Field guide · May 2026
Google just told us how to actually optimize for AI search.
A plain-English breakdown of Google’s May 2026 guidance — what to do, what to ignore, and the popular “AI optimization” tactics your agency should stop charging you for.
- SEO is still relevant for AI search. That's Google's own first answer.
- AI Overviews are grounded in your existing rankings. No rankings, no citations.
- Commodity content is out. First-hand expertise and a real point of view are in.
- “AEO” and “GEO” are marketer terms. Google says it's still SEO.
- Structured data isn't required for AI features. That surprised us too.
- Five popular tactics Google specifically tells you to skip.
On May 15, 2026, Google quietly published the clearest official statement yet on how to optimize a website for AI search. The guide is short. The implications are not.
If you’ve been reading marketing newsletters this year, you’ve been told a dozen contradictory things. That you need an llms.txt file. That you need to “chunk” your content. That you need to hire an AEO specialist. That you need to rewrite every page in question-and-answer format. Most of it doesn’t survive contact with what Google actually published.
Here’s what the guidance says, what it means for your strategy, the tactics Google explicitly tells you to skip, and the questions clients send us most often.
01 — How AI search works
Two concepts behind every AI Overview
Google opens with their own question: is SEO still relevant for generative AI search? The short answer is yes, and the reason matters.
The AI features in Google Search — AI Overviews, AI Mode — aren’t separate from regular search. They’re built on top of it. Two techniques drive the whole system, and once you understand them, most of the popular AI optimization advice falls apart on its own.
Grounding (retrieval-augmented generation)
When you see an AI Overview, the model isn’t pulling answers from training data. It runs live searches against Google’s normal index, retrieves real pages, reads what’s on them, and generates a response with clickable links back to the sources.
Google calls this grounding. The technical term is retrieval-augmented generation. The practical consequence is simple: if your page isn’t already eligible to appear in regular Google Search, it cannot appear in an AI Overview either.
“A page must be indexed and eligible to be shown in Google Search with a snippet.”
That's the entry ticket to AI features. Your content has to be crawlable, indexable, and pass the basic Search technical requirements before any “AI optimization” matters at all.
Query fan-out
When someone types a conversational query, the model doesn’t run just that one search. It generates a cluster of related searches behind the scenes and combines the results into an answer.
Google’s own example: a search for how to fix a lawn that’s full of weeds might fan out into best herbicides for lawns, remove weeds without chemicals, and how to prevent weeds in lawn, all running concurrently. Your page needs to hold up across that whole constellation, not just the original phrasing.
Optimizing for generative AI search is optimizing for the search experience, and thus still SEO.
Google Search Central, May 2026
02 — What to actually do
Three foundations that move the needle
Google organizes effective work into three buckets. Get these right and you’re optimizing for AI search, by name or not.
1. Create valuable, non-commodity content
Google is explicit that this is the single highest-leverage thing you can do. They draw a sharp line between two categories:
Commodity — avoid
- “7 Tips for First-Time Homebuyers”
- “What Is Freight Brokerage?”
- “How to Choose a PEO”
- “Benefits of Professional Organizing”
- Anything an AI could write from public knowledge alone
Non-commodity — invest here
- “Why We Waived the Inspection and Saved Money” — first-hand
- “What Happens When a Bill of Lading Goes Missing” — case study
- “PEO vs ASO: Cost Breakdown From an Actual Switch”
- “Inside a 3-Day Garage Reset” — before and after
- Original data, expert analysis, lived experience
Google’s broader list of what wins here:
- A unique point of view. First-hand reviews and original takes outperform summaries of existing content.
- People-first writing — the same standard Google has applied to helpful content for years.
- Clear structure with paragraphs, sections, and headings that help a reader navigate.
- High-quality images and video. AI Overviews can surface visual content, which gives you more ways to appear.
- Don’t manufacture a page for every fan-out variation. Google names this as a scaled content abuse trigger.
2. Build and maintain a clear technical structure
This section is short in the source document, and it’s where most websites quietly fail. If Google can’t crawl, render, and index your page cleanly, none of your content investment matters.
Technical foundation checklist
- Page is indexable and eligible to appear in normal search with a snippet. This is the prerequisite for AI features.
- Crawling is unblocked. No robots.txt blocks, no rendering issues, no broken canonicals.
- JavaScript SEO best practices are followed if your site leans on a JS framework.
- Semantic HTML where reasonable. It helps screen readers and browsing agents alike.
- Page experience is solid. Core Web Vitals, mobile rendering, low latency, clear main content.
- Duplicate content is minimized so Google doesn't spend crawl budget on pages you don't care about.
- Search Console is verified so you can actually see what's happening.
