How to Rank in AI Overviews: Google’s 2026 SEO Guide

Google just published its first official guide on how to optimize for AI Overviews and AI Mode, and the headline from inside the document is harsher than the polite framing suggests. The phrase Google chose, repeated three times across the page, is that optimizing for generative AI search is “still SEO.” That sentence quietly demolishes three cottage industries that have built themselves around AI search in the last 18 months: the llms.txt vendors, the “chunking your content for LLMs” consultants, and the GEO and AEO course sellers charging $497 for tactics Google now says you should ignore.

The guide went live on May 15, 2026 on the official Google Search developer documentation. It is short, declarative, and unusually direct for a Google publication. After six-plus years running SEO end to end for law firms through my agency hey-ash.com, and reviewing AI tools here at CriticNest, this is the cleanest signal I have seen in a year about what Google actually rewards in its generative experiences. Below is the full breakdown, plus what I have observed move rankings on real client sites and what to do about it this week.

TL;DR: How to rank in AI Overviews, per Google’s 2026 guide

If you only read one paragraph today, read this one. Google is telling website owners that the systems behind AI Overviews and AI Mode are not a separate algorithm. They are the same ranking and quality systems used for regular Search, with two AI techniques layered on top: RAG to ground responses in current web content, and query fan-out to expand a single user question into several related searches. To rank in AI Overviews, the page must already be eligible to rank in regular Search with a snippet, which means it must be indexed, crawlable, and pass the standard technical bar.

Beyond eligibility, Google is asking for one thing more than anything else: non-commodity content. The guide uses a specific contrast. “7 Tips for First-Time Homebuyers” is commodity content because anyone could write it from common knowledge. “Why We Waived the Inspection and Saved Money: A Look Inside the Sewer Line” is non-commodity content because it carries first-hand experience that an AI model cannot fabricate without grounding. That single distinction explains roughly 70 percent of the actionable advice in the document.

What are AI Overviews and AI Mode, exactly?

AI Overviews are the AI-generated summary block that appears at the top of certain Google search results. They cite source URLs, link to those sources, and show a short generated explanation rather than (or in addition to) the traditional ten blue links. AI Mode is the standalone conversational search experience, accessed through the AI Mode tab on Google Search, where users can ask follow-up questions and refine their query inside an extended dialogue. Both surfaces draw from the same Google index and the same core ranking systems.

The reason this distinction matters for publishers is simple. People who use AI Overviews are still on the standard SERP and can click through to your page in the traditional way. People who use AI Mode are inside a chat-style interface where the model often satisfies the query without a click. Both are now part of the Google Search experience and both are sourced from the same content pool. There is no separate “AI index” you can opt into or out of.

Worth noting: a page must already be eligible to appear in standard Google Search with a snippet before it is eligible for AI Overviews. If your page is blocked by robots.txt, noindexed, or technically broken, it cannot appear in either. This is the floor.

How AI Overviews pick what to cite: RAG and query fan-out, explained

Google’s guide names two specific AI techniques that operate inside AI Overviews and AI Mode. Understanding both is the difference between writing for what you imagine the algorithm wants and writing for what Google has now publicly confirmed it uses.

Retrieval Augmented Generation (RAG)

RAG, also referred to as grounding, is the technique that connects the language model to live web content. When a user asks a question in AI Overviews or AI Mode, Google’s core ranking systems first retrieve relevant, current web pages from the index. The language model then reads passages from those retrieved pages and generates a response that is grounded in that retrieved content, with clickable citations back to the pages. The model is not pulling answers from a frozen training corpus. It is pulling from the live index in real time.

The practical consequence is that ranking signals you already optimize for, link authority, content quality, freshness, indexability, page experience, are exactly the signals that determine which pages get retrieved as grounding sources. A page that ranks well for a query is the page most likely to be cited inside an AI Overview for that query. There is no separate “RAG ranking” to game.

Query fan-out

Query fan-out is the technique where the model expands a single user query into several related concurrent searches. Google’s example in the guide: the original query “how to fix a lawn that’s full of weeds” can fan out into “best herbicides for lawns,” “remove weeds without chemicals,” and “how to prevent weeds in lawn.” Each fan-out query runs through the same ranking systems and returns its own set of grounding candidates. The final AI Overview blends citations from all of them.

This is the part that quietly rewards comprehensive content. A page that addresses the parent question (weed removal) plus the natural follow-ups (chemical versus organic, prevention versus treatment) within the same URL is more likely to surface across multiple fan-out queries simultaneously. Google explicitly warns against creating separate thin pages for every possible fan-out variation, calling that behavior a scaled content abuse violation. Build depth into one page, not breadth across many.

