Google AI Overviews & AI Mode SEO: the technical guide

How AI Overviews and AI Mode choose their sources#

Both features are built on Search itself. Google’s AI optimization guide says they “are rooted in our core Search ranking and quality systems” and names two techniques: retrieval-augmented generation, where core ranking retrieves pages from the Search index and a Gemini model writes an answer linking to them, and query fan-out, “a set of concurrent, related queries generated by the model.”

Fan-out is the part that changes how you plan content. Google’s own example: “how to fix a lawn that’s full of weeds” becomes “best herbicides for lawns,” “remove weeds without chemicals” and “how to prevent weeds in lawn.” Each of those is a separate search with its own winners, and each can contribute a link to the final answer.

1. 1

   Google decides whether an AI answer adds anything

   AI Overviews are “only shown when our systems determine that it is additive to classic Search, and as such, often don’t trigger.” Question-shaped and long queries trigger them far more often: Pew saw summaries on 60% of question queries against 8% of one- or two-word ones.

2. 2

   The query fans out into subtopics

   The model issues several related searches at once, across subtopics and data sources. AI Mode’s Deep Search can issue “hundreds of searches” for a single report.

   Your lever: Cover the sub-questions around your topic in depth — as a strong page or a tight cluster, not one thin page per variant.

3. 3

   Core ranking retrieves candidate pages

   Each fan-out query is answered from the Search index by the normal ranking and quality systems. Pages that are not indexed, or not allowed a snippet, never enter the pool.

   Your lever: Be indexed, snippet-eligible and set to Include in Search Console.

4. 4

   Gemini writes a grounded answer

   Gemini 3 has been the default model for AI Overviews since January 2026; Gemini 3.5 Flash has powered AI Mode since I/O in May 2026. Google notes the two “may use different models and techniques, so the set of responses and links they show will vary.”

   Your lever: Give it passages worth grounding on: specific, first-hand, clearly structured.

5. 5

   Links are attached to the claims they support

   Since May 2026 Google places more inline links next to the generated text, previews site names on hover, labels links from publications the user subscribes to, adds a “further exploration” list of in-depth articles, and surfaces forum and social perspectives with the creator attributed.

6. 6

   AI Mode keeps the conversation going

   Follow-up questions — including ones started from an AI Overview — continue in AI Mode with context carried over. In Search Console, each follow-up counts as a new query.

Eligibility: the switches that decide if you can be cited#

Google’s AI features documentation sets one bar: a page must be “indexed and eligible to be shown in Google Search with a snippet.” It adds that “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 to be eligible.” Include is the default — but check it, because anyone with owner access can change it.

IndexedRequired

Confirm with Search Console’s URL Inspection. A page Google hasn’t indexed cannot be retrieved for any fan-out query. Indexing “isn’t guaranteed,” so internal links and sitemaps still matter.

Snippet-eligibleRequired

No nosnippet, and no max-snippet value so low it starves the model. Both explicitly limit what AI Overviews and AI Mode may use as input.

Included in Search generative AIRequired

Search Console → Settings → Search generative AI must read Include (or inherit Include from a parent property). Excluded sites “won’t receive any traffic or impressions from these features.”

Googlebot allowed — everywhereRequired

In robots.txt and at your CDN, WAF and bot-management layer. Google calls this out: an “AI bot” blocklist that catches Googlebot takes you out of Search and its AI features together.

Crawlers and controls: what each one actually does#

There is no separate AI crawler for Search. Google is explicit: “AI is built into Search and integral to how Search functions, which is why robots.txt directives for Googlebot is the control.” The token most people reach for, Google-Extended, does something else entirely.

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

# Search, AI Overviews and AI Mode — never block this
User-agent: Googlebot
Allow: /
 
# Optional: opt out of Gemini training and Gemini-app grounding.
# Has no effect on AI Overviews or AI Mode.
User-agent: Google-Extended
Disallow: /

Every control, and what it really switches off

The Search Console control, rolled out worldwide on 31 August 2026, is the only one built for this job. It works per property — domain, subdomain or URL-prefix, so a URL-prefix property can scope it to one folder — takes effect in “a few days,” and Google says it is “not used as a ranking or inclusion signal affecting other parts of Search.” It does not touch training (that’s Google-Extended) or Merchant Center and Ads participation.

To keep one 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 negotiated per contract.
</div>

Technical requirements, and the things that don't matter#

Because AI features sit on Search, the technical bar is Search’s bar. The useful work is making sure Google can render, understand and prefer your page — and not wasting weeks on files and markup Google has said it ignores.

JavaScript renderingHelps

“Google is able to process content within JavaScript as long as it isn’t blocked.” Don’t block JS or CSS in robots.txt, and check the rendered HTML in URL Inspection. Server-rendered text is still safest — and it’s what every other engine needs.

