How to get cited by AI Overviews: beyond traditional SEO rankings
Last updated 1 October 2026
Guide
AI Overviews pick sources passage by passage and favour claims the wider web agrees with. Ranking gets you into the pool of eligible pages; it is not what decides who gets cited.
To get cited in Google AI Overviews, your page must be indexed and eligible for snippets, contain self-contained passages that answer specific sub-questions, and be corroborated by other sources on the web, especially forums like Reddit. A top ranking helps, but it is not the mechanism that decides who gets cited.
The key message: SEO gets you into the pool of eligible pages, but AI Overviews pick sources passage by passage and favour claims the wider web agrees with. Optimising only for your position on one keyword misses how the citation decision is made.
Key terms
- AI Overviews — the generated summaries Google shows above organic results for many queries, with links to supporting sources.
- Generative engine optimisation (GEO) / answer engine optimisation (AEO) — two names for the practice of making content easy for AI systems to retrieve, quote and cite. For a fuller definition, see what GEO is and how it differs from SEO.
- Query fan-out — Google's term for how its AI search features break one question into several related searches and pull sources for each. Google described this when it introduced AI Mode.
- Citation vs mention — a citation links to your page as a source. A mention names your brand in the answer text without necessarily linking. You want both, and they respond to slightly different signals.
Why ranking first is not enough
The old model was linear: pick a keyword, rank for it, collect the click. AI Overviews break that model in two ways.
First, the answer is assembled from multiple retrievals. Because of query fan-out, a single search like "best way to onboard remote employees" can trigger sub-searches on checklists, tools, legal issues and first-week schedules. A page that ranks fourth for the head term but has the clearest paragraph on one sub-question can be the one that gets quoted.
Second, the unit of selection is a passage, not a page. The model needs a chunk of text it can lift with confidence. A 4,000-word post that buries its answer in paragraph nine competes poorly against a shorter page that states the answer in the first two sentences under a clear heading.
Vendors have published studies on how often AI Overview citations overlap with top-10 organic results, and the reported figures vary a lot between studies and over time. The honest reading is that the relationship is real but loose, and it shifts as Google changes the system. Treat ranking as a prerequisite, not a strategy.
Step 1: confirm you are eligible
Google's own documentation on AI features says there is no special markup or separate technical requirement. To appear as a supporting link, a page needs to be indexed and eligible to be shown in Search with a snippet. That makes the checklist short, and it is worth running before anything else:
- Googlebot can crawl the page. Check robots.txt and any bot-blocking rules at your CDN or firewall. Aggressive bot protection sometimes blocks legitimate crawlers.
- The page is indexed. Use the URL Inspection tool in Search Console. "Crawled, currently not indexed" is common on thin or near-duplicate pages.
- You have not restricted snippets. nosnippet, a very low max-snippet value, or data-nosnippet on your key paragraphs can keep content out of AI features.
- The answer is in the HTML. If your main content only appears after heavy client-side rendering, retrieval may miss it.
- Do not confuse Google-Extended with AI Overviews. Blocking Google-Extended controls certain Gemini uses of your content. It does not control whether Googlebot-crawled pages appear in AI Overviews.
If you also want citations from ChatGPT, Claude and Perplexity, their crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot) need access too. Many sites block them by accident through a blanket rule, so check each one separately. Try our GEO checker for a preliminary check of your site against some of these; a full site audit is available in the Geograph app.
Step 2: write passages that can stand alone
A citable passage can be copied out of the page and still make complete sense. Practically, that means:
- Put the answer first. The first sentence under a heading should answer the question the heading asks. Context and caveats come after.
- Name things. Replace "this approach" and "the tool" with the actual noun. A passage that depends on the previous paragraph for meaning is hard to quote.
- Make one claim per paragraph. Short paragraphs with a single point are easier to retrieve and attribute.
- Attribute every number. "Conversion rose 18% (our Q1 data, 212 trials)" is citable. "Conversion rose significantly" is not.
- Use headings that look like real questions or clear topics. "How long does SOC 2 take for a startup" matches how people (and fan-out queries) are phrased. "Our approach" matches nothing.
A before and after shows the difference.
Before: "Many founders wonder about pricing. There are lots of factors, and it really depends on your situation, but let's look at what typically happens."
After: "A freelance SEO consultant typically charges a monthly retainer, while agencies charge more and often take two to three months to show results. The right choice depends on whether you have someone to manage the work."
The second version can be dropped into an AI answer unchanged. The first cannot. For more patterns, the guide on how to write blog posts AI actually cites goes deeper on structure.
Step 3: cover the fan-out, not just the head term
Pick the question your buyer asks, then list the sub-questions an AI system would need to answer it. Take the question "What is the best invoicing tool for freelancers?" Likely sub-questions:
- What features do freelancers actually need in invoicing software?
- How do tools handle multiple currencies and VAT?
- What do the main options cost per month?
- How do they compare on late-payment reminders?
- What is the switching cost from spreadsheets?
Each one is a chance to be cited. You do not need a separate page for each. A single well-organised page with a heading and a direct answer per sub-question works, as long as every section is substantive. Sections of two vague sentences add length without adding citability.
