Open any marketing newsletter in 2026 and you’ll be sold the same panic. Old SEO is dead, AI search runs on its own secret rules, and you need a new GEO playbook before your competitors get one. There’s a course, a checklist, a “semantic proximity” framework, and no shortage of voices online swearing that llms.txt is the new sitemap. Guess what? Almost none of it holds up.
Optimizing for generative AI search is still SEO. That is Google’s own position, not a marketer’s hot take. In two official guides published in late 2025 and updated through June 2026 — the guide to optimizing for generative AI search and its companion on how AI features work — it explains that AI Overviews and AI Mode run on the core Search ranking systems, not a separate engine. The same guides sort the popular GEO tactics into what works and what doesn’t. The ones they reject by name include llms.txt files, content “chunking,” and special AI schema, and they confirm there is no separate AI crawler you need to court.
This piece walks through what those guides actually say. How your pages get pulled into an AI answer, the three high-hype tactics Google just named as useless, and the short list of moves that genuinely improve your visibility.
Is GEO different from SEO?
No. In Google’s own words, “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” AI Overviews and AI Mode (the AI-written answers Google now shows at the top of the results page) run on Google’s core ranking and quality systems, not a separate AI engine. They retrieve pages from the regular Search index using a technique called grounding, then generate an answer with clickable links. There is no special AI crawler, no required schema, and no secret format. The single biggest lever Google names is unique, first-hand, non-commodity content.
How Google Actually Pulls Sources Into AI Answers
Google pulls sources into AI answers using the same system it uses to rank everything else. AI Overviews and AI Mode (the AI-written answers Google now shows at the top of the results page) don’t run on a separate engine. Google says both features are “rooted in our core Search ranking and quality systems.” In plain terms, the same Google that has always decided which websites show up for a search is the one deciding which websites get named inside an AI answer. That is the reassuring part. Nothing about your website suddenly became obsolete, and once you see how Google picks a page to use, the rest makes sense.
The Crawlers That Really Matter: Googlebot vs. Google-Extended (and the “AI Crawler” Myth)
There is no special “AI crawler” for Google Search, and that single fact dissolves a lot of the hype. Here is how the crawling actually works, and the controls you really have.
There Is No Separate “AI Crawler”
The bot that fetches your pages for AI Overviews and AI Mode is the same Googlebot that has crawled the web for years. A crawler is just the automated program a search engine sends out to read and copy your pages so it can index them. Google is explicit about this. “AI is built into Search,” its documentation says, “which is why robots.txt directives for Googlebot are the control for site owners.” If Googlebot can reach a page, that page is eligible to show up in an AI answer. If you block Googlebot, you vanish from regular Search too. You can’t have one without the other.
What Google-Extended Actually Controls
The one crawler name that genuinely confuses people is Google-Extended, and it does something narrow. Google-Extended is not a crawler in the usual sense. It’s a setting in your robots.txt file that controls whether your content can be used to train and ground some of Google’s other AI products, the broader Gemini ecosystem, rather than Search itself. Blocking Google-Extended does not remove you from AI Overviews, because those run on Googlebot. It only opts you out of training and grounding elsewhere.
The Controls You Actually Have
So what should you do? Allow Googlebot to crawl. If you want to limit how much of a page Google can quote inside an AI answer, use the controls Google provides. The nosnippet tag tells Google not to show a text snippet of a page at all. The data-nosnippet attribute lets you wrap a single paragraph so that one passage stays out of snippets while the rest of the page stays eligible. And noindex removes the page from Search entirely. Those are the levers. Everything else marketed as “AI crawler optimization” is noise.
Grounding and Query Fan-Out: How RAG Picks What to Cite
Google picks what to cite using two techniques it now names openly: grounding and query fan-out. Grounding is how it gathers real pages to build an answer from. Query fan-out is how it turns one question into many at once. Together they decide whether your page becomes a source or stays invisible.
Grounding: The AI Answers From Real Pages, Not From Memory
Grounding is Google’s word for a mouthful of a term, retrieval-augmented generation, usually shortened to RAG. Before it writes a single word, the AI pulls a handful of live web pages from Google’s normal index, reads what they say, and builds its answer out of what it found rather than from memory. Say a small-business owner asks Google’s AI which email platform is best for a Shopify store. The AI doesn’t recite an answer from training data; it pulls up current pages that actually compare email tools for ecommerce, reads what they report, and assembles its answer from those pages. Google describes it as relying “on our core Search ranking systems to retrieve relevant, up-to-date web pages,” then reviewing “the specific information from those retrieved pages.” Those clickable links under an AI Overview are the pages it pulled from, and that link is your citation.
