Google's New AI Search Guide Is Right About Google. Our Copilot Data Covers the Rest.

Google has an official guide on optimizing for generative AI features in Search, first published in May, updated July 10, and making the rounds in SEO media again this week. It includes a section that names names: llms.txt files, content "chunking," rewriting copy for AI systems, chasing mentions. Google's verdict on all of it, for Google Search, is that you can ignore it. The guide's own recap puts it plainly: "Prioritize effective SEO strategies over 'AEO/GEO hacks'."

Plenty of people will read that as "AEO is dead." We read it differently, because two weeks ago we published a week of our own AI citation data showing our most-cited page in Copilot sitting at position 66 on Google. Google's guide is accurate about Google. It is also, by its own careful wording, about nothing else. Both halves matter, so let's take them one at a time.

What does Google's new AI guide actually say?

The guide says visibility in AI Overviews and AI Mode comes from standard SEO: indexed pages, snippet eligibility, crawlable content, good page experience, and unique "non-commodity" content. Its mythbusting section lists five things you can skip for Google Search: llms.txt files, chunking, rewriting for AI, inauthentic mentions, and treating structured data as an AI requirement.

The reasoning is straightforward. Google's AI features answer questions through retrieval-augmented generation, which means the model pulls pages from Google's Search index before it writes a word. If your page is in that index and ranks, it can be retrieved. If it isn't, no special file or markup rescues it. There's no secret AI door around the side of the building. For Google's surfaces, the front door is the only door.

Where is Google right?

Almost everywhere, within its own walls, and it's worth saying so out loud because a lot of what gets sold as AEO deserves the mythbusting.

Nobody should be paying an agency to write an llms.txt file for a local service business. Google says its search doesn't use them, and server log studies back that up for most site types. Nobody should be shredding readable pages into robotic 40-word fragments because a LinkedIn thread said AI prefers chunks. And the guide's warning about spinning up a page for every possible query variation comes with teeth: Google explicitly files scaled query-variation pages under its scaled content abuse spam policy.

The guide's positive advice is also the same thing we tell clients on day one: original data and first-hand experience beat recycled summaries. Google contrasts commodity content like "7 Tips for First-Time Homebuyers" with a first-person account of waiving an inspection and what it cost. That's the bar now, and honestly, good.

If your only concern is Google's AI surfaces, the guide is complete. Do real SEO, write things only you could write, and you're done. But that "if" is carrying a lot of weight.

What does the guide leave out?

Read it again and notice the qualifier. The mythbusting list opens with "Here are a few things you can ignore for Google Search," and the recap closes the same way: "For Google Search, you can ignore tactics like 'chunking' content." The llms.txt line specifies that Google Search doesn't use them. The whole list is bookended by the same scoping. That's not sloppy writing. Google is a precise publisher, and the precision is the story: this is a document about one retrieval system, written by the company that owns it, and it makes no claims about the others.

The others are where a growing share of answers come from. Copilot grounds its answers in Bing's index. Ahrefs measured the gap across 15,000 prompts: only 8.6% of the URLs Copilot cites rank in Google's top 10 for the same prompt. ChatGPT, Perplexity, and Claude each run their own retrieval with their own source preferences. A guide that's fully correct about Google's index tells you nothing about whether those systems can find you, and most businesses have never once checked.

What happens when you check both scoreboards?

We know because we ran it on ourselves. In our July 7 dataset, Copilot and its partner systems cited evoix.io 79 times in a week. The page that earned 60 of those citations averaged position 66 on Google with zero clicks in 28 days. Meanwhile the service page we most want cited had never been crawled by Google, and our GEO guide picked up Copilot citations while sitting in Google's "crawled, currently not indexed" bucket. We call this the Two-Index Problem: a site can look like a failure on one index and be quietly winning on the other, and a Google-only audit will never show you either fact.

Now hold Google's guide up against that table. Its advice would tell us the position-66 page is doing nothing and the fix is better SEO. Reasonable, for Google. But that page is simultaneously our single most productive asset on the index Google's guide never mentions. Same URL, two scoreboards, opposite verdicts.

So is AEO dead or not?

No, but the version of AEO that deserved to die is dying, and Google just helped. Here's the honest split:

If you care about	The playbook is
Google AI Overviews and AI Mode	Exactly what Google's guide says: indexation, snippet eligibility, non-commodity content, standard technical SEO
Copilot	Bing's index, which most businesses have never audited, plus Bing Webmaster Tools
ChatGPT, Perplexity, and others	Their own retrieval systems, where third-party mentions, citable original data, and clear answer structure decide who gets named
All of the above	Original, first-hand content. This is the one recommendation every engine rewards, and it's the least optional thing on this list

Notice what's not in that table: llms.txt as a flagship tactic, chunking rituals, keyword-variation page farms. Google mythbusted the snake oil, and the snake oil needed it. What survives is less glamorous: be indexable everywhere, publish things worth citing, and measure each engine against its own index instead of assuming Google's scoreboard speaks for all of them.

What should a small business actually do this week?

Fifteen minutes, three steps. First, open Google Search Console and confirm your money pages are indexed and snippet-eligible, because Google's guide is right that nothing else matters there. Second, open Microsoft Clarity's free AI Visibility dashboard and see which of your pages Copilot actually cites, because that answer will not match your Google report. Third, put the two lists side by side and look for the gaps in both directions. That comparison is the whole diagnosis, and we walked through it step by step in the Two-Index post.

Google wrote a good guide about Google. Treat it as the floor, not the map.

Stephane Morera is the founder of EVOIX, an AI consulting company in DeLand, Florida that helps small businesses get found and recommended by AI search. Want to know what AI engines say about your business? Start with our free AI readiness audit.