Official Google Guidance on AI Search

One of the biggest questions in SEO over the past few years has been what to do to optimise for AI search?

Because the thing is, if there are no rules and no guidance, then people just make things up. That’s how it’s always been. Until, of course, the powers that be catch up with the shenanigans on the interwebs and do something about it.

Like that time when ChatGPT was mercilessly used to generate tons of content by people with zero writing talent. A win for them but a major loss for seasoned copywriters with decades of experience.

But then, eventually, Google cracked down on this sort of behaviour with its scaled content abuse updates. It took about 18 months if I recall correctly, and that’s far too long to wait for an ambulance if you have an emergency condition, but when they did finally deal with this abuse of technology, I smiled. Schadenfreude. Karma.

So yes, Google has been, at times. telling us what we _can’t_ do.

But what is it that we _can_ do?

Since late 2022 we’ve had no guidance, until now…

Optimizing your website for generative AI features on Google Search

So what have we been doing all these years to optimise for AI search?

Well, we’ve been keeping our ears to the ground, lingering in the fora, reading industry news, and gathering anecdotal and otherwise evidence.

Some of the things we’ve been doing have been adding schema markup, generating and adding LLMs.txt files of markdown site synopses, and even “chunking” content so that it’s readable and quotable for use in AI search.

I read a piece recently that said LLMs.txt has “little to no effect” on AI search inclusion. I was dubious of its use, but I erred on the side of using it anyway to see what the results might be, It’s an unofficial but proposed standard, so it was worth a try as a “low effort” benefit. But now supposedly, it doesn’t work.

And schema has been battered for a while now. FAQ schema can still be created and added to webpages, but only official Government and health industry websites have had their FAQ schema read and in the results at Google. For all other industries we’ve placed the schema but supposedly had no impact on AI search. And now, only this week, Google killed off FAQ schema for good, saying it has zero influence or bearing on the search results. My personal caveat here is, if it’s not for the SERPs, then what else might it be a relevant use case for? And if not Google, then who else finds it useful? I don’t know.

So we try things, see if they work, and Google tells us no.

But they never say yes. Until yesterday’s release of the official guidance Optimizing your website for generative AI features on Google Search

The Official Google AI Optimisation Guide

Yes, it’s official. And it’s very welcome too. After nearly four years of scratching around looking for answers, processes, and procedures, Google has finally given us a definitive direction.

First up, in their guidance, Google said

“…people are increasingly gravitating to generative AI experiences