To earn that citation, three things all have to be true. Your page has to be in Google’s index. It has to rank well enough to get pulled in among the handful the AI actually reads. And it has to contain the specific, quotable fact the answer needs. Miss any one of the three and the AI grounds on someone else’s page instead of yours.
Query Fan-Out: One Question Becomes Many Searches
Query fan-out explains why thin pages lose. When someone asks a question, Google doesn’t run a single search. It quietly fires off a cluster of related searches at the same time to cover the subtopics. Google’s own example is a query about fixing a lawn full of weeds. The fan-out searches behind it might include “best herbicides for lawns,” “remove weeds without chemicals,” and “how to prevent weeds in the lawn.” The AI gathers results across all of them and assembles one answer.
That changes what your page is eligible for. A page that answers only the surface question is in the running for one of those searches. A page that also answers the surrounding questions can be pulled in for several of them at once, which is several more chances to be the source the AI names.
What This Means for How Deep Your Page Should Go
Cover the whole cluster of questions on one strong page. You don’t write a separate flimsy page for each little version of a question, because Google treats that kind of mass-produced filler as spam. You write one thorough page that genuinely answers the real questions a person has about the topic, laid out in clear sections the AI can pull from. One deep page beats ten thin ones, for the reader and for fan-out alike.
Debunking the Myths: Three High-Hype GEO Tactics Google Just Killed
Google has now publicly named a list of “GEO hacks” that don’t work, and three of them dominate the sales pitches you’re hearing in 2026. Plenty of agencies are still charging good money for these exact tactics, so they’re worth knowing cold. Google’s language is blunt. Many circulating tactics “aren’t effective or supported by how Google Search actually works.” Here are the three to watch for, and what to do instead.
Why “Keyword Stuffing for AI” Does Nothing
Cramming keywords and query variations into a page to win AI citations is a waste of time, and it can get you penalized. The pitch sounds plausible. If AI runs many fan-out searches, surely you want every keyword variation on the page so you match more of them. Google says no. Its systems “understand synonyms and general meanings of what someone is seeking,” so you “don’t have to worry that you don’t have enough long-tail keywords or haven’t captured every variation.” The AI connects intent to meaning, not to exact-match strings.
The deeper risk is the part the hype skips. Producing a swarm of near-identical pages, each tuned to a slightly different query variation, is exactly what Google’s scaled content abuse spam policy targets. Scaled content abuse is Google’s term for generating lots of low-value pages mainly to manipulate rankings rather than to help readers. Google warns that doing this “primarily to manipulate rankings or generative AI responses” violates that policy and is “an ineffective long-term strategy,” because “a high quantity of pages doesn’t make a website higher quality.” So the tactic doesn’t just fail. It puts you in spam territory. Write for meaning, cover the real questions in one place, and skip the keyword tricks.
Why Fake Mentions and AI Link Networks Get Ignored
Buying mentions, planting brand name-drops across forums, or paying into an “AI citation network” doesn’t earn you AI visibility. This one is trickier, because there’s a grain of truth underneath it. Google confirms its AI features “can show what’s being said about products and services across the web,” in blogs, videos, and forum threads, so real third-party discussion of your brand does matter. The hype industry took that grain and built a manufactured-mention economy on top of it.
Google addresses it directly. “Seeking inauthentic mentions across the web isn’t as helpful as it might seem,” because its core ranking systems “focus on high-quality content while other systems block spam,” and the AI features lean on both. In plain terms, the same spam defenses that have spent two decades discounting paid links and fake reviews also sit between a bought mention and an AI citation. The folklore version of this, the idea that there’s a hidden “mention count” you can purchase your way up, is not how the system works. There is no subscribed citation signal you can buy.
What actually earns mentions is the boring, durable thing. Real coverage. A genuinely useful page that a journalist, a forum regular, or a podcaster references because it helped them. A first-hand result worth talking about. Those mentions survive the spam filters precisely because they’re authentic. For most businesses, the fastest route to being mentioned across the web is being worth mentioning, which loops right back to content quality. If you want help building that kind of earned presence rather than buying a fake one, that’s the work our content marketing team does.