The SEO best practices Google says still work for AI Overviews

This section of the guide is short on surprises and long on reaffirmation. Every classic SEO discipline that mattered in 2020 still matters in 2026. The framing has shifted only in that Google now describes each one in the context of AI Overviews eligibility.

Non-commodity content. Google’s top priority for visibility in AI Overviews. The guide describes it as content that goes beyond common knowledge and provides unique expert or experienced takes. First-hand reviews, behind-the-scenes accounts, original data, unusual perspectives, the kind of content that an AI model cannot produce on its own from a frozen training set. If a competitor could publish the same article tomorrow by paraphrasing the top ten results for your query, your content is commodity.

Helpful, reliable, people-first content. The phrasing is identical to the long-running Helpful Content guidance. Write for humans, not search engines or LLMs. If a real reader would find the page satisfying, you are on track.

Clear structure and headings. Use semantic HTML where it helps human readers. Sections, paragraphs, descriptive headings. The guide is explicit that perfect semantic HTML is not required, but readable structure helps both human visitors and assistive technology.

High-quality images and video. AI Overviews can surface images and video clips from your pages in addition to text. The standard Google image SEO and video SEO practices apply directly. Original, relevant media is rewarded.

Crawlability and indexability. If Googlebot cannot reach the page, the page cannot appear in AI Overviews. Robots.txt rules, server response codes, render-blocking JavaScript, and crawl budget for large sites all still matter exactly as before.

Page experience. Latency, mobile rendering, intrusive interstitials, and Core Web Vitals continue to influence visibility. Nothing has been demoted or removed in the AI era.

Duplicate content reduction. Duplicate URLs waste crawl budget and confuse the system about which version to surface. The guide flags this as worth fixing if you have the bandwidth.

Structured data for rich results. Use it where it qualifies you for a rich result (Product, Recipe, FAQ, How-To, Article, and similar). Do not use it as a magic incantation to influence AI Overviews. Google says structured data is not required for AI search and there is no special schema.org markup that makes a page more visible in AI surfaces.

Local and ecommerce signals. If you sell products or run a local business, Google Business Profile and Merchant Center feeds remain the path for surfacing product cards and local business details inside AI responses.

AEO, GEO, and llms.txt: 5 myths Google just officially killed

The most useful section of Google’s guide is the part that calls out what does not work. Five tactics, all of which have been sold as essential at one conference or another in the last 18 months, are now explicitly named as things you can ignore. This is the part to forward to anyone trying to sell you a $497 AI search optimization course.

Myth 1: You need an llms.txt file

The llms.txt proposal asked websites to publish a special text file at the root that gives LLMs a curated map of their content. Google’s position is direct: you do not need to create any new machine-readable files, AI text files, markup, or Markdown to appear in generative AI search. Google may discover and index such files like any other file, but they are not treated specially. The llms.txt file is, for Google Search purposes, dead. Anthropic, OpenAI, and Perplexity have not committed to honoring it either. If you have one and it is accurate, leave it. If you are about to spend hours generating one, do not.

Myth 2: You need to “chunk” your content for LLMs

The chunking pitch is that LLMs read content in small, isolated pieces, so you should rewrite every page into LLM-friendly short blocks of self-contained meaning. Google’s response: there is no requirement to break your content into tiny pieces. Their systems understand the nuance of multiple topics on a page and surface the relevant piece to users. There is no ideal page length. Write for your audience, not for an imagined chunking algorithm. The agencies selling content restructuring services on this premise are selling a non-problem.

Myth 3: You need to rewrite content in “AI-friendly language”

A close cousin of chunking. The pitch is that you should rewrite existing content using long-tail conversational keywords, question-based H2s, and specific syntactic patterns that supposedly help LLMs parse meaning. Google explicitly says: AI systems understand synonyms and general meaning. You do not need to capture every variation of how someone might phrase a query. Write naturally. If your writing is clear to a human, it is clear to the model.

Myth 4: You need to chase inauthentic mentions across the web

One of the more aggressive GEO tactics has been to pursue brand mentions across blogs, forums, Reddit threads, and YouTube transcripts on the theory that LLMs are scraping these mentions and using them as ranking signals. Google’s answer is that their AI features do reflect what is being said about products and services, but inauthentic mention-seeking is not as helpful as it might seem because the core ranking systems already filter for quality, and spam systems block manipulation attempts. Buying or fabricating mentions has the same downside as buying links: detection risk plus opportunity cost.

