Straight from Google: Want AI Search to Cite You? Don't Kill SEO — but 5 Things Really Need to Stop

A couple of days ago, I scrolled past an official document from Google.

The document has a notoriously long name: "Optimizing your website for generative AI features on Google Search." In plain English — do you want your site to be cited by AI search features like AI Overviews and AI Mode? If yes, Google has put together an official set of recommendations for you.

My first reaction after reading it: man, this document is a wake-up call for everyone panicking about whether "SEO is dead."

A bracing wake-up call, to be precise. The kind that sobers you up.

First, the question you actually want answered

Is SEO still useful?

Almost every friend I know who runs a website has been asking this over the past year. The moment AI search arrived, people freaked out — if everyone goes and asks an AI, who's going to click through to my page? Does doing SEO even mean anything anymore?

Google's own answer is: yes, it's useful. More foundational than ever, in fact.

Why do they say that?

Because Google's AI search features are not built from scratch. They run on the same core ranking system Google has always used. AI just layers two things on top of it.

One layer is called RAG (Retrieval-Augmented Generation). In plain terms — when the AI answers you, it first has to use Google's ranking system to retrieve a batch of relevant pages, then read the information on those pages, and finally organize it into an answer for you. It doesn't make things up out of thin air.

The other layer is called query fan-out. When a question comes in, the AI automatically splits it into several related queries and goes fetching results for all of them at once.

An example will make it obvious.

You search "my lawn is full of weeds, what do I do." The AI won't just look at that one sentence — it'll also go search "which lawn herbicide is best," "how to kill weeds without chemicals," "how to prevent weeds in your lawn." It scoops all of that up, then hands you a combined answer.

Do you see what's happening?

The essence of AI search is still about fetching web pages. The only difference is that before, you clicked the links yourself; now the AI clicks for you, reads for you, and summarizes for you.

So whose pages does the AI pick? The ones with good content and solid rankings, of course.

SEO isn't dead. SEO is the foundation of AI search.

So what kind of content does Google actually want?

The answer is so plain you might not believe it.

What Google wants is "non-commodified" content.

What's commodified content?

The kind that runs under a headline like "7 tips for first-time homebuyers." A hundred versions of it are online, all written more or less the same way, and you could publish any of them just by swapping the title. It's built on common knowledge, anyone could write it, and it adds no unique value for the reader.

So what's non-commodified?

"Why I waived the home inspection and ended up saving money — a field investigation of sewer lines." Only someone who actually lived through that could write a headline like this. It carries first-hand experience and the kind of details only an insider would know.

Google puts it bluntly: their AI systems read a lot of sources, so you need a unique angle that can stand out. First-hand reviews, personal retrospectives, the pitfalls only insiders know — that's the material AI is scrambling to cite.

Flip it around: if what you write could just as easily be generated by an AI, on what grounds would you get cited?

And that's the first — and heaviest — piece of advice in the document. The gist of the original line is this: the content itself matters more than every other recommendation in this guide combined.

In one sentence — stop writing what a machine could write.

On the technical side, a few things still need doing

Content covered, now the technical stuff. Google states it clearly: the way AI systems access your pages is the same way Google's crawlers always have. So all those old-school technical SEO practices still work.

I'll pull out the most critical ones from what the document keeps hammering.

First, your page has to be indexed by Google and able to show a featured snippet (the summary box at the top of search results) before it's even eligible to appear in AI features. That's the entry bar.

Second, your site has to be crawlable. AI models depend on public, crawlable web pages to learn patterns and give answers. If you block crawlers, you don't even have a ticket to get in. If your site is especially large or updates very frequently, you also need to check whether your crawl budget (how much of your site Google will crawl in a given timeframe) is being wasted.

Third, semantic HTML. There's an interesting point here — Google itself says your pages don't need to be written as perfectly semantic code; the whole internet is full of messy HTML, and Google can still read it fine. But the closer you get to spec, the better it is for tools like screen readers. Don't obsess over code details for the AI's sake; put human readability first.

Fourth, if you use JavaScript, the relevant SEO standards still apply. Just because it can be processed doesn't mean you can stop worrying.