Why Hidden “AI-Only” Text (and llms.txt Tricks) Backfire
Hiding text on your page for the AI to read, or leaning on an llms.txt file as your AI strategy, ranges from useless to actively harmful. Start with the harmful end. Cloaking, which means showing one version of a page to Google and a different one to human visitors, is a long-standing spam violation. The newer flavor of it is stuffing invisible “AI-optimized” text into a page, white-on-white paragraphs or off-screen blocks meant only for the crawler. It backfires for a concrete reason Google states in its AI guidance. Your structured data, the behind-the-scenes labels that describe your content to search engines, has to “match the visible text on the page.” Hidden content that contradicts what users see is exactly the mismatch Google penalizes.
Then there’s llms.txt, the year’s most oversold file. The idea was borrowed from robots.txt and proposed a special text file that hands AI systems a tidy summary of your site. Google’s verdict is unambiguous. “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search, as Google Search itself doesn’t use them.” It will neither help nor hurt your Google visibility, because Google ignores it. Keep one if another service you care about reads it, but never let it sit at the top of your AI strategy.
The same skepticism applies to the broader “rewrite everything for machines” pitch, the one about chunking your content into tiny passages and adding exotic schema. Google says there’s no required chunking, no ideal page length, and “no special schema.org markup you need to add” for AI search. Notice the pattern across all three myths. Every one of them is a shortcut aimed at the algorithm instead of the reader. Google’s whole message is the opposite. Its systems reward pages written for people, in plain language.
The Actionable Playbook: What Google Says Actually Moves the Needle
Google’s actual playbook comes down to three moves: lead with the answer in plain language, stand behind a credible author and brand, and publish non-commodity, first-hand content. None of it is exotic, and that’s good news if you’re not technical. All of it is hard to fake, which is exactly why it keeps working.
Answer-First, Plain-Language Structure
Lead every section with the answer, write in clear prose, and organize the page with real headings. Google’s guidance asks you to “organize content in a way that helps your readers,” noting that people “appreciate it when web pages are organized by paragraphs and sections, along with headings that provide a clear structure.” This isn’t an AI trick. It’s basic readability that happens to make your page easy for grounding to extract a clean answer from.
The mechanism connects back to how retrieval works. When grounding pulls your page and looks for the specific fact an answer needs, a section that states its point in the first sentence is far easier to lift than one that buries the point in paragraph four. So put the conclusion up top, then explain it underneath. Write the way you’d answer a customer who asked you the question to your face: plainly, the answer first, the reasons after. Do that and you’re already optimized, without ever thinking about AI. Google is clear that you do not need to write “in a specific way just for generative AI search.” Clear writing for humans is the optimization.
Build a Real Author and Brand Entity (E-E-A-T)
Make it obvious who wrote the page and why they know the subject. Trust is the one thing that survives every filter between your page and an AI citation. Google scores this with a framework called E-E-A-T. The letters stand for Experience, Expertise, Authoritativeness, and Trustworthiness, which is really just Google’s way of asking one plain question. Do the person and the business behind this page actually have the standing to say it? Part of that comes down to being a recognizable “entity.” An entity is simply a real, identifiable thing Google can put a name to and connect to other things it already knows. A named author. A real company. A clear topic. The opposite is an anonymous block of text with no one behind it.
In practice, that comes down to a few concrete things. Put a real, named author on the page with a genuine bio, not a faceless “admin.” Keep your business details the same everywhere Google can find them. Google specifically points to keeping your Business Profile “up-to-date” as a way to show up in AI answers for local questions. Show that you’ve actually done the thing, rather than just claiming you’re an expert. Google rewards the first and shrugs at the second. A page that proves the author has real, first-hand experience carries weight that no piece of code can fake. Structured data (the behind-the-scenes labels on your page) can help describe all this, and it’s worth keeping, but Google says plainly it’s “not required” for AI search and it won’t stand in for the real thing. The substance is what counts. The labels are just tidy packaging on top.
The Non-Commodity, Information-Dense Standard
Publish content a model couldn’t have written without you. Google names this as the single biggest lever for AI visibility. Its language leaves little room for interpretation. Creating content people find “unique, compelling, and useful will likely influence your website’s presence in generative AI search in the long run more than any of the other suggestions in this guide.” That wording is deliberate. More than any other suggestion. Everything else supports this one thing.
Google even hands you the test. Commodity content, its example being “7 Tips for First-Time Homebuyers,” is built from common knowledge anyone or any model could assemble, and it adds little. Non-commodity content, its example being “Why We Waived the Inspection and Saved Money: A Look Inside the Sewer Line,” delivers a first-hand, specific take that goes beyond common knowledge. The difference is lived experience and real information the internet didn’t already have. One is a summary. The other is a source.
