The short answer
Google's 2026 AI search guide asks for what its ranking systems already reward: content with a point of view a generative model could not have produced, plain structure, and a current Business Profile. It also names the tactics it ignores, llms.txt among them. Most of what we changed afterward was what we stopped doing.
On May 15, 2026, Google published a page it had never needed before: an official guide to appearing in its own AI features, AI Overviews and AI Mode included. Most of it says what not to bother with. One paragraph says what gets a site penalized. We publish for these features every week, so we read Google's AI search guide the way a contractor reads a code update, checking what we already do, what turns out not to matter, and what is now a violation. This post covers what the guide asks for, what it says to ignore, where the scaled content line now sits, what we changed on our site, and how to read the Generative AI report in Search Console. For the discipline itself, start with our plain explainer on AEO.
What does Google's AI search guide actually ask for?
One hard thing and a few familiar ones. Google's generative AI guide, published May 15 and last updated July 10, asks for "unique, non-commodity content" and defines it with a real estate example. Commodity content is "something like '7 Tips for First-Time Homebuyers'," which "could originate from anyone." Non-commodity content is "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line," a page only the person who waived that inspection could have written. The next line is the one we pinned above the desk: "Don't just recycle what others on the internet have already said, or could easily be produced by a generative AI model."
The rest is the SEO you already know. Write for people, use paragraphs, sections, and headings, and add images or video where they carry something the text cannot. Keep your Business Profile current, because Google says Business Profile and Merchant Center data can surface in AI responses. Google's position is that appearing in generative AI features is still SEO, not a second discipline. That matches what we see on the organic search and local visibility side: pages pulled into an AI Overview already rank for the question underneath.
What does Google say does not help?
Five things, in a section headed "Mythbusting generative AI search: what you don't need to do." It names AEO and GEO and says many hacks sold under those labels "aren't effective or supported by how Google Search actually works."
llms.txt and other special files. "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search." Keeping one for other systems is fine, and "Google Search ignores them."
Chunking. "There's no ideal page length," and no need to break content into small pieces for AI.
Rewriting for AI, long-tail variants included. "You don't need to write in a specific way just for generative AI search." The systems handle synonyms.
Chasing mentions. Seeking inauthentic mentions across the web "isn't as helpful as it might seem."
Special schema. None is required for generative AI search. Keep what you have for rich results.
Two caveats. This is Google describing Google, and ChatGPT and Perplexity retrieve differently, so the cross-engine citation playbook covers what those assistants quote. And "does not help" is not "hurts," so the only question is whether the hours are worth it.
Where is the line between variants and scaled content abuse?
This is the part with teeth. Right after saying you do not need long-tail variants, Google adds: "While it might be tempting to create separate content for every possible variation of how people might search (for example, by focusing on other queries that people have asked, or fan-out queries), doing so primarily to manipulate rankings or generative AI responses in Google Search violates Google's scaled content abuse spam policy."
A lot of programmatic SEO is built on that idea. The spam policies page, last updated August 28, defines scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users," and lists generative AI tools as its first example. Fifty hand-written city pages that say the same thing are the same violation as fifty generated ones.
Google enforced this in August. Its Search status dashboard lists the August 2026 spam update as running from August 18 to August 21, and Google described it as a normal spam update, all languages, no new policies. There has been no core update since the May one finished in early June. If a site dropped in late August, check scaled or doorway content first, then run the older helpful content self-assessment.
What did we keep, stop, and add on our own site?
We checked our own blog and service pages against the guide. Most survived, one thing got demoted, and the biggest change was to a temptation rather than anything shipped.
| Practice | Call | Why |
|---|---|---|
| Direct answer at the top of every post, question-phrased headings, on-page FAQ | Keep | Serves the reader first, and it is what AI features lift |
| BlogPosting schema with a named author and a stated role | Keep | Not needed for AI features, still earns rich results and names the author |
| llms.txt, built from the same source as our sitemap | Keep, off the Google list | Google ignores it. Other engines and tools read it |
| Metro or trade copies of one post, rewrites "for AI," chasing mentions | Stop | Copies are Google's scaled content example. The other two do nothing |
| A modified date on material revisions, a monthly read of the Generative AI report | Add | The schema carries the date. The report is the only Google-side measure |
The item we argued about was llms.txt. It stays for the engines that read it, and it left our proposals as a Google lever. The metro copies were the closer call. Spinning every post into Orange County, Los Angeles, and San Diego versions was tempting, and the fan-out sentence closed that door before we opened it. One post per question, one post per angle, which is also how we scope our AI answer visibility work for clients.
How do you measure it in Search Console's Generative AI report?
Impressions by page, and that is all so far. Google announced the report on June 3 for a subset of sites, and the help page now reads "As of August 31, 2026, we've rolled out these insights to all websites worldwide." It counts how often links to your site appeared in a generative AI feature on Search, AI Overviews and AI Mode included, by page, country, device, and date. No clicks, no click-through rate, no queries.
It is not in the Search Console API either. The search type filter still accepts only web, image, video, news, discover, and Google News, so you read this report in the UI or use its export button. Google mentions "adding additional metrics over time," with no date attached.
The AI impressions are a subset of the main Performance report, so the useful read is the ratio. Once a month we sort both by page and look for pages where most impressions now come from AI features. Those pages are being summarized rather than clicked. The Pew click data on AI Overviews shows what that does to traffic, and the follow-up question is whether the quoted passage carries your name and a reason to call.
FAQ
Does Google's AI search guide say llms.txt helps?
No. The guide lists llms.txt under things you can ignore for Google Search and says Google does not use such files. Keeping one is allowed and "will neither harm nor help" your visibility or rankings. Other engines and tools may read it, which is the only reason to maintain one.
Is AEO pointless now that Google calls the hacks myths?
Google calls the hacks myths, not the goal. Its guide says appearing in AI features is still SEO, earned with the same content, structure, and site health that earn rankings. What changes is measurement, and how consistently your business is described across the web, which still decides citations on engines outside Google.
What counts as scaled content abuse under the 2026 spam policy?
Google defines it as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users," however they were made. A page for every variation of how people might search, made primarily to manipulate rankings or AI responses, is the AI guide's own example.
Can I see clicks from AI Overviews in Search Console?
Not in the Generative AI report. It shows impressions by page, country, device, and date for AI Overviews and AI Mode, and nothing about clicks or queries. The data is not in the Search Console API either. Google has said more metrics may come over time and has not given a date.
A review of your pages against Google's guide is the first step of every engagement we run. We tell you what to keep, what to stop, and what is worth adding, on a written scope and a fixed price.