What Google's 2026 AI Search Guide Adds
Google published its first official guide for AI Overviews and AI Mode on May 15, 2026. The headline reaction was the myth-bust. The more useful read is additive: here are the extras Google specifically asks for on top of the AEO playbook you already run for ChatGPT, Claude, and Perplexity.
On May 15, 2026, Google Search Central published its first official, consolidated guide for optimising for generative AI features on Google Search. It is a short document, but it lands hard -- because it tells the industry what is actually different about AI Overviews and AI Mode versus regular Google, and what is identical. The honest read: most of what your team already does for ChatGPT, Claude, and Perplexity still applies. Google is asking for a small number of extra things on top.
5
things Google explicitly says you do NOT need to do for AI Overviews -- including llms.txt and AI-specific markup
This is not a "scrap your AEO stack" post. The guidance is Google-only. ChatGPT, Claude, and Perplexity still consume llms.txt, still reward structured data, still respond to the broader GEO playbook -- because they are not Google AI Overviews. What follows is the additive read: what Google added on top of the playbook you already run.
Scope: this guide is about Google, not "all AI"
Read the title carefully: Optimizing your website for generative AI features on Google Search. The document covers AI Overviews and AI Mode. Both are features of Google Search, built on top of Google's core ranking systems. The guidance does not extend to ChatGPT, Claude, Perplexity, Grok, or any other LLM platform -- and Google does not claim otherwise.
The distinction matters, because on the non-Google platforms, the tactics Google calls "not required" are genuinely effective. Anthropic, GitHub, Perplexity, Stripe, and Zapier all publish their own llms.txt files. Cursor, Continue, and Aider actively read them. For those audiences, our llms.txt guide still applies as written.
"From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO." -- Google Search Central, May 2026
How AI Overviews and AI Mode actually pick sources
Two mechanisms do the work. Both are documented for the first time in this guide.
Retrieval-augmented generation (RAG). Google grounds every AI response in pages from its Search index. Models do not "remember" your brand -- they retrieve documents in real time and quote them. That means the eligibility bar is identical to classic Search: the page must be indexed and snippet-eligible. If your page is blocked from regular search results, it is also blocked from AI Overviews — same index, same gate.
Query fan-out. Instead of running a single query, AI Overviews issues many related sub-queries and stitches the answers together. A single user question can pull supporting links from a much broader set of pages than the classic ten blue links. That is good news for second-page content with strong topical depth.
The extras Google specifically asks for
This is the part worth acting on. None of the items below are reframings of older SEO advice -- they are net-new asks tied to the AI features.
1. Agent-ready websites
The guide explicitly links out to web.dev's agent-friendly UX article. AI agents are the next user type Google is preparing for -- autonomous systems that interpret, plan, and act on a user's behalf. They perceive sites through three channels: vision (screenshots), the DOM, and the accessibility tree.
Concrete asks: use real <button> and <a> tags (not <div onclick>), associate every form input with a <label for=...>, keep click targets above 8 square pixels, avoid ghost overlays, and keep layouts stable. The line that matters: everything Google suggests to make a site agent-ready also makes it better for humans. That is a useful sanity check.
2. Snippet eligibility, treated as primary
Snippet eligibility used to be a footnote. In this guide it is the primary requirement. Three Google directives tell search engines not to show a text preview of your page: nosnippet (a meta tag that says "no preview at all"), max-snippet:0 (the preview length capped at zero characters), and data-nosnippet (an attribute you wrap around specific sections of HTML to keep them out of previews). The snippet is exactly what an AI Overview quotes from, so any of those directives on a key landing page opts that page out of AI Overviews entirely. Audit them against your most valuable pages and remove them unless you have a specific reason.
3. Google-Extended controls (explicit opt-out)
Google-Extended lets you opt out of having your content used for AI training while staying in regular Search. Most sites do not want this -- it removes one of the few places Google proactively learns about your brand -- but the guide makes the lever explicit for sites that do.
4. Preferred Sources (publishers + news brands)
The Preferred Sources feature lets users pin a domain so its coverage ranks higher in news queries. Publishers can promote with a downloadable button and a deeplink (google.com/preferences/source?q=example.com). For news and publisher brands, this is a real lever.
