1,000 AI Overviews Analyzed: Citation Pattern Study

Google's AI Overviews are now the dominant SERP feature for informational and definitional queries in the US — and unlike the ten blue links era, the only traffic they hand out is the citation. Either you are in the citation list, or your visibility for that query is effectively zero. We analyzed 1,000 AI Overviews across 10 query intents to figure out exactly which sites get into that list, how the slots are allocated, and what an editorial team can change this quarter to improve their odds.

The dataset was collected over two weeks in April 2026 — 100 queries per intent class, sampled from US-English desktop search, with the full citation panel captured for every AIO that rendered. Where AIOs failed to render (about 11% of the original sample) the query was replaced with a same-intent alternate so the per-intent counts remained even. Page-level signals — domain authority, content length, schema markup, age — were joined from a separate crawl of the 4,200 unique URLs that ended up in the citation lists.

The headline finding is concentration. The top 1% of cited domains — roughly 12 sites including Wikipedia, Reddit, Forbes, Healthline, Investopedia, NYT, the larger gov and edu domains — capture 47% of all citations. The next 9% capture another 31%. Everything else shares the remaining 22%. If your agentic SEO program is built on the assumption that AIO citations are evenly distributed, the dataset says it is not — and the corollary is that the page signals that get a non-top-1% domain into the list are worth understanding precisely.

The mechanical change is simple. When an AIO renders for a query, the click-through rate to the ten blue links collapses — third-party CTR studies through Q1 2026 put the drop at 30–60% depending on vertical, with informational and definitional queries hit hardest. What replaces that traffic is not a click on the AIO panel itself (those are rare) but the citation links inside it: a small number of attributed sources, typically 3–5, that the model has chosen to cite inline with its answer.

Those citation slots are now the only meaningful traffic vector for a large class of queries. If your page is in the citation list it picks up a high-intent click; if it is not, the query is invisible to your site regardless of where you rank in the unrendered blue links. That is why understanding citation pattern — which sites Google chooses, why, and what page-level signals correlate with inclusion — has become the defining question of 2026 SEO. The answer is not what the conventional ranking-factors lists predict.

We constructed the query set by taking 100 queries per intent class across 10 classes — informational, commercial, navigational, comparison, how-to, definitional, statistical, troubleshooting, review, and transactional. Within each intent we sampled across ~30 verticals (health, finance, B2B SaaS, consumer electronics, travel, home, legal, education, food, automotive, plus a long-tail of smaller categories) so no single industry would dominate the citation distribution.

Queries were issued from US-English desktop search between April 8 and April 22, 2026, in private sessions with rotated IPs to control for personalization. Every AIO that rendered was captured in full — answer text, citation panel, and the rendered URL of every cited source. Of the original 1,123 queries issued, 1,000 produced a renderable AIO (the remaining 123 were replaced with same-intent alternates so the per-intent count remained at exactly 100).

The 4,243 unique URLs that ended up in citation lists were then crawled for page-level signals: word count, schema markup type and presence, last-modified date, named-source citation count, table presence, HTTPS, and a domain-authority proxy from a third-party backlink dataset. The cited-vs-not analysis compares the 4,243 cited URLs against a control set of the next 50 organic results for the same query (~50,000 URLs total) on the same signals.

The most striking finding in the dataset is concentration. Of the ~1,200 unique domains that appeared at least once in a citation list, roughly 12 — the top 1% — accounted for 47% of all citations. The next 9% (about 100 domains) accounted for another 31%. The long-tail 90% of cited domains together share the remaining 22%. That is a more concentrated distribution than the underlying organic SERP, where the top 1% of domains typically capture closer to 25–30% of position-1 listings.

The composition of the top group is what you would expect: Wikipedia and Reddit lead by a wide margin (each appearing in roughly a quarter of all AIOs), followed by a cluster of large editorial publications (Forbes, NYT, Healthline, Investopedia) and a long tail of high-authority gov and edu domains that win specific verticals (CDC and NIH for health, IRS for tax, .edu for academic queries). Notably, the top-cited “agency-style” outlets — Forbes, HubSpot, Moz, Ahrefs, Search Engine Journal — are present in the second tier and represent the citation ceiling that commercial editorial sites can realistically aim for.

The practical read is that the long-tail 22–25% of citations is the realistic battleground for a non-top-1% editorial site. Inside that slice, the page-level signals discussed in section 05 — schema, DA, length, named sources — have a much larger relative effect than they do in the highly-competitive top tier where domain reputation already saturates the model's confidence.

Across the full 1,000-AIO sample, the average citation count is 4.2, with a range of 2 to 9 and a median of 4. Only 8% of AIOs cite more than 7 domains, and only 4% cite fewer than 3. The distribution is tight enough that “design for a 4-citation list” is a reasonable editorial assumption. What changes sharply is the per-intent breakdown — Google is meaningfully more conservative on commercial-intent queries (where the citation list is shorter and skews to credentialed publications) and more generous on definitional and how-to queries (where the model pulls from a wider source pool to construct the answer).

The intent breakdown changes how an editorial roadmap should allocate effort. Definitional and how-to queries are the highest- opportunity intent classes for a non-top-1% site — citation lists are longer, the source pool is wider, and the page-level signals (especially schema and structure) carry more weight relative to pure domain authority. Commercial intents are the hardest; attempting to win citation share against Wirecutter and Consumer Reports on “best X” queries is rarely the right roadmap call for an emerging editorial brand.

With domain effects partially controlled by the within-query baseline, four page-level signals stood out as having the largest independent effect on citation rate: domain authority, schema markup presence, content length, and the presence of named-source citations in the body of the page. Each is a lever an editorial team can pull this quarter — and each compounds with the others.

The dataset turns into a four-stage editorial workflow that any in-house team or agentic SEO program can ship in a quarter. The order matters — schema and structure first because they are the cheapest lifts, then length and credibility, then programmatic citation tracking, then domain-authority work last because it is the longest-cycle.

The mistake we see most often is teams attempting stage 4 first — spending a quarter on PR and outreach for marginal DA gains while their editorial pages remain unstructured and under-sourced. The order in this playbook is deliberate: schema and editorial structure first because the multipliers are larger and ship in weeks; tracking second so the gains are visible; DA last because the gains are real but slow and the leverage is lower without the page-level work in place.

For multi-brand portfolios the workflow is the same, but the sequencing changes — start the schema rollout on the brand with the highest existing DA, because the same 2.3× multiplier applied to a higher base produces the largest absolute citation count increase. Use the slower brands as a test bed for the editorial restructure work in stage 2 before rolling it across the portfolio.