Key content in the first 2 MBRequired

Googlebot fetches the first 2 MB of a URL’s HTML; anything after the cut-off is “not fetched… not rendered… not indexed.” Put the title, canonical and main content early, and avoid inlining huge scripts or base64 images.

Page experienceHelps

Works on every device, loads quickly, and makes the main content clearly distinguishable from navigation and ads.

Internal links & duplicatesHelps

Make deep pages “easily findable through internal links,” and consolidate duplicates with canonicals so ranking signals aren’t split.

Original images & videoHelps

Google’s guide recommends high-quality media. YouTube was the single most-cited domain in AI Overviews in Ahrefs’ 2026 data, at 5.6% of cited URLs.

Structured dataNo effect

Not required for AI features and there is “no special schema.org structured data.” Keep it for rich results — and make sure it matches the visible text.

llms.txt & AI text filesNo effect

“You don’t need to create new machine readable files, AI text files, or markup.” Google Search ignores llms.txt; it neither helps nor hurts here.

Chunking & rewriting for AINo effect

Google says there is no need to chunk content or rewrite it for AI, and “no ideal page length.” Write clear sections for people.

Merchant Center feedHelps

For shopping answers. AI Mode shops from the Shopping Graph (60B+ listings), so keep prices, shipping, returns and attributes current. Agentic checkout is live for select US merchants.

Google Business ProfileHelps

For local answers — hours, categories, reviews. AI Mode’s agentic booking for local services launched in the US in summer 2026.

What gets cited: ranking, fan-out and non-commodity content#

Ranking still matters — it’s how pages enter the pool — but the typed query is no longer the only race. Fan-out means your page can be cited for a sub-question you rank for even when you’re nowhere for the original query. The data shows the share of citations coming from outside the top 10 growing fast.

Ahrefs attributes the fall to more aggressive fan-out after Gemini 3 arrived. What that means in practice: the most reliable way into an answer is to be the best result for one of its sub-questions.

• Core ranking and quality

  Official

  Grounding runs on the same ranking and quality systems as Search. Helpful, reliable, people-first content and E-E-A-T apply unchanged.

• Non-commodity content

  Official

  Google contrasts “7 Tips for First-Time Homebuyers” with “Why We Waived the Inspection & Saved Money.” First-hand experience and original data are what a generic answer can’t supply.

• Subtopic coverage

  Observed

  Pages that win a fan-out sub-query get cited even when they don’t rank for the head term — the likely driver of the 76% → 38% shift.

• Video and forums

  Observed

  YouTube leads cited domains, and since May 2026 Google surfaces forum and social “perspectives” with attribution. Being discussed off-site is a second path in.

• Clear sections

  Our read

  Google recommends clear headings and sections. It doesn’t endorse answer-first formatting specifically — we find direct openings get lifted more, but that’s practitioner evidence.

• Per-variant pages backfire

  Official

  Creating pages for every fan-out variant “primarily to manipulate generative AI responses” is, per Google, scaled content abuse. Build depth, not doorways.

What AI Overviews do to your clicks#

The honest framing: for uncited pages, AI Overviews cost clicks. For cited ones, they cost less — and sometimes pay. That makes being cited as much a defensive move as a growth one.

Seer’s numbers put it most concretely: per million informational impressions, about 33,500 clicks with no AI Overview, about 20,700 when the brand is cited, and about 9,400 when it isn’t. Seer notes this is correlation, not proof. Google’s own position is that AI-feature clicks are “higher quality,” though it hasn’t published data for that.

Measuring AI Overviews and AI Mode#

Measurement changed more in 2026 than anything else. On 3 June Google launched the Search generative AI performance report, rolled out worldwide by 31 August. It’s the first first-party view of AI visibility — with one big gap.

• Impressions count when a link to your site is shown in an AI feature — for AI Overviews, only once it’s scrolled or expanded into view. Two links in one answer count as one impression in the property chart.
• Position: an AI Overview occupies one position, and every link in it is assigned that position. In AI Mode, positions count as in regular Search.
• Follow-ups in AI Mode are logged as new queries.
• GA4: a new AI Assistant default channel (May 2026) groups assistants such as ChatGPT, Gemini and Copilot. Google’s own AI features are excluded — those clicks still land in Organic Search from google.com.

The practical setup: watch the AI report for which pages gain impressions, check those pages’ engagement and conversions in GA4, and run a fixed weekly prompt panel for the citations Search Console can’t show. Google’s own caveat applies to every tool, ours included: “no third-party tool has access to our internal ranking or AI systems.”

Myths worth dropping#

The action checklist#

Everything above, in the order we’d do it. Tick items off as you go — your progress is saved in this browser.

Google AI Overviews & AI Mode SEO: frequently asked questions#