For a business owner, that should come as a relief. You can’t out-publish a content farm, and you don’t have to. What you have that a model lacks is your own results, your own data, your own hard-won opinions, the inspection you actually waived. Put those on the page. Lead with a real number or a quoted primary source instead of vague generalities. Cover the real questions a reader has so you’re eligible across the fan-out, the way a proper AI search strategy layers depth into one page. Stop trying to be the page that summarizes the topic. Be the page the summaries have to cite.
TL;DR
- GEO and AEO are just SEO. Google states it directly: optimizing for generative AI search is optimizing for the search experience. There is no separate AI ranking system.
- AI Overviews and AI Mode run on Googlebot, grounding (RAG), and query fan-out. No special AI crawler. Allow Googlebot, control quoting with nosnippet / data-nosnippet / noindex, and know that Google-Extended only governs training and grounding for Google’s other AI products.
- Three GEO myths are dead per Google: keyword stuffing for AI (ineffective and spam-adjacent), manufactured mentions and paid “citation networks” (blocked by the same spam systems), and hidden AI-only text plus llms.txt tricks (cloaking risk, and Google ignores llms.txt).
- The real playbook is three moves: answer-first plain-language structure, a credible named author and brand entity (E-E-A-T), and non-commodity, first-hand, information-dense content.
- The number one lever is non-commodity content. Google says unique, expert, first-hand content influences AI visibility more than anything else in its guide. Be the source, not the summary.
Ready to Win AI Search Without the Hype?
If you’ve been handed a GEO proposal full of llms.txt files, schema “for the AI,” and a monthly mention-building retainer, now you know how much of it Google has publicly written off. The work that actually earns AI citations is the same work that has always earned good rankings, done well and grounded in something only your business can say.
Adotme builds that for national DTC brands and local businesses alike. Our organic SEO services cover the technical and content foundations AI features actually reward, and our writing team specializes in the non-commodity, first-hand content Google names as the top lever. If you want a second opinion on whether your current AI strategy is real or repackaged hype, reach out to our team and we’ll tell you straight.
Frequently Asked Questions
Is GEO different from SEO?
No. Google’s official guidance states that “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” GEO (generative engine optimization) and AEO (answer engine optimization) are vendor terms for the same discipline. Google’s AI features run on its core ranking and quality systems, so the work that improves your AI visibility is the same work that improves your regular Search visibility. Treat any “GEO hack” that contradicts standard SEO with suspicion.
Is there a special AI crawler I need to optimize for on Google?
No. Google states that “AI is built into Search,” and the same Googlebot that crawls for regular Search also fetches pages for AI Overviews and AI Mode. There is no separate AI crawler to court with special files. The one related name, Google-Extended, is a robots.txt control that only governs whether your content trains and grounds Google’s other AI products, not Search. Allow Googlebot and you’re eligible for AI features.
What is query fan-out, and why does it matter?
Query fan-out is Google’s term for running several related searches at once to answer one question. For a query about fixing a weedy lawn, Google might also search “best herbicides for lawns” and “how to prevent weeds.” The AI then assembles one answer from results across all those searches. It matters because a page that thoroughly covers the surrounding questions can be retrieved for multiple fan-out searches at once, while a thin page competes for only the surface query.
Does llms.txt help me rank in Google’s AI features?
No. Google’s documentation is explicit: “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search, as Google Search itself doesn’t use them.” An llms.txt file will neither help nor hurt your Google visibility because Google ignores it. It’s fine to keep one if another service you use reads it, but it should never be the center of your AI strategy.
Does adding more schema markup improve my AI Overview citations?
Not on its own. Google says structured data “isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” Keep schema for its legitimate uses, such as rich-result eligibility and clear entity labeling, and make sure it matches the visible text on the page, since mismatches get penalized. But schema is a label on your content’s quality, not a substitute for it. The content itself is what gets cited.
What actually improves my visibility in AI Overviews and AI Mode?
According to Google, the biggest lever is unique, non-commodity, first-hand content that goes beyond common knowledge. Beyond that: clear answer-first structure with real headings, a credible named author and consistent brand details (E-E-A-T), crawlable pages, and content that genuinely covers the questions readers ask. Leading each section with specific facts and a clear, direct answer helps as well.
External references: Google Search Central — Optimizing for generative AI search · Google Search Central — AI features and your website · Google Search Central — Scaled content abuse spam policy · Google Search Central — Robots meta tag, data-nosnippet, and X-Robots-Tag
