Information-gain assets to add before publish:
- Your own citation audit: which of your eight live posts get pulled into AI Overviews, which do not, and screenshots of both.
- Search Console AI-mode impression data for bijayworks.com, annotated.
- One worked example: a query where a competitor page is cited and yours is not, with your read on why.
The short answer
Google states there are no special optimisations for AI Overviews. The independent citation data agrees with that in an uncomfortable way: what gets cited is deep, specific pages that answer sub-questions, and roughly two thirds of cited pages do not rank in the top ten for the query at all.
What Google actually says is required
Google’s documentation on AI features in Search is unusually direct. There are no additional requirements to appear in AI Overviews or AI Mode, and no special optimisations are necessary. The only technical requirement is that the page is indexed and eligible to be shown with a snippet.
The same page says explicitly that you do not need to create new machine-readable files, AI text files, or markup to appear in these features. That single sentence invalidates a large share of what is currently being sold as generative engine optimisation.
What Google does recommend is the same list it has recommended for a decade: allow crawling, link internally, keep important content in text, make sure structured data matches what is visible, and give the page a decent experience.
What the citation data shows
Two independent studies published in 2026 give a reasonably consistent picture. Neither is peer-reviewed and both come from SEO vendors, so treat the numbers as directional rather than exact. They still point the same way.
| Finding | Figure | Source |
|---|---|---|
| Citations pointing to deep pages, two or more clicks from the homepage | 82.5% | BrightEdge |
| Citations pointing to homepages | 0.5% | BrightEdge |
| Cited pages that rank for the main query only | under 20% | Surfer SEO |
| Cited pages ranking for fan-out sub-queries are more likely to be cited | 161% more likely | Surfer SEO |
| Cited pages not ranking top ten for either the main query or a fan-out | about 68% | Surfer SEO |
Deep pages get cited, homepages do not
BrightEdge’s analysis of millions of cited URLs found 82.5% of citations went to pages sitting two or more clicks from the homepage, and only 0.5% to homepages. If your strategy is to make the homepage rank for everything, AI Overviews will not cite you. The unit of citation is a specific page answering a specific thing.
Ranking for the sub-questions beats ranking for the headline query
Surfer SEO looked at 10,000 keywords and 173,902 URLs, extracting roughly 33,000 fan-out queries. Pages ranking for those fan-out queries were 161% more likely to be cited than pages ranking only for the main query. Fan-out-only rankings produced about 30% of citations; main-query-only rankings produced about 20%.
The practical reading: an AI Overview is assembled from answers to several sub-questions, not from one page that ranks first. Cover the sub-questions properly and you enter the pool for several of them.
Most cited pages are not in the top ten
The same study found roughly 68% of cited pages did not rank in the top ten for either the main query or a fan-out query. This is the finding that should change how you think about the work. Classic ranking and citation eligibility overlap, but they are not the same system, and a page that is invisible on page one can still be the source Google quotes.
What this means for how you build a page
- Write to sub-questions, not to a keyword. Pull the People Also Ask set and the related searches, and give each real question its own heading with a direct answer beneath it.
- Answer before you elaborate. A 40 to 60 word answer immediately under the question heading, then the nuance. An extractable answer buried in paragraph six is not extractable.
- Keep the page deep in the site, not on the homepage. The citation data is clear about where citations land.
- Use tables for comparisons and numbered lists for processes. Structure that a machine can parse is structure a reader can scan. The two goals are aligned here, which is rare.
- Say something the other results do not. Every cited page is competing against nine near-identical summaries of the same consensus.
What does not work
- llms.txt. Google’s own documentation says no new machine-readable or AI text files are needed. There is no evidence it influences citation.
- Content chunking as a tactic. Writing in artificially short blocks does not create eligibility. Clear structure does; performative fragmentation does not.
- Stuffing a page with thirty FAQ questions. Relevance and quality decide extraction. Volume of questions does not.
- Manufacturing brand mentions to influence models. This is the 2026 version of buying links, and it will age the same way.
A caution about the numbers circulating
One claim doing the rounds is that guides over 2,000 words receive three times more AI Overview citations than shorter pages. The BrightEdge research usually cited for this does not measure word count at all. It measures page depth. Length and depth get conflated constantly, and only one of them is in the data.
This matters because writing to a word count is the exact behaviour Google warns against. Depth of coverage is the variable. Length is a side effect of covering something properly.
How to measure whether you are being cited
- Check Search Console for impressions on queries where an AI Overview appears. Google surfaces AI Search data in performance reporting.
- Run your ten most important queries manually, in an incognito window, and record whether an AI Overview appears and who is cited.
- Log the cited domains and pages, not just whether you appear. The pattern of who gets cited tells you what shape of page qualifies.
- Repeat monthly. A single snapshot tells you almost nothing, because AI Overview presence for a given query is unstable.
Frequently asked questions
Does adding llms.txt help my site appear in AI Overviews?
No. Google’s documentation on AI features states you do not need to create new machine-readable files, AI text files, or markup to appear in these features. Sites publishing llms.txt have not demonstrated citation gains attributable to the file itself.
Do I need to rank on page one to be cited?
No. In Surfer SEO’s analysis of 173,902 URLs, roughly 68% of cited pages did not rank in the top ten for either the main query or a related fan-out query. Ranking helps, but citation eligibility is a partly separate system.
Does FAQ schema get me into AI Overviews or People Also Ask?
Structured data helps Google understand a page, and Google requires that it match visible content. It does not purchase placement. Treat schema as a description of what is already on the page, not as a lever for visibility.
Should I write longer articles to get cited more?
Write until the question is fully answered, then stop. The citation data measures page depth within a site, not word count. Google’s own guidance warns specifically against writing to a target length because someone claims that length ranks better.
How often do AI Overviews change for the same query?
Frequently. AI Overview presence and the set of cited sources both fluctuate, which is why a single manual check is unreliable. Track the same query set monthly and read the trend rather than any individual result.
Sources
- AI features and your website – Google Search Central
- Spam policies for Google web search – Google Search Central
- Ranking for AI Overview fan-out queries boosts citation odds – Search Engine Land, reporting Surfer SEO research
- Google AI Overviews cite deep pages – Search Engine Land, reporting BrightEdge research