How to Get Cited in Google AI Overviews (Small-Site Edition)

If you run a small site and you're asking how to get cited in Google AI Overviews, the honest answer in two sentences: pick long-tail question queries where the big sites haven't written the cleanest answer, publish a page with an answer-first passage backed by a first-party number, and accept that the citation itself will send roughly 1% of clicks. The win isn't traffic. The win is being the page Google quotes when your buyer asks the question.

That reframes everything. You're not optimizing for a #1 ranking that gets clicked. You're optimizing to be the source named inside the summary that gets read.

TL;DR

• Google says no special markup, file, or schema makes you eligible — it's the regular index. (source)

• Pew (July 2025) measured citation clicks at 1% of visits when an AI Overview appears.

• Ahrefs (December 2025) measured a 58% CTR drop on #1 results when AIOs show up.

• Small sites win on long-tail question queries, not head terms with enterprise competitors.

• Optimize for branded impressions, answer-first passages, and first-party data — not click volume.

1. How Google selects AI Overview sources: documented vs speculated

The documented part is short. Google's own guidance on AI features states there is no special markup, file, or schema required to be eligible — eligibility runs through the standard Search index and standard SEO requirements (Google Search Central, AI features documentation). If your page is indexed and ranks for a query that triggers an AI Overview, it's in the eligible pool. That's it. Everything else is pattern observation.

The speculated part is what every SEO blog will sell you: structured data tricks, custom "AI schema," entity-stuffing, llms.txt files. None of these are confirmed by Google. Some are actively contradicted by it. We cover the llms.txt situation separately in what is llms.txt and does Google use it.

What we can observe from the field: AI Overviews appear to prefer passages that directly answer the user's question in the first 1-2 sentences, that are corroborated by other indexed sources, and that come from pages already ranking in the top 10-20 organic results for the underlying query. Treat that as the working model, not gospel.

The practical implication: AI Overview optimization is regular SEO with a passage-level layer on top. You still need to rank. You just need to rank with a page structured so the model can lift a clean, factual passage without rewriting it.

2. The 1% problem: what a citation is actually worth

Here's the math nobody selling you an "AI visibility" subscription wants on the homepage. Pew Research, in a July 2025 study of real Google sessions, found that users click a source cited inside an AI summary in just 1% of visits. Even traditional organic results take a hit: users click any traditional result in 8% of visits when an AI Overview is present, versus 15% without (Pew Research, July 2025).

Ahrefs' updated December 2025 study went further on the head-result impact: AI Overviews reduce the organic click-through rate of the #1-ranked page by 58%, up from 34.5% in its earlier 300,000-keyword study (Ahrefs, December 2025).

So why bother? Because the 99% who don't click still read your brand name attached to the answer. If 4,000 people see an AI Overview that cites Entropy & Co. this month, roughly 40 click through. The other 3,960 just got an unpaid impression with implicit endorsement from Google. For a DR-0 site, that's the asset you can actually accumulate. We unpack this difference further in answer engine optimization vs SEO.

The right KPI for small-site AIO work isn't sessions. It's cited impressions per month and the share of those impressions on queries that match a buying intent. Optimize for that and the click number, when it comes, comes free.

Five gates between an indexed page and a paying click. Most pages fail at gate 2.

3. Query selection: where small sites actually get retrieved

The SERPs where a DR-0 site can win AI Overview citations have three features in common. They're framed as a question. The answer is specific enough that a generic enterprise page can't summarize it well. And the existing top-10 results are mostly forum threads, outdated blog posts, or thin affiliate content.

In our client work, the queries that produce citations for small sites typically look like:

• "how long does [specific process] take for [specific business size]"

• "what does it cost to [specific action] in [specific city or vertical]"

• "can you [specific edge case] with [specific tool] when [specific constraint]"

Head terms like "best CRM" or "how to do SEO" trigger AI Overviews that pull from the usual suspects — HubSpot, Search Engine Journal, Moz. You will not displace those on a six-month timeline with a new domain. Stop trying. Pick the question your customer actually types when they're three weeks into evaluation, when their search has gotten specific.

A rough filter we use: if the top 3 organic results all have a domain rating above 70 and the AI Overview already exists and reads cleanly, walk away. If the top 3 are a Reddit thread, a 2021 blog post, and a Quora answer, that's a query you can win on craft alone.

4. Page patterns that get quoted

Four patterns show up repeatedly in the pages we've seen pulled into AI Overviews for small-site clients. None of them are tricks. All of them are good writing made retrieval-friendly.

Answer-first sections. The first 40-60 words under an H2 should answer the question that H2 poses, in plain language, with the relevant number or named mechanism. Background and context come after. Models lift the top of the section; bury the answer and you don't get quoted.