5. Universal Commerce Protocol (ecommerce)
Google's guide points at Universal Commerce Protocol as the emerging standard for agentic commerce. Early days, but worth tracking if your buyers are likely to delegate purchases to agents.
What Google says you do NOT need (for Google AI specifically)
Google is direct here. Five tactics are called out as "not required" for AI Overviews and AI Mode. We are repeating the list verbatim because it is the part of the guide that drew the loudest industry reaction, including Search Engine Journal's "AEO and GEO are still SEO" framing.
- llms.txt files. "You do not need to create new machine readable files, AI text files, markup, or Markdown" for Google AI features.
- Content chunking. Google's systems "understand the nuance of multiple topics on a page" -- artificially splitting content into chunks does not help retrieval.
- AI-specific rewrites. No need to capture every long-tail variant or write in a specific machine-style. Write for people.
- Inauthentic mentions. Mass-produced brand mentions across the web violate the scaled content abuse spam policy.
- Special schema.org markup for AI. Structured data still helps rich results -- but there is no "AI-only" schema lever.
The footnote that almost everyone missed: these statements are about Google's AI features. For ChatGPT, Claude, and Perplexity, llms.txt still has documented uptake, and schema markup still meaningfully changes citation patterns. The advice from those of us tracking the non-Google platforms is unchanged.
What still matters (and Google leans into hard)
The fundamentals do not change. The guide reinforces them:
- Helpful content. The Who / How / Why framework -- clear authorship, transparent automation disclosure, and content that serves people first -- is now the explicit quality bar for AI inclusion.
- E-E-A-T with trust paramount. Author bylines, credentials, dated content. Same playbook we cover in our E-E-A-T playbook for AI citations.
- Technical foundations. Crawlability, canonical URLs, sitemap.xml, robots.txt, HTTPS, mobile viewport. None of these are new -- they are now non-negotiable.
- High-quality images and video. Google explicitly calls these out as supporting AI inclusion when relevant.
What this means if you track AI visibility
One sentence: Google's optimisation playbook is not the same as ChatGPT's, Claude's, Perplexity's, or Grok's -- and any single-platform view of AI visibility is going to mislead you. The honest read of Google's guide is that AI Overviews map closely onto traditional SEO, while the other platforms continue to reward the broader AEO/GEO toolkit. The brands winning on AI right now treat the five major platforms as five separate surfaces with overlapping but distinct rules.
That is also why we built Ranqo as a multi-platform tracker from day one. Our Page Audit now includes Agent Readiness checks (semantic HTML, label association, click affordance, skip-to-content) derived directly from Google's 2026 guidance -- alongside the citation-readiness, structured-data, and authority checks that move the needle on the non-Google platforms.
Audit your page against the new asks
Run any URL through Ranqo's free AI Readiness audit to see how your page scores on the six categories that now include agent-readiness -- semantic interactive elements, form label association, cursor affordance, and skip-to-content links.
Try the free AI Readiness auditWritten by
Nisha Kumari
Nisha Kumari is Co-Founder at Ranqo, where she leads growth strategy and client acquisition. With a background in digital marketing and financial management, she specializes in SEO, Generative Engine Optimization, and helping brands build visibility across AI platforms.
Share this article
Related articles
How to Optimize for Google AI Mode
Google AI Mode decomposes every query into a multitude of sub-queries before it answers -- in Google's own words. Yet every ranking how-to recycles the same syndicated list that conflates AI Mode with AI Overviews. The optimization target isn't the head query; it's the fan-out set behind it. Here's the playbook, plus what Google itself says about llms.txt and AI Mode.
Peec AI Alternatives: 8 Tools Compared (2026)
Peec AI tracks prompts well and stops there, which is fine until you want to know why you are missing. These eight were not picked, they were computed: sort every platform we track by how much of the loop it closes, take the top eight. We build one of them, and there is a section on where Peec still wins.
Reddit and Forums Out-Cite Your Entire Website
Reddit and community forums are 3.3% of citations in our published study. That sounds small until you see what it sits next to. A brand's entire own website is 2.9%. Forums also outrank Wikipedia and software review sites. Video still leads overall, and the honest read is that the sources shaping how AI describes you are places other people publish.