Myth 5: Structured data is the secret to AI Overview visibility

Schema.org markup is genuinely useful for rich results in standard Search. Recipe cards, FAQ accordions, Product price displays, How-To carousels. None of that is changing. What Google is now saying clearly is that structured data is not required for AI Overviews and there is no special schema for AI search. The agencies positioning schema as the key to AI rankings are overstating it. Use schema where it earns you a rich result. Do not stuff your pages with FAQPage and Article schema and expect AI Overview citations to follow.

What I’ve seen actually move rankings in AI Overviews (client-site notes)

Google’s guide is correct about the principles. Here is the practical take from running SEO end to end for the last six-plus years, mostly for law firms through my agency hey-ash.com, with every review and analysis you read on CriticNest researched, audited, and published by me personally. No team, no outsourced writers, no recycled press releases. Reading Google’s May 2026 guide against what I see day to day, here is where I think the document gets it most right.

Original data is the strongest leverage point. Google’s “non-commodity content” framing is essentially a request for content an AI model cannot generate from common knowledge. In practice, the cleanest examples are first-party numbers and observations: surveys you run yourself, internal benchmarks, aggregated case data, pricing tests, and reproducible methodology. AI models cannot fabricate primary data without grounding, so they cite the source. This is the single biggest gap I see on most client sites and most affiliate sites.

Named-author E-E-A-T matters more in generative responses than in blue links. A real author byline, a credible bio, and links to off-site credentials (LinkedIn, bar association, professional bodies, published work) all give the model something to attribute a quote to. That attribution layer is more load-bearing inside an AI Overview than it is on a standard SERP, because a generative response stakes the credibility of the answer on the source it cites. Anonymous content is harder for the model to confidently surface.

First-hand product testing crushes summary content. Hands-on reviews, real screenshots, pricing structures verified by logging into the actual product, and specific errors or quirks you only find by using the tool are exactly the signals Google’s guide is asking for under “unique point of view.” Paraphrased press releases and AI-spun roundups are exactly what the guide is steering visibility away from. Doing the test work yourself is the moat.

Where budget consistently underperforms. llms.txt generation projects (now officially confirmed by Google as zero ROI), content-rewrite engagements pitched as “LLM optimization,” schema-stuffing where every blog post gets FAQPage plus HowTo plus Article markup whether it qualifies or not, and link-bait blog comments or Reddit threads trying to seed brand mentions. Each of these has been pushed hard at conferences in the last 18 months. Google’s May 2026 guide explicitly names all four as ineffective.

Where budget reliably pays back. Original photography and video for product, location, and process content. Case studies with real numbers and disclosed methodology. Author bio buildouts with verifiable credentials. Internal link architecture that connects topical clusters into something the model can navigate as one coherent body of work. Technical audits to fix indexability issues that pre-date the AI era. Every dollar spent here pays back in both standard Search and AI Overviews simultaneously, because both surfaces draw from the same ranking signals.

How to rank in AI Overviews: a 12-step checklist for this week

Here is the actionable list, in priority order, that takes Google’s May 15 guide and turns it into something you can run against any property by Friday.

1. Verify the site in Google Search Console. Without this you are flying blind on indexing and impression data.
2. Confirm key pages are indexed and eligible for a snippet. Use the URL Inspection tool on your five highest-priority pages.
3. Audit your robots.txt and meta robots tags. Make sure nothing important is accidentally disallowed or noindexed.
4. Run a content commodity audit. For each top 20 page on the site, ask: could a competitor publish a substantially similar article tomorrow by paraphrasing the existing top 10? If yes, the page is commodity. Plan a non-commodity replacement.
5. Add named-author bylines with verifiable credentials on every content page. Link the author to a real bio page, LinkedIn profile, and any external authority signals.
6. Replace stock imagery with original photography and video where feasible. Especially on product, service, location, and case study pages.
7. Run a structured data review. Keep schema where it qualifies for a rich result (FAQ on FAQ pages, Recipe on recipes, Product on commerce pages). Remove duplicates and remove schema added “for AI.”
8. Test page experience. Core Web Vitals, mobile rendering, layout shift. Use PageSpeed Insights as the starting point.
9. Identify your top three commodity pages and commit to one non-commodity rewrite each over the next 90 days. Original data, customer interviews, or behind-the-scenes process content.
10. Sunset llms.txt generation projects, “AI content rewrite” engagements, and inauthentic mention buying. Redirect the budget toward original content.
11. Set up Merchant Center and Google Business Profile if you sell products or run a local business. AI Overviews surfaces both.
12. Monitor AI Overview citations weekly for your top ten target queries. Track which sources Google is citing today and reverse-engineer what makes them citable.