3. Optimize your local and commerce details
For brick-and-mortar businesses and retailers, AI responses pull commercial and local information from structured sources first, not from your blog posts. The highest-leverage actions:
- Google Business Profile — fully complete. Hours, services, photos, products, Q&A, and reviews. This is what AI Overviews pull from for local results.
- Merchant Center feeds — every attribute populated, accurate, and current. This is the gateway to AI-powered shopping.
- Business Agent — Google’s emerging conversational experience that lets customers chat with your brand from Search. Worth exploring if your model supports it.
03 — Stop doing this
Tactics Google explicitly tells you to ignore
If your current agency is selling any of the following, point them at the documentation. The page exists. It just doesn’t say what they’re telling you.
“You need an llms.txt file so AI can read your site.”
What Google says: You don't need to create new machine-readable files, AI text files, special markup, or Markdown to appear in generative AI search. Standard HTML is enough.
If you already have one, leave it. If you don't, it isn't worth building. There are good case studies showing how little it moves.
Two caveats. This is Google's position — other engines may read it. And Google itself is inconsistent here: llms.txt isn't needed for AI features, yet Lighthouse now checks for the file in an experimental agentic-browsing audit.
“Break your content into AI-friendly chunks.”
What Google says: There's no requirement to break content into tiny pieces and no ideal page length. Google's systems already understand multi-topic pages. Write for your audience, not for theoretical AI parsing.
Content is still king, but it rots. We've watched sites that chunk aggressively lose rankings and traffic over time.
Quality over quantity. If your strategy is “more is more,” rethink it. Treat your content like a garden and yourself as the gardener — if you never pull the weeds, the flowers stop looking like much.
“Rewrite content for every long-tail variation.”
What Google says: Google's AI understands synonyms and intent natively. You don't have to capture every phrasing, and doing it at scale triggers the scaled content abuse policy.
Agreed, and this should have been settled years ago. AEO brought the tactic back in 2025 and Google is now putting a lid on it.
It's genuinely tricky, because other engines don't behave like Google and it may still work there. But this is your own copy on your own site — you can't claim to be a spam victim. Google could reasonably read it as a red flag. More measurement needed, and SEOs may be the slowest to adjust.
“Buy mentions on forums and blogs to teach AI your brand.”
What Google says: Inauthentic mentions get filtered by Google's existing spam systems before they ever influence AI features. You're paying for noise that gets stripped out.
Don't write off ChatGPT, Copilot, and Claude here. When noise is all that exists on a topic, we see it get cited.
Pay attention to the word filtered. It can't be a penalty, or your competitors would weaponize it against you. It can't hurt you. It can be ignored.
Last thought: authorship. Engines are using listicles, but looking harder at reputation and E-E-A-T signals.
“Add Schema.org markup everywhere to optimize for AI.”
What Google says: Structured data isn't required for generative AI search, and there's no special schema to add. Keep using it for rich results in regular search.
Use it where it makes sense. Don't overstuff your HTML with it.
This is where Google's left hand isn't talking to its right again — around the same time, Google announced FAQ schema is sunsetting in the SERPs. Schema matters for data communication and indexation. It is not a lever that hacks your brand to the top, and it doesn't deserve the LinkedIn hype. Ignoring it entirely, especially in e-commerce, is also a mistake.
“Hire an AEO or GEO specialist.”
What Google says: Optimizing for AI search “is optimizing for the search experience, and thus still SEO.” The work is real. The separate specialty isn't.
Let's push “SEO is dead” out a few more years.
Regardless of the engine, the interface, or whether someone types or speaks, SEO keeps evolving. That's what makes the job interesting.
And in practice: we can't name a single serious SEO who isn't already looking at answer engine work as an extension of what they do.
04 — What’s coming next
Agentic experiences and the Universal Commerce Protocol
Worth knowing about. Not worth panicking over yet.
Google is preparing for a future where autonomous AI agents — software acting on a user’s behalf — browse websites to complete tasks like booking a reservation, comparing specs, or making a purchase. These agents analyze visual renderings, inspect the DOM, and interpret the accessibility tree to work out where to click.
The emerging standard for agent-driven transactions is the Universal Commerce Protocol, and Google has confirmed it’s building toward it.
For most businesses this is a stay-informed item, not an urgent one. But if you run e-commerce or take bookings online, the groundwork pays off in both directions: clean structured data, working transaction flows, machine-parseable pricing, and semantic HTML are good for users today and required by agents tomorrow.
The “should we use ChatGPT for our blogs” question has its own answer.
This guide covers Google's optimization guidance. A separate document covers what happens when you use AI to create the content. The short version: AI-assisted writing is fine, but pure AI output with no human judgment tends to fail the quality bar and can trigger scaled content abuse penalties.
Using generative AI content is the source to point people to.
05 — Questions we keep getting
The answers we give clients
What we send when a client emails asking what they should be doing differently. Copy them. Use them. Send them to your team.
Is traditional SEO still relevant now that Google uses AI?
Yes, and Google's guidance opens by answering this exact question. AI Overviews are grounded in the normal search index using retrieval-augmented generation. The links under an Overview are real ranking results.
The practical implication: if your page isn't indexed and eligible to appear in regular Search with a snippet, it cannot appear in AI features. SEO isn't a separate game from AI search. It's the same game.
Do we need to hire someone for AEO or GEO?
No. Google's exact framing is that optimizing for generative AI search “is optimizing for the search experience, and thus still SEO.”
What people call AEO or GEO is foundational SEO executed well — strong content with real expertise, clean technical structure, and optimized local and commerce signals. If an agency sells you AEO as a separate line item, ask what specifically they're doing that good SEO doesn't already cover. The answer matters.
Should we add an llms.txt file or special AI markup?
No. Google has directly stated you don't need machine-readable AI text files, special markup, or Markdown variants to appear in generative AI features. Standard HTML and clean technical SEO is what the systems read.
If a vendor is charging for “AI-readability” services built around these files, they're selling something Google's systems don't use.
What about Schema.org structured data? Doesn't that help AI?
This one catches people by surprise. Google explicitly says structured data isn't required for generative AI search, and there's no special markup to add for AI features.
Keep using it anyway. It's still how you qualify for rich results in regular Search — products, FAQs, recipes, events, reviews — and those affect click-through. Just don't believe anyone claiming schema is the secret to AI Overviews.
Can we use AI to publish hundreds of pages targeting question-based searches?
That's one of the fastest ways to get penalized. Google's guidance names it: producing content primarily to manipulate rankings or generative AI responses violates the scaled content abuse policy. Creating a page for every fan-out variation is the example they give.
Google handles synonyms and intent natively, so you don't need separate pages for “best CRM for realtors,” “top realtor CRMs,” and “real estate agent CRM software.” One comprehensive page outperforms ten thin ones and avoids the penalty entirely.
How do we get our products or local business featured in AI results?
AI features pull commercial and local data from structured sources first. In order of leverage: complete your Google Business Profile (hours, services, photos, products, Q&A, review responses); maximize your Merchant Center feed if you sell online, with every attribute and variant accurate; explore Google's Business Agent rollout; and make sure foundational SEO is solid, because AI features can't pull from a site that isn't indexed.
What's query fan-out, and should we optimize for it?
Query fan-out is what happens behind the scenes when someone enters a conversational search. Google's model silently runs several related searches at once and combines the results into an answer.
You don't optimize for it by writing more pages. You optimize by writing deeper ones. A piece that thoroughly covers a topic, its sub-topics, common objections, and edge cases lands in more fan-out branches than ten shallow pages each chasing one phrase.
Our competitor publishes thin AI content and ranks anyway. What gives?
Two explanations usually cover it. Enforcement isn't immediate — sites can ride thin content for months before a core update or manual action catches up. And what looks thin from outside sometimes carries quality signals you can't see: real reviews, original imagery, internal expertise, brand authority.
Either way, betting the business on the strategy Google's policy explicitly targets is a poor risk profile. Recovery from a scaled content abuse hit is long, expensive, and uncertain.
What should we actually invest in?
Four priorities, in this order.
First-hand expertise content. Case studies, original research, behind-the-scenes process, before-and-afters, proprietary data. Anything an AI can't synthesize from the public web.
Technical foundation. Indexable pages, clean crawling, working JavaScript rendering, solid Core Web Vitals, minimal duplication. Without this, nothing else matters.
Local and product surfaces. Business Profile and Merchant Center are where AI features pull commercial information, and they're chronically underinvested.
Topical depth, not breadth. Fewer pages, each comprehensive enough to land in the fan-out branches around your topic.
Note what's missing: schema-as-secret-weapon, AEO retainers, llms.txt files, content chunking, mention campaigns. Google tells you to skip all of it.
AI Search Optimization
Find out where your site actually stands
We'll audit your domain against Google's current AI-search criteria — content depth, technical foundation, Business Profile and Merchant Center health, and scaled-content risk. Then we'll tell you straight which of the five tactics above you're already paying someone for.
Book an AEO CallNo commitment. If your foundation is already solid, we'll say so.
Primary source: Google Search Central — Optimizing for generative AI features on Google Search, last updated May 15, 2026.
Related Google documentation: Using generative AI content · Creating helpful, reliable, people-first content · Scaled content abuse spam policy · Search technical requirements.
Last reviewed May 2026. Google's AI search guidance is still moving. If you're reading this more than six months out, check the primary source for the current version.