Visible, schema-marked FAQs. Not because schema is magic — Google has been clear it isn't required — but because the Q&A structure forces the writing pattern that models prefer. The schema helps Google parse the boundaries cleanly.

First-party numbers. "In our work with 14 brokerages between January 2025 and May 2026, the median deployment took 18 days from kickoff to first agent action." That sentence, with a specific number and a defined cohort, is more quotable than a regurgitated industry statistic. Generic stats are everywhere. Your number is yours.

Named mechanism, not vibes. A page that says "we use AI to make things efficient" gets ignored. A page that says "we wire the CRM, inbox, and billing into one decision loop with a human approver above $5k spend" gets quoted. Specificity is retrieval bait.

One more pattern worth naming: the pages that get cited tend to be the ones a competent operator would actually screenshot and send to a colleague. That's not a coincidence. The model is optimizing for the same thing — a clean, defensible passage.

5. What not to bother with

A short list of things consuming small-site SEO budgets right now that have no documented effect on AI Overview citations.

llms.txt files. Google has not confirmed any use of llms.txt. As of June 2026, it's a proposal with adoption from some AI vendors but no signal from Google Search. Publishing one costs nothing, so do it if you want. Don't pay anyone for it.

Custom "AI schema." There is no Google-recognized schema specifically for AI Overview eligibility. The standard Schema.org types — Article, FAQPage, HowTo, Product — are the supported set. Anyone selling "AI-optimized schema" as a separate product is selling regular structured data with a markup.

Citation-guarantee services. No vendor can guarantee an AI Overview citation. Google's selection is non-deterministic and changes weekly. If a service promises citations, they're either selling backlinks under a new name or they're going to refund you when it doesn't work.

Most "AI visibility" SaaS at full price. The category has useful tools, but the monthly subscriptions in the $200-$500 range are mostly tracking dashboards a small site can replicate manually for the first six months. We broke down the pricing tiers in AI visibility tools pricing — check published vendor pricing pages before committing.

6. Measuring your citation share without a $499/mo tool

For the first three to six months, manual measurement is fine and arguably more honest than a dashboard.

Build a tracking sheet with three columns: the query, the date checked, and the citation state (cited / not cited / no AIO triggered). Pick 20-40 queries that match the long-tail pattern from section 3. Check them in an incognito window once a week. Log the result. After eight weeks you have a baseline citation share and a trend line. That's enough to make decisions.

For each query where you're cited, also note the position of your link within the AIO source list (1st, 2nd, 3rd+) and whether your passage appears verbatim in the summary text. Verbatim passage = your writing made it into the answer itself, which is the highest-value placement.

If you want to spend money, spend it on a virtual assistant who runs the check every Monday morning for two hours. That's roughly $60-$120 a month depending on rate. Cheaper than any AIO tracking SaaS and the data is the same.

When you cross 50+ tracked queries or your client load requires reporting, then evaluate the tools. Not before.

If you want a second set of eyes on which queries to chase and how to structure the pages, that's a regular part of our SEO work.

FAQ

Does Google publish how it picks AI Overview sources?

Partially. Google states that no special markup, file, or schema is required for eligibility, and that AI features run through the standard Search index. Beyond that, selection logic is not publicly documented and appears to change frequently. Treat field observations as working hypotheses, not confirmed mechanics.

How long does it take a new site to earn its first AI Overview citation?

In our client work, sites that pick the right long-tail queries and publish answer-first pages typically see first citations 8-16 weeks after publication, once the pages are indexed and accumulate a small handful of organic ranking signals. Head-term citations on new domains are rare on any realistic timeline.

Is the 1% click rate from Pew the same across all industries?

No, and Pew didn't claim it was. The 1% figure is an aggregate across observed Google sessions in their July 2025 study. Click rates likely vary by query intent, vertical, and AI Overview length. Use 1% as a planning baseline, then measure your own queries to refine.

Should I add llms.txt to my site to help with AI Overview citations?

There's no documented benefit for Google AI Overviews as of June 2026. Some non-Google AI vendors have indicated support for the proposal. The file is cheap to publish if you want coverage with other systems, but don't expect it to influence Google citation selection.

What's a realistic AI Overview citation share for a DR-0 site after six months?

Based on operator experience rather than a public dataset: getting cited on 10-20% of a focused 30-query long-tail set within six months is achievable with disciplined publishing. Higher shares are possible on tightly defined niches. Anyone promising specific numbers without seeing your queries is guessing.

Where to start this week

Pick 10 long-tail question queries your buyer actually types. Write one page per week that answers each in the first 60 words, backed by one first-party number you can defend. Check citations every Monday for the next eight weeks. If you want a second set of eyes on the query list before you commit the time, send it over.