How to Rank in Google AI Overviews (and Actually Get Cited)

Which queries do Google AI Overviews actually cite?

Google AI Overviews cite sources heavily for informational and health-and-money questions and barely at all for local searches, which means citation is a realistic goal for some of your pages and a waste of effort for others. Our crawl of 500 queries, 100 in each of five intent categories, measured the citation rate for each engine on identical prompts.

Source: Novastacks, AI Search Is Three Channels, 500 queries, 6,083 unique AI-cited URLs, April 2026.

Google's AI answer for a local query mostly stays home: the Map Pack wins that slot, so a service-area business chasing an AIO citation on "plumber near me" is optimising for a surface that fires 10 times in 100. Google and ChatGPT are also not one channel. Across all 500 identical queries the two engines did not share a single citation, which is why our ChatGPT SEO playbook is a separate document rather than a section of this one.

What does a citable page look like structurally?

A page that wins both a Google top-10 position and an AI Overview citation is a denser document than a page that wins only one of them, and the density shows up in links and structure rather than word count. Of the 6,576 usable pages in our crawl, only 15% won both, and that group is measurably different.

Source: Novastacks, The Pages That Win Everywhere, 8,397 URLs crawled, 6,576 analysed.

Outbound citation density is easy to overlook. Fifty-six external links is roughly double what a single-system winner carries, and Seer Interactive's crawl of 6,354 AI-Overview-winning pages found definitional-query winners citing roughly twice as many .gov and .edu sources as pages that do not win (a figure we take from our own study write-up rather than Seer's report directly). Surfer's 57,000-URL analysis agrees from another angle: cited pages carry 38% more key facts, and pages with 10 or more discrete facts are cited at more than double the rate of pages with fewer than 5.

Density beats length. Ahrefs measured the correlation between word count and citation at a Spearman of roughly 0.04, which is effectively nothing.

Does schema get you into AI Overviews?

No, schema does not get you into AI Overviews, and the number that settles it is that 48.4% of Google's AI-cited pages carry any schema at all. That is the lowest adoption of the three engines we measured. It sits well below ChatGPT's 80.8%, though ChatGPT's figure rests on 182 cited pages against Google's 2,355, so read the ChatGPT column as directional rather than settled.

Source: Novastacks, AI Engine Rulebooks, 5,090 AI-cited pages, JSON-LD only. Per-engine sample: Google AI Overview 2,355 pages, Perplexity 2,870, ChatGPT 182.

Google's own documentation agrees. Its AI optimization guide states plainly that "structured data isn't required for generative AI search, and there's no special schema.org markup you need to add," and that you "don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search." That last clause covers llms.txt, which several competing guides still present as a headline tactic.

Do it anyway, for the right reason. Schema earns classic rich results, keeps your entity legible, and costs an afternoon. What it will not do is carry a thin page into an AI answer. Read the 80.8% as a description of which pages happen to get cited, not as an instruction that produces citations.

What actually works, and what is worth the effort?

Eight signals survive contact with the data, and most of them are Google's own published guidance translated into something checkable. The table below is the working version we run, scored by the effort a team has to spend against the return we see.

Fan-out is the mechanism most rival guides leave out. Google expands a query into roughly five sub-queries and grounds its answer across all of them. Ranking for the main query plus at least one sub-query makes a page 161% more likely to be cited, at a Spearman correlation of 0.77, and only 19.6% of top-3 AIO citations rank for the main query alone. One caveat: Ahrefs and Surfer both publish that figure, and Ahrefs credits the underlying research to Joshua Hardwick working alongside Surfer, so it is one study reported twice, not two confirmations.

Brand mentions look like a stronger correlate than backlinks, on a narrow sample. Ahrefs analysed the top 50 websites mentioned across 76.7 million AI Overviews and found branded web mentions correlating at 0.664 with AI Overview visibility, against 0.326 for domain rating and 0.218 for backlinks, with YouTube strongest at 0.740. Fifty websites is a thin cut, so treat that ordering as a strong hint rather than settled law. It does track with where citations come from: Surfer attributes roughly 23.3% of AIO sources to YouTube, 18.4% to Wikipedia and 16.4% to Google's own properties.

Write the sentence so it can be lifted. The test is whether a single paragraph, removed from the page and read alone, still answers the question and names its subject.

