How to appear in Google AI Overviews and AI Mode
A practical guide to getting cited in Google's AI Overviews and AI Mode: how the systems retrieve and ground answers, and the page-level tactics that earn citations.
5 min read
How AI Overviews and AI Mode actually pick sources
AI Overviews (the summary box at the top of some Google results) and AI Mode (Google's fuller conversational search experience) are not pulling answers from a model's memory. They are grounded in live search results. Google takes your query, and behind the scenes runs a technique it calls query fan-out: it decomposes one question into several related sub-queries, runs them in parallel against its index, and assembles an answer from passages across the pages that rank for those sub-queries. The links shown are citations for the claims in that answer.
Two consequences matter. First, classic ranking is still the gatekeeper: if a page does not surface in the underlying search results for the relevant sub-queries, it cannot be cited. Indexability, crawlability, and conventional relevance signals remain the entry ticket. Second, the unit being selected is the passage, not the page. Google is looking for a specific paragraph or sentence that directly answers a sub-query. A page that ranks but buries the answer in marketing prose loses to a page that states it plainly.
This is why you often see pages cited in an AI Overview that are not the number-one blue link. The system rewards the clearest, most directly extractable answer to the decomposed question, not necessarily the highest-authority domain overall.
Write for the sub-questions, not just the headline keyword
Because of query fan-out, the highest-leverage move is to anticipate the sub-queries Google will generate and answer each one explicitly on the page. If your target query is "best CRM for a small agency", the fan-out likely includes things like "CRM pricing for small teams", "CRM with client portals", "easiest CRM to set up", and "CRM vs spreadsheet for agencies". A page that addresses only the headline phrase competes for one passage; a page that has a clean, self-contained answer to each sub-question is eligible to be cited across several of them.
Practically: use descriptive H2/H3 headings phrased as the questions real people ask, and put a direct, complete answer in the first one or two sentences underneath. Lead with the answer, then justify it. "The most affordable CRM for agencies under five people is X at £Y per user, because Z" is extractable. "In today's fast-moving landscape, choosing a CRM can feel overwhelming" is not. Each section should make sense if lifted out on its own, since that is exactly what the system does.
Tables and tight comparison lists are disproportionately useful here. A specification table or a labelled "X vs Y" list gives the model structured, unambiguous facts it can quote with confidence, which is far safer for it to surface than a claim it has to infer from a wall of text.
Earn citations with specificity and evidence
AI answers favour content that reduces the model's uncertainty. Concrete numbers, dates, named methods, prices, and first-hand data are easier to cite than vague assertions, because the system can attribute a precise claim to a precise source. A sentence like "agencies in our 2026 survey of 400 users cut onboarding time by 31%" is citation-bait in the good sense; "our tool saves significant time" is not, because there is nothing specific to attribute.
Original information is the strongest moat. Proprietary benchmarks, pricing breakdowns, step-by-step methods, and genuine test results give Google a reason to cite you rather than a competitor restating common knowledge. If five pages say the same generic thing, the system has little reason to prefer yours. If you are the only source of a specific data point or comparison, you become the natural citation for any query that touches it.
Match the format to the intent, too. Definitional and "how-to" queries reward clear ordered steps and plain definitions near the top of the page. "Best" and comparison queries reward balanced, honest assessment that names trade-offs, including where you are not the right choice. Answers that read as even-handed are more likely to be treated as trustworthy reference material than ones that read as a sales pitch.
Get the technical foundations right
None of the above works if Google cannot crawl and render the page. Make sure key content is in the server-rendered HTML rather than injected by client-side JavaScript that the crawler may not execute, keep important answers out of tabs or accordions that hide them from extraction, and confirm the page is indexed. AI Overviews draw on Google's standard index, so anything blocked in robots.txt or marked noindex is invisible to them.
A separate, newer control governs AI Mode and Gemini-grounded experiences: the Google-Extended user-agent token. Allowing it keeps your content eligible to be used for grounding in these surfaces; disallowing it can remove you from consideration there while leaving normal Search ranking intact. Check this deliberately rather than inheriting a blanket block someone added to keep out AI crawlers, because the same instinct that protects your content from training can also exclude you from being cited.
Structured data helps the system parse what your page is about. Article, FAQPage, Product, and HowTo schema where genuinely applicable, plus clean semantic HTML and a clear last-updated date, all make a page easier to interpret and trust. Freshness matters more than most teams assume: AI answers visibly favour recently updated sources for anything time-sensitive, so revisiting and re-dating cornerstone pages is a real ranking input, not housekeeping.
Build the off-page consensus the model reads
AI answers lean on what the wider web agrees about you. For comparison and recommendation queries especially, Google's systems weigh how often and how consistently your brand appears in independent third-party content: roundups, review sites, forum threads, comparison articles, and industry directories. If your category's "best tools for X" listicles and Reddit threads never mention you, you are missing from the exact sources the fan-out retrieves for those high-intent queries, no matter how good your own pages are.
The work here is digital PR and category presence, not link-buying. Get included in credible roundups, encourage genuine reviews on the platforms your buyers cite, contribute substantive answers in the communities Google surfaces, and make sure your name is associated consistently with the specific use cases you want to win. Consistency of description matters: if some sources call you a "project tool" and others a "CRM", you dilute the consensus the model can form.
This is also where measurement closes the loop. Because AI answers vary by phrasing and change over time, you cannot infer your standing from your own analytics alone; you have to query the assistants the way buyers do and watch which brands and sources get cited. Tracking those citations over time tells you which pages are working, which sub-queries you are losing, and where a competitor's third-party consensus is beating your on-page effort, which is the difference between guessing and steering.