 12 prompts put the odds of getting the same brand list twice at under 1 in 100, and Search Engine Land's write-up of the same work puts the odds of that list in the same order closer to 1 in 1,000.
A 2026 variance-components study reports run-to-run noise large enough to swamp real differences in small samples, and Ronald Sielinski's work on quantifying uncertainty in AI visibility puts confidence intervals around the same problem. Neither paper hands you a headline number to quote; both report a direction, which is why we treat several weeks of answers, across phrasings and engines, as the minimum honest test of whether you took the slot.
Two things decide whether the smaller site takes it. The first is resolution: a brand that names itself three ways across three pages is harder to resolve than an incumbent with a settled name, and harder for readers and engines to credit. Treat consistent naming as hygiene, not a retrieval lever: in a 2026 study of 443 entity-oriented retrieval configurations on a news benchmark, entity signals added at most 0.051 MAP once they could not peek at the answers. The second is prior association. A 2026 study of brand bias in LLM recommendations for skincare products found well-known brands were recommended every time when the products had the same specifications, and that the edge disappeared once a competitor had less than a +0.1-star rating advantage. Read that as a tiebreaker, not a moat: the prior matters most when nothing else separates you.
What does owning a topic cluster actually buy you?
Three concrete things, none of them a site-level score. A cluster raises the odds that a page exists for the sub-question a fan-out generates, it gives you somewhere to put evidence an answer can quote, and it forces the terminology consistency that lets an engine treat your brand as one entity across pages.
The evidence part has the strongest support behind it. The GEO study (Aggarwal et al., KDD 2024) found that adding statistics, quotations, and citations lifted citation visibility by roughly 30 to 40% in its benchmark, while the same error-bar table showed keyword stuffing did nothing; the KDD proceedings entry records the conditions tested. Page count matters only because it creates somewhere to put that evidence.
Consistency is the quieter half. Google says no new markup earns you a place in AI features, and markup still does other work. Structured data markup and the conventions in schema.org's getting started documentation give machines a stable description of what a page covers, which helps the resolution problem above. Placement matters after retrieval too: Liu et al. documented that where a passage sits inside a model's context changes how reliably the model uses it.
Where does cluster thinking waste your budget?
It wastes budget when the cluster becomes a publishing quota. The rewrite tactics that read as cluster hygiene have thin evidence behind them: C-SEO Bench tested conversational-SEO methods across two tasks and six domains and found most of them largely ineffective and often harmful to citation rank, and the NeurIPS 2025 record points toward traditional retrieval-side improvements as the more dependable work. Rewriting phrasing for the model is mostly noise, and the retest found little more from added statistics, citations and quotations. Fetchability is a gate you can verify; proof is the better bet because it serves the reader either way.
The second waste is assuming your cluster covers the whole battlefield. Profound's citation research found more than 4 in 10 AI citations point to sources brands do not control, so a complete cluster still leaves much of the cited surface off your domain. 6sense found the average buying-group member knows 75% of the vendors on the shortlist by the time it is filled, and nearly four out of five competitors are named to it on day one, mostly on the strength of prior experience, which no page on your site can supply.
The third is persuasion creep. Scarcity and exclusivity framing measurably reduces how often a model recommends a product, in work that ran the test across three sets of 10 fictitious products (coffee machines, cameras and books), so the marketing voice that lifts a landing page can cost you inside an answer. A citation and a visit are also separate outcomes: Ahrefs' December 2025 re-run measured a 58% lower average clickthrough rate for the top-ranking page when an AI Overview is present.
How does Trovance turn topic ownership into evidence you can check?
Trovance treats a topic as a set of tracked questions. You define the buyer questions that make up the territory you intend to own, and the system runs them repeatedly across AI engines as answer runs, preserving each answer snapshot with who was mentioned, who was cited, who was recommended, and which sources carried the answer. Share of answer then reads as a map of where you appear and where a competitor holds the answer.
That record turns cluster planning into a decision. When the snapshots show a competitor winning on their own page, the gap is evidence on your side of the line. When they show answers built entirely on review sites and community threads, no page you publish closes it, and the recommended action says so rather than filling a calendar. Your Brand Core holds the claims you are entitled to make and the proof behind each one, so drafts are produced from approved claims, and a person reviews and approves everything before it publishes.
Each analysis cycle reruns the same tracked questions after your asset goes live, so you can compare appearance rates on the sub-questions you targeted against the run-to-run variance the research above documents rather than one reading on one afternoon.
What Trovance will not do is promise that owning a cluster produces citations, rankings, or recommendations. The engines are probabilistic and your competitors publish too. There is no single AI-visibility score in the product, because the measurement evidence does not support one, and nothing ships without a person approving it. What you get is an observed record of where you appear today and a shorter list of things worth building next.
What should you do this week?
Start with the territory before the calendar. Write down the ten to fifteen questions a buyer asks inside the one topic you have a legitimate claim to, and ask each one in two or three phrasings on two engines, weekly, so you hold appearance rates. For every answer, note whether the cited page was yours, a competitor's, or a third party's.
Then fix in order. Check that your key pages return full content to a fetch with JavaScript disabled, because nothing downstream matters if the crawler sees an empty shell. Put at least one defensible number or quotable source on every page that answers a tracked question. Settle your naming and entity description so the same brand string appears everywhere, and only then start filling the gaps in the question space.
Retrieval fixes show up only when engines next fetch the page, evidence improvements follow re-crawling, and displaced third-party coverage takes sustained earned presence that no publishing schedule replaces. Semrush's AI Visibility Index analyzed 126 million AI search prompts to reach category-level readings, a useful reminder that your panel describes your questions and nothing wider. If you would rather have the tracked questions and the comparison running continuously, start a free Trovance trial and watch one topic instead of arguing about a domain score.
Build the territory
Topic clusters and AI authority - how cluster structure reads to a retrieval system.
Buyer-intent queries, end to end - where the question list for a territory comes from.
Can likely buyer questions improve AI retrieval? - the evidence for question-shaped coverage.
AI search authority on a startup budget - what to skip when the budget is small.
Check the evidence
Google rankings vs AI citations - why the two sets overlap less than people expect.
What gets cited by AI - the page traits that correlate with being quoted.
The most-quoted study in AI search, quoted correctly - what the GEO benchmark actually tested.
There is no such thing as an AI visibility score - why a site-level number hides the rungs.