• Weak, because it depends on its neighbours "As we discussed above, this can vary quite a bit depending on the factors involved, though most teams find that it improves over time."
• Strong, because it survives extraction "Google AI Overviews cite sources on 96% of informational queries and 10% of local ones, per Novastacks' April 2026 crawl of 500 queries. Local pages should be measured on Map Pack presence instead."

The second version carries the number, the source, the timeframe and the consequence, and works alone.

Entity clarity is the quiet prerequisite. Google's AI builds a category-level model of what your brand is before deciding whether to name you in an answer about that category. If your homepage, Organization schema, LinkedIn and third-party listings describe the company differently, the model has nothing stable to attach a citation to. Fix the naming everywhere before rewriting a single article.

Where do the studies disagree, and what should you believe?

The studies disagree most on how much AI Overview citation depends on already ranking, and the disagreement is large enough that averaging the numbers would produce a figure nobody measured. Here they are side by side.

Those four are not measuring the same thing, which is why the direction of travel looks contradictory. SE Ranking's 92.36% counts whether any cited domain also ranks top-10, a much easier bar than Ahrefs asking whether this cited page ranks. BrightEdge's number counts citations ranking at any organic position, not the top 10; its own breakdown puts just 16.7% of citations in top-10 results, with most growth coming from positions 21 to 100. And Ahrefs improved its citation parsing between the two studies, which limits comparability, while its own stated reading of the shift is that AI Overviews now lean more on fan-out query results than on the original SERP.

The practical conclusion holds across all four. Ranking well makes citation substantially more likely and it is nowhere near sufficient. On Ahrefs' newer numbers, 31.2% of citations come from pages ranking 11 to 100 and 31.0% from pages not in the top 100 at all. Build for extraction, not only for position.

Freshness has a cleaner answer. SE Ranking found only 12.32% of AIO citations come from content published within the previous 30 days, against 28.76% from 2025 content and 26.85% from 2024, so AI Overviews favour established rather than brand-new material. Update pages when the facts change and keep dateModified honest.

On content length the field is split and our data breaks the tie. Ahrefs argues for concision, citing Dan Petrovic's finding that grounding plateaus around 540 words and effectiveness declines past 2,000; the long-guide camp argues the opposite. Our crawl says the variable separating winners is structural density, not word count. Write until the question is fully answered, structure it so any part can be lifted, and stop.

How do you measure AI Overview presence in Search Console?

Google does not give AI Overviews their own filter in Search Console, so you measure them by inference from the Performance report rather than by reading a number off a dashboard. Three reads do most of the work.

Watch the impressions-to-clicks gap on your question queries. Filter Performance to queries containing "how", "what", "why" or "best", then compare impressions against clicks over a six-month window. Since AI Overviews trigger on 57.9% of question queries, a page holding impressions while its click share erodes is the standard signature of an answer being lifted above it. Position holding steady while CTR falls is the same signal from another angle.

Split AI Mode traffic out before reading anything. AI Mode queries land in the same Performance export as classic search and skew the averages, so separate them first. Those exports also carry bare fragment queries and prompt-shaped strings from monitoring tools; strip those or you will read your own tracking as buyer demand.

Track what Search Console cannot see. Search Console reports impressions and clicks, never whether you were named in an answer. Pair it with prompt-level citation tracking on the queries that matter commercially, and with the engagement metrics Google itself points at: time on site and conversions per session from organic, not CTR. If citations are working, branded query volume rises before non-brand clicks do. That leading indicator is the early read, not the finish line, and it is the one we report against. More on the full measurement stack in measuring AEO.

How Novastacks approaches this

We run this as a named method, Source & Echo: Source is the work of becoming the thing engines retrieve, which is the entity clarity, fact density and passage construction described above. Echo is the work of being repeated elsewhere, the independent mentions and third-party coverage that the correlation data keeps pointing at. Programmes that do only one wonder why citations do not follow.

• AEO & SEO Solutions We run the crawl, intent segmentation and passage-level rebuild described above as a delivery programme, and report on citation share rather than deliverables shipped.
• Fractional Growth Partner Senior principals hold strategy, execution and reporting for the life of the engagement, with no account-management layer between the work and the decision.

FAQ: Ranking and Citation in Google AI Overviews