What’s next: agentic search, UCP, and browser agents

The final section of Google’s guide flags an emerging area that I think is more important than the rest of the document, though it gets the least space. AI agents are autonomous systems that perform tasks on behalf of users: booking a reservation, comparing product specifications, completing a checkout. They access websites the way humans do, through DOM inspection, accessibility tree parsing, and visual rendering analysis.

Google references the Universal Commerce Protocol (UCP), a proposed standard for letting AI agents transact on a user’s behalf across websites. The guide is light on specifics, but the direction is clear. Within 12 to 24 months, expect to see AI agents browsing your site to compare you against competitors, attempting to complete bookings or purchases, and surfacing the result back to a user inside a chat-style interface. If your site is hostile to agent traffic (broken without JavaScript, requires CAPTCHA on every form, has accessibility tree gaps that confuse screen readers), you will lose this traffic before you understand it exists.

The practical preparation overlaps almost entirely with good accessibility and clean HTML structure. Sites that work for screen readers tend to work for agents. Sites that require visual interpretation to navigate do not. Read Google’s agent-friendly website best practices if this is relevant to your business.

FAQ: How to rank in AI Overviews

Is SEO still relevant in the age of AI search?

Yes. Google’s May 2026 guide is the most direct statement on this question to date: optimizing for generative AI search is optimizing for the Search experience, and therefore still SEO. The same ranking and quality systems that determine standard Search visibility determine AI Overview citations.

Do I need an llms.txt file to rank in AI Overviews?

No. Google explicitly states you do not need to create any new machine-readable files, AI text files, or special markup for AI search. The llms.txt file is not required and is not given special treatment. If you have one, it is harmless. If you do not, skip it.

What is the difference between AEO, GEO, and SEO?

AEO (Answer Engine Optimization) and GEO (Generative Engine Optimization) are terms coined to describe AI search visibility work. From Google’s perspective, both are subsets of SEO. The underlying ranking systems are the same and the foundational best practices are the same.

How does RAG affect my SEO strategy?

RAG (Retrieval Augmented Generation) means AI Overviews retrieve grounding pages from the live Google index in real time. The pages most likely to be cited are the pages that already rank well for the query. Ranking signals you optimize for (content quality, link authority, freshness, indexability) are exactly the signals that determine RAG retrieval.

What is query fan-out and how should I structure content for it?

Query fan-out is the technique where AI Overviews expand a single user question into multiple related searches and blend citations from all of them. The way to benefit is to write comprehensive content on a topic that naturally addresses follow-up questions inside the same page, rather than splitting the topic across many thin pages.

Should I create separate content for every long-tail query variation?

No. Google flagged this in the guide as a potential violation of the scaled content abuse spam policy. The model understands synonyms and intent. Build depth on one strong page, not breadth across many thin ones.

Does structured data help me rank in AI Overviews?

Not directly. Structured data qualifies pages for rich results in standard Search (FAQ accordions, Recipe cards, Product details). It is not required for AI Overviews and there is no special schema that boosts AI visibility. Use it where it earns a rich result, not as a generative ranking lever.

How do AI Overviews handle images and video on my page?

AI Overviews can surface images and video clips alongside text citations. The standard image SEO and video SEO practices apply: descriptive filenames, alt text, captions, structured data where relevant, fast load times. Original media (real photography, original video) outperforms stock at the AI ranking layer.

How long does it take to see AI Overview visibility from new content?

Timelines vary by niche competitiveness, indexing speed, and whether AI Overviews already surface for the target query. Google’s grounding model retrieves pages from the live index in real time, so once a page is indexed and eligible for a snippet, it is in the pool the model can draw from. For low-competition queries with a live AI Overview, citations can appear quickly once the page is indexed. For competitive YMYL niches (legal, medical, financial), expect four to eight weeks while link signals and topical authority build. The trajectory is meaningfully faster than traditional blue-link rankings, but the exact timing depends on signals you do not control.

What is the single most important thing I can do this month for AI Overviews visibility?

Replace one piece of commodity content with one piece of non-commodity content. Original data, first-hand experience, or unique insight that an AI model could not generate from common knowledge. The ROI on that single move outperforms every other tactic on the list.

Sources and further reading

• Optimizing your website for generative AI features on Google Search (the official guide, May 15, 2026)
• Google Search Central blog
• Google Search Essentials
• Creating helpful, reliable, people-first content
• Agent-friendly website best practices

Written and published by Ashikur Rahman, an SEO operator with over six years building search visibility for law firms and AI tools. Founder of hey-ash.com, editor at CriticNest. Every article on this site is researched, drafted, and shipped by me personally. Last updated May 16, 2026.