Google AI Overviews & AI Mode SEO: how a page earns the link

An AI Overview is a generated summary at the top of an ordinary results page. Google shows one “only when our systems determine that it is additive to classic Search,” so many queries never get one. AI Mode is the conversational mode you open on purpose, or reach by asking a follow-up from an Overview: since 27 January 2026 that hand-off works globally on mobile, with the Overview’s context carried into the chat.

They share everything a site owner controls — eligibility, the opt-out and the reporting — and differ in almost everything a site owner sees.

The answer-size and citation rows come from Ahrefs’ comparison of 540,000 query pairs (US data, September 2025). Its headline finding is the one that matters for planning: the two features’ answers were 86% similar in meaning on average, yet they cited the same URLs only 13.7% of the time. Google says as much — the two “may use different models and techniques, so the set of responses and links they show will vary.” You can’t optimise for one slot in one feature; you win by being a strong source across the subtopics both of them search.

Google’s guide to optimizing for generative AI features, added on 15 May 2026, names the two techniques behind both features. Retrieval-augmented generation (“also known as grounding”) uses Google’s “core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index.” Query fan-out is “a set of concurrent, related queries generated by the model to request more information.”

Google’s own example: for how to fix a lawn that's full of weeds, the fan-out queries might include these three. Each is a separate search with its own winners, and any of them can put a link in the answer — which is what the figure at the top of this page shows.

• remove weeds without chemicals
• best herbicides for lawns
• how to prevent weeds in lawn

1. 1

   Google decides whether an AI answer is worth showing

   Overviews are reserved for queries where Google thinks they add something. Question-shaped and longer queries trigger far more often: in Pew’s March 2025 browsing data, 60% of searches starting with who, what, when or why produced a summary, against 8% of one- or two-word searches and 53% of searches of ten words or more.

   Your lever: Aim at the longer, question-shaped searches your buyers actually make. Short head terms rarely trigger an Overview.

2. 2

   The model fans the question out

   It issues several related searches at once, across subtopics and data sources. Google says this lets it surface a wider set of links than a classic results page would.

   Your lever: Cover the sub-questions around your topic in depth — in one strong page or a tight, linked cluster.

3. 3

   Core ranking answers each sub-query

   Every fan-out search is answered from the Search index by the normal ranking and quality systems. Pages that aren’t indexed, or can’t show a snippet, never enter the pool.

   Your lever: Pass the three switches in the next section.

4. 4

   Gemini writes an answer grounded in those pages

   Google made Gemini 3 the default model for AI Overviews in January 2026, and Gemini 3.5 Flash the default for AI Mode at I/O in May 2026.

   Your lever: Give it passages worth grounding a claim on: specific, first-hand, and clearly attributable to you.

5. 5

   Links attach to the claims they support

   Since May 2026, links sit inline beside the text they back, show the site name or page title on hover (desktop), and many answers end with an “Explore new angles” list of in-depth articles. In October 2026 Google said it was showing inline links higher, adding thumbnails to related-article links, and testing a block of official sites at the top of some answers.

What the citation studies show

Read these as a direction, not one true number. The methods differ: Ahrefs’ 2025 run counted only the three most visible citations per answer, and Ahrefs says its 2026 parser detects more citations, so the two runs aren’t strictly comparable. Ahrefs’ explanation, which it hedges, is more aggressive fan-out after Gemini 3 took over AI Overviews; Google hasn’t confirmed any change. What every study agrees on is that ranking for the typed query is one way in, not the only one.

Google’s AI features documentation sets the bar: a page must be “indexed and eligible to be shown in Google Search with a snippet,” and “there are no additional technical requirements.” The 2026 optimization guide adds one more: “a site must be included in Search generative AI features in Search Console.” Include is the default, but owners can change it per property, so check rather than assume.

IndexedMust be on

Confirm with URL Inspection in Search Console. A page Google hasn’t indexed can’t be retrieved for any fan-out query, and indexing “isn’t guaranteed,” so internal links and sitemaps still do work here.

Snippet-eligibleMust be on

No nosnippet, and no max-snippet value so small it starves the model. Google’s robots meta spec says both limit what may be used as “direct input for AI Overviews and AI Mode.” Google also honours robots meta tags placed in the body, so a stray tag injected by a plugin counts.

Search generative AI: IncludeMust be on

Search Console → Settings → Search generative AI. Child properties follow their parent by default but can override it, and owners can switch it. Excluded sites “won’t receive any traffic or impressions from these features.”

Crawlable end to endMust be on

Google asks you to make sure crawling is allowed “in robots.txt, and by any CDN or hosting infrastructure.” A bot-management rule that challenges Googlebot takes you out of Search and its AI features together.

Content inside the first 2 MBMust be on

Googlebot crawls the first 2 MB of each HTML, CSS or JavaScript file (uncompressed) and indexes only what it fetched; PDFs get 64 MB. Keep the main content early and avoid inlining huge scripts or images.

JavaScript that rendersHelps

Google renders JavaScript that isn’t blocked. Check the rendered HTML in URL Inspection; server-rendered text remains the safest choice.

A two-minute eligibility check

1. 1Open Search Console → Settings → Search generative AI on every property you own, including URL-prefix properties that may override the domain property, and read the value.
2. 2Run URL Inspection → Test live URL on a key page. You want indexing allowed and a successful page fetch.
3. 3In the source and the rendered HTML, search for nosnippet, max-snippet, data-nosnippet and noindex, and account for each one.
4. 4Check your CDN or firewall logs for challenges or blocks served to verified Googlebot, not just your robots.txt.

There is no separate AI crawler for Google Search. Google’s documentation says robots.txt rules for Googlebot are the control for crawling, for Search and its AI features alike. The token most people reach for, Google-Extended, does something else: per Google’s crawler documentation it governs training of future Gemini models and grounding in the Gemini app and Vertex AI, it “does not impact a site’s inclusion in Google Search,” and it has no user agent of its own.

So a robots.txt that opts out of Gemini training while staying fully eligible for AI Overviews and AI Mode looks like this:

# Google Search, including AI Overviews and AI Mode. Never block this.
User-agent: Googlebot
Allow: /

# Optional: opt out of Gemini training and Gemini-app / Vertex AI grounding.
# No effect on AI Overviews or AI Mode.
User-agent: Google-Extended
Disallow: /

The Search Console control, live for all sites worldwide since 31 August 2026, is the only lever built for this job. It works per property — domain, subdomain or URL-prefix, so a URL-prefix property can carve out a folder — and Google says content is excluded “within 1-2 days,” though caching can stretch that. It “isn’t used as a ranking or inclusion signal affecting other parts of Search,” doesn’t override Merchant Center or Ads participation, and doesn’t stop Google using your pages to “power Google Search as a whole,” such as understanding queries.

To keep a single passage out of AI answers without losing the page — a price you don’t want quoted out of context, say — wrap it in data-nosnippet:

<p>Plans start at $29 a month.</p>
<div data-nosnippet>
  Enterprise pricing is agreed per contract.
</div>

Eligibility gets a page into the pool. Being chosen is ranking and quality — applied not just to the typed query but to every search Google fans out to. These are the signals worth working on, with how sure anyone can be about each.

• Core ranking, applied per sub-query

  Official

  Grounding runs on the same ranking and quality systems as Search, so helpful, reliable, people-first content and E-E-A-T apply unchanged — once for each fan-out search.

• Non-commodity, first-hand content

  Official

  Google contrasts “7 Tips for First-Time Homebuyers” with “Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line.” Experience and original material are what a generic answer can’t supply.

• Owning a sub-question

  Observed

  Most cited pages no longer rank top 10 for the typed query (Ahrefs, March 2026). Being the best result for one sub-question is a route in that doesn’t require beating the head-term leaders.

• Video beside the text

  Observed

  Google’s guide recommends supporting text with high-quality images and video. In Ahrefs’ March 2026 data YouTube was the most-cited domain in AI Overviews, at 5.6% of cited URLs.

• Real discussion, not manufactured mentions

  Official

  Answers now preview “Community Perspectives” from forums and social posts, with the creator named. Google also warns that “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.”

• Clear paragraphs and sections

  Our read

  Google: people “appreciate it when web pages are organized by paragraphs and sections,” and there’s no need to chunk text for AI. Opening each section with its direct answer is a practitioner habit Google hasn’t endorsed.

Mapping the sub-questions, step by step

1. 1Write down one buyer question you want to be cited for, in the words a buyer would type.
2. 2List the searches Google would plausibly fan it out to. People Also Ask, related searches and AI Mode’s own follow-up suggestions are good prompts.
3. 3Search each sub-question in a private window and note who wins it and how strong those pages are.
4. 4Pick the sub-question where the winners are weakest and you hold first-hand material — data, tests, photos, a process only you run.
5. 5Build or improve the page that answers it best, and link to it from the hub page for the main question.

2026 gave publishers ways to stand out in Google’s AI search that don’t depend on rank alone. Each label appears only for some users or some links, but each is something you can act on.

The preferred sources guide documents a JavaScript button and a plain deep link. The link needs no script and works in newsletters and social posts as well as on your site:

<a href="https://www.google.com/preferences/source?q=example.com">
  Add us as a preferred source on Google
</a>

Google says people are “twice as likely to click through” to a Preferred Source, that more than 345,000 sources had been selected by May 2026, and in October that use had grown more than 7× since January. Those are Google’s own figures. Google doesn’t describe preferred sources as a ranking signal: the label shows when your page is already in the answer.

A lot of 2026 GEO advice is aimed at engines other than Google. For Google’s AI features, its own guide has a mythbusting section, and the 15 June 2026 documentation update made the llms.txt answer explicit.

• Skip

  You need llms.txt or a Markdown copy of your pages.

  Google: “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search.” Doing so “will neither harm nor help.”

• Skip

  Content has to be chunked into short passages for AI.

  “There’s no requirement to break your content into tiny pieces,” and there’s “no ideal page length.” Write sections that read well to people.

• Skip

  FAQ schema gets you into AI answers.

  There’s no special schema for AI features, and FAQ markup no longer earns FAQ rich results either: Google stopped showing them on 7 May 2026 and removed the documentation on 15 June. Keep structured data that matches the page for the rich results still supported.

• Skip

  Blocking Google-Extended keeps you out of AI Overviews.

  It doesn’t. Use the Search Console control, nosnippet or noindex — each with the side effects in the table above.

• Skip

  GEO for Google is a separate discipline with its own tricks.

  Google: “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” It tells site owners to prioritise effective SEO over “AEO/GEO hacks.”

For pages that aren’t cited, AI Overviews cost clicks. For pages that are, they cost less. That makes earning the citation a defensive move as much as a growth one.

The table is Seer Interactive’s 2026 update: 53 brands, 2.43 billion organic impressions, January 2025 to February 2026. Seer is explicit that it shows correlation, not cause — higher-authority brands are also more likely to be cited — and one account makes up 47% of the uncited impressions. Ahrefs’ −58% is modelled from desktop Search Console data. Google, for its part, says clicks from results pages with AI Overviews are “higher quality,” without publishing numbers.

On 3 June 2026 Google introduced Search generative AI performance reports, first for a subset of sites and for all sites worldwide by 31 August. It’s the first first-party view of AI visibility, with one large gap: no clicks.

• AI Overviews take one position on the page, and every link inside gets that same position. A click on an external link counts as a click.
• AI Mode positions follow the normal results-page method, and each follow-up question is logged as a new query.
• Impressions in the report’s chart are aggregated by property: two of your links in one answer count once. The page table counts by page.
• Gaps: Search Labs experiments aren’t counted, and the report may not appear at all for sites with too few AI impressions or set to Exclude.

A workable routine: each month, list the pages gaining AI impressions and compare their Search clicks and GA4 conversions with the month before. Each week, run a fixed set of 20–30 buyer questions in a private window and note which pages Google cites — by hand, the same questions every time. Google’s caveat applies to any tool that promises more: “No third-party tool has access to our internal ranking or AI systems.”

AI answers on Google also carry product listings, product details and local business information, and those come from Google’s own data sources as much as from your pages. Google’s guide points to three of them.

• Merchant Center: keep feeds complete — prices, availability, shipping, returns and attributes — so products can appear in AI responses and other results.
• Google Business Profile: hours, categories, photos and reviews feed local answers. Agentic booking for local experiences and services was rolling out to everyone in the US from summer 2026.
• Business Agent: Google suggests considering its conversational brand agent, depending on your business type and goals.
• Agents on your site: browser agents may read pages through screenshots, the DOM and the accessibility tree. Google links to web.dev’s agent-friendly website practices and notes the emerging Universal Commerce Protocol.