Auto Alpha AdvisoryA business owner asked me whether generative engine optimisation is a real thing or whether they're being sold snake oil. The problem is real, but a lot of what gets sold to fix it is ritual. Google has now said in writing that for Google Search, GEO is SEO, and it has named the popular extras Google Search doesn't need. On ChatGPT, Claude and Perplexity, Google's guide says nothing, and no checklist can tell you whether those engines name your business. To find that out, you have to measure it.
What Google says you can skip
Google's Search Central changelog dates the guidance. On 15 May 2026 Google added a guide on optimising for generative AI features, including a section that mythbusts common AEO and GEO misconceptions. On 15 June 2026 it added a note on llms.txt. The guide itself was last updated on 10 July 2026 and covers AI Overviews and AI Mode.
The mythbusting section of Google's guide lists five things you don't need to do: create llms.txt files or other special markup, chunk your content into small pieces, rewrite content just for AI systems, chase inauthentic mentions, or overfocus on structured data.
On llms.txt, Google's guide is blunt. You don't need new machine-readable files or markup to appear in Google Search, "as Google Search itself doesn't use them." On chunking, the same guide says "There's no requirement to break your content into tiny pieces for AI to better understand it." On length, it says "There's no ideal page length."
Google's guide also has the sentence that answers the snake-oil question: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO." It adds that Google's generative AI features are "rooted in our core Search ranking and quality systems."
As a CA(SA), I read that the way I'd read an issuer's own disclosure: it outranks any vendor's summary of it. If someone sells you GEO for Google Search as a discipline separate from SEO, Google's documentation disagrees with them.
This replaces my own earlier advice
In March I published a GEO explainer that rated structured data as critical, named a word-count sweet spot and counted llms.txt as a GEO signal. The April follow-up leaned on structured data too. For Google Search, Google's guidance now contradicts those points, and this post supersedes that advice. I'd rather correct it in public than leave it standing.
Why llms.txt is still in my audit engine
My audit engine's GEO score still checks server rendering, schema, citation density and llms.txt. That doesn't contradict Google, because those checks aren't only about Google Search.
Google's guide says it's "completely fine" to keep llms.txt files "for other services or systems that use these files", and that doing so will neither help nor harm your Google Search visibility. Google's own Chrome team treats llms.txt as something for agents rather than for search. Chrome's Lighthouse documentation calls it "an emerging convention used to provide a machine-readable summary of a website's content, specifically designed for LLMs and AI agents", and marks the audit not applicable when the file is missing, since providing one is optional.
Whether anything actually reads the file is another matter. The Ahrefs study of server logs across 137,000 domains found about 38,000 with a valid llms.txt, and 97% of those received no requests for it at all in May 2026. The Ahrefs study also notes that "no major AI platform has ever committed to reading it."
So the check stays in my audit as cheap housekeeping for AI agents. It isn't a ranking lever or a citation promise. If someone prices an llms.txt file as your fix for AI visibility, that's the snake oil you were worried about.
Structured data is the same story, with one real Google benefit on top. Google's guide still recommends it "as part of your overall SEO strategy, as it helps with being eligible for rich results on Google Search." That's reason enough to keep checking for it. It's no reason to expect AI citations: in my study of 832 AI answers, pages with FAQ schema averaged 3.6 citations and pages without it averaged 4.2.
What Google's guide can't tell you
Google's guide can't tell you how engines Google doesn't run will treat you. It describes itself as best practice for "generative AI features in Google Search (such as AI Overviews and AI Mode)". ChatGPT, Claude and Perplexity crawl, retrieve and cite on their own terms, and they don't behave alike.
ChatGPT often skips retrieval. In my 832-answer study, run on 5 June 2026, ChatGPT grounded 50 answers in sources while Gemini and Perplexity grounded 208 each. In an earlier study of three South African industries, it answered from memory without citing a source roughly half the time. When a model answers from memory it doesn't read your site, so changing your pages today won't touch that answer.
A strong Google ranking also helps more on some engines than on others. Semrush's July 2025 comparison found Perplexity had over 91% domain overlap and 82% URL overlap with Google's top 10, while ChatGPT had the weakest overlap of the platforms it tested. On that evidence, ranking well on Google does a lot more for you on Perplexity than on ChatGPT.
Each engine runs its own crawlers, and they do different jobs. OpenAI's crawler documentation separates OAI-SearchBot, which "is used to surface websites in search results in ChatGPT's search features", from GPTBot, which gathers training data, and Anthropic and Perplexity draw the same line. A site can turn one away without realising it: in that same earlier study, two property agencies were blocking AI crawlers without knowing. A blocked crawler doesn't by itself explain how an engine answers about you, but it is the first thing to rule out. Rendering differs too. Vercel's December 2024 analysis of crawler traffic on its network found that "none of the major AI crawlers currently render JavaScript", whereas Google renders pages in headless Chromium before indexing them. On that data, a page built on client-side JavaScript can look fine to Google and empty to the AI crawlers.
For a business owner, the biggest gap is this: Google's guide says nothing about whether AI answers in your category name you at all. It points you to a Generative AI performance report in Search Console for Google's own features, and nothing in that report covers ChatGPT, Claude or Perplexity.
What moved citations when I measured
The things that moved citations in my own studies weren't rituals. Question-matched pages won: service pages named for the thing itself, and articles shaped exactly like the buyer's question. Ask an engine about subscription versus ownership solar and it cites the article called subscription-vs-ownership. Freshness helped too. In the 832-answer study, pages updated within the last three months averaged 6 citations against 3.6 for stale ones.
The sites that act as middlemen in a category mattered. In insurance, hippo.co.za appeared in 41 answers. In one crowded market, ten third-party sites carried nearly a third of every citation. In thin markets without those middlemen, a small brand could become the definitive source itself. A reputation nobody had written down counted for nothing.
Most of that is what Google already calls SEO: current pages that answer a real question, and a presence on the sites your buyers already trust. The difference is whether you can see the result. Search Console shows you how you perform in Google's AI features. On the other engines, the only way to see it is to ask them the questions your buyers ask and count the answers.
The practical split
For Google, do SEO properly and stop paying for the extras Google says you don't need. Make pages crawlable and server-rendered, write them to answer the questions buyers actually type, and keep them current. Use structured data where it earns rich results, and watch the Generative AI performance report.
For ChatGPT, Claude and Perplexity, start with access. Check that your robots.txt and firewall aren't turning away the crawlers that fetch pages for AI search answers (OAI-SearchBot for ChatGPT, Claude-SearchBot for Claude, PerplexityBot for Perplexity), and that your main pages render without JavaScript. GPTBot and ClaudeBot are training crawlers, and blocking them is a separate decision. Then measure. Ask the same buyer questions every month, record whether you're named and which sources get cited, and fix what the numbers point at. If a competitor keeps appearing through a comparison site you're not listed on, get listed. If the engines cite a competitor's page because it answers the question better than yours, write a better page.
Don't buy an llms.txt package sold as the answer, a rewrite that chunks your site into fragments, schema sold as a citation lever, or a campaign to manufacture mentions. On the last one, Google's guide says that seeking inauthentic mentions across the web "isn't as helpful as it might seem."
If you want the Google side done properly, that's my SEO & GEO work. If you want to know whether ChatGPT, Claude, Perplexity and Gemini name you when a buyer asks, AI visibility tracking measures that every month.