Fifth, page experience. How fast it loads, whether it renders correctly across devices, whether the body content is clearly separated from the rest of the junk on the page — all of this figures into the score.

Sixth, cut duplicate content wherever you can. A crawler's bandwidth is finite — don't let it drain away on duplicate URLs.

For local businesses and e-commerce, Google adds a specific note: feeding product data through Merchant Center and maintaining business info via Google Business Profile makes you more visible both in AI answers and in regular search results.

None of these is new. It's precisely because nobody wants to do this slow grind that it's become the barrier to entry.

Now for the myth-busting segment. These 5 things, Google says outright, don't work

This is the part of the document I found most valuable.

After AI search blew up, a pile of new acronyms surfaced online: AEO, GEO, Answer Engine Optimization, Generative Engine Optimization. A dozen different names, all selling the same thing — "optimization specially built for AI."

This time, Google called out the most popular "hacks" by name and said: we don't recognize these.

Let me walk you through them one by one.

One: llms.txt, and any "AI-specific" special files or special markup.

Some people say you need to drop an llms.txt file in your site's root directory to tell the AI what your content is about. Google says: our search doesn't use this thing at all. It will discover and crawl all kinds of files, but that doesn't mean it gives any one file special treatment.

Two: "chunking" your content.

There's a claim that you should slice your articles into small chunks so the AI can understand them more easily. Google's response: unnecessary. Our system can understand multiple topics within a single page and will automatically pick out the relevant part. Make your article as long as it needs to be — based on what your reader wants, not what the AI wants.

Three: rewriting content specifically for AI.

Some people worry that they "haven't stacked enough long-tail keywords" or "haven't covered every possible search phrasing." Google says the AI understands synonyms, and understands what someone is looking for, even if the words you use don't exactly match what the user searched. Stop chasing every variant.

Four: mention-farming.

That means going around to various blogs, videos, and forums getting people to mention your product and say nice things about you, hoping to influence the AI. Google says: yes, we do show online discussions of products, but those are natural discussions. Mention-farmed buzz can't move the needle — the core ranking system weighs high-quality content, not gamed mentions.

Five: stuffing structured data.

Some people feel they have to add AI-specific schema.org markup. Google says: AI search does not require structured data, and there is no "exclusive schema" for it. But structured data should still be used where it makes sense, because it helps you earn rich results in regular search. That's a side benefit — not something you do for the AI.

Do you see the common thread across these five?

They all treat AI as a brand-new species that needs to be courted separately. But Google has been telling you all along: AI search is just the core search system with a layer of AI on top. Court the core system — that's enough.

Courting the core system, in plain terms, comes down to the most unglamorous thing of all: make the content good, and make the technical foundation solid.

One more thing Google dropped in passing: AI agents

Toward the end of the document, Google brought up something fairly cutting-edge — AI agents.

What's an AI agent? It's an AI system that can do things on a person's behalf. Like booking a restaurant for you, or comparing the specs of two products.

What's interesting is that this kind of agent accesses your website differently than a human would. It looks at screenshots, scrapes the DOM structure, and reads the accessibility tree (the data tree screen readers use to navigate a page). Some are also starting to use emerging protocols like the Universal Commerce Protocol, which lets search agents do more directly on your site.

If you run the kind of business that agents will go after, the document suggests you take a look at the agent-friendly best practices.

This area isn't mature yet. But the direction is worth keeping an eye on: the visitors to your website in the future won't only be humans — they'll also be machines running errands on humans' behalf.

Finally, Google's own one-line summary

The document wraps up with a closing note, and I'll translate its meaning directly: a lot of content ranks perfectly well in Google Search (AI search included) without doing any noticeable SEO at all. You don't have to do every single thing on this list.

In other words, Google itself is telling you not to be anxious.

So what's actually worth being anxious about?

It's this: if you're still writing "7 tips for first-time homebuyers" — the kind of thing anyone could write — then no matter how AI search evolves, you won't stand out.

Google's document boils down to a single sentence: go back to the content itself. Go back to the thing only you can write — the thing no one else (AI included) could produce.

SEO isn't dead. What's dead is the kind of SEO that was all about tricks and never about content.