A quick way to build the list: search your target question, read the AI Overview, and note which sub-topics it covers. Then check the "People also ask" box and the first page of Reddit threads for the same question. The gaps between what exists and what is missing are your editorial calendar.
Step 4: earn corroboration outside your own site
This is the part pure on-page SEO misses. AI systems are cautious about claims that only one source makes. A passage on your site saying you are the best option for a niche carries little weight unless other places say something compatible.
Corroboration comes from several places:
- Reddit and forums. Reddit content is heavily represented in AI answers, and Google signed a data licensing agreement with Reddit that Reuters reported in early 2024 at around $60 million a year. Genuinely useful replies in relevant subreddits create indexable, quotable text that names your product in context. Reddit also rewards the opposite of spam, so the approach matters. We cover it in how to promote your product on Reddit without getting removed.
- Contextual backlinks. A link from a relevant article, placed within a sentence that explains why your product is useful, tells both search engines and AI systems something that a directory listing does not. See what a backlink is for the quality factors.
- Consistent descriptions. If your homepage, your listings and your Reddit replies all describe the product in roughly the same words, models learn the association. If every source describes you differently, they do not.
Corroboration is slow to build, but it compounds, and competitors cannot copy it by rewriting a page.
Step 5: give the system something only you have
Generic explainers are the easiest content for an AI system to replace, because it can generate them itself. What it cannot generate is your first-party information:
- Your own benchmark data, even from a small sample, with the sample size stated.
- A specific teardown of a workflow you ran.
- Pricing you have verified yourself, with a visible date.
- A named, repeatable method you invented (original framing gives models something to attribute).
Keep dated pages current. A "last updated" line only helps if the content actually changed, and stale pricing or outdated screenshots are a reason to be skipped.
What does not work
- Mass-publishing thin posts. Fifty generic articles do not create fifty citation opportunities. They create fifty pages Google may decline to index.
- Stuffing keywords like GEO or AEO into headings. The retrieval is semantic. Clear writing beats repetition.
- Relying on llms.txt. Google representatives have said they do not use it for Search, so do not expect it to move AI Overviews.
- Schema as a shortcut. Structured data helps search engines understand a page, but Google says no special markup is required for AI features. It will not rescue a vague page.
- Tracking only rankings. A rank tracker will tell you where you sit for one keyword and nothing about whether you were quoted.
How to measure it
AI Overview visibility is awkward to measure, so keep the method simple and consistent:
- Choose 20 to 30 questions your customers really ask.
- Run them monthly in Google (logged out, same location) and note whether an AI Overview appears and which domains it cites.
- Record whether you are cited, whether a competitor is, and which passage was used.
- Use Search Console for clicks and impressions. Google counts AI Overview appearances in the Web search type totals, but does not break them out, so expect to infer patterns rather than read them directly.
Note when you changed a page and when a citation appeared. Over a quarter, those pairs tell you which of the changes above are working for your niche.
Where Geograph fits
Most founders can do the checklist above, but doing it weekly alongside everything else is where it falls apart. Geograph covers the repeatable parts in one place: it monitors relevant subreddits for questions and drafts replies that follow each subreddit's rules (you approve every one before it posts), generates 15 optimised articles a month, runs weekly audits that include crawlability for Googlebot and the AI search bots, and places contextual backlinks through its member network. If your question is research depth rather than execution, the Geograph vs Ahrefs comparison lays out which job each tool is built for.
The takeaway
Pick five questions your best customers asked you in the last month. Search each one, read the AI Overview, and write down which sub-question you cannot answer in two direct sentences on your own site. Fix the weakest one this week with a self-contained, dated, specific answer, then find the Reddit thread where someone asked it and add a genuinely useful reply. To see which technical blockers might be hiding your site from AI crawlers before you start, run your URL through Geograph's free GEO site review.
Frequently asked questions
- Do I need to rank in the top 10 to be cited in AI Overviews?
- No, but being indexed and ranking for related queries improves your odds. Because of query fan-out, a page can be cited for a sub-question even when it does not rank highly for the main keyword.
- Is GEO different from SEO?
- They overlap heavily. SEO focuses on ranking pages for queries, while GEO focuses on being quoted inside generated answers. Technical SEO is the foundation for both.
- How long does it take to get cited?
- It varies by niche and competition. Geograph's own stated timeline for AI citations is two to six weeks after starting a consistent workflow, but treat any timeline as a rough guide rather than a promise.
- Can I block AI Overviews from using my content?
- You can limit snippets with nosnippet or max-snippet controls, though that also affects how you appear in regular results. Blocking Google-Extended does not stop AI Overviews.
- Does Reddit really matter for Google's AI answers?
- Reddit threads are frequently surfaced in Google results and AI answers, and Google has a licensing deal with the platform. Helpful, rule-compliant replies create text that models can retrieve and associate with your product.
Get found everywhere buyers already look.
Join the waitlist and lock in a founding rate before we launch.
Get found firstPricing and figures mentioned are accurate as of publish (1 October 2026) and may have changed since - check each provider's site for current numbers.
Written by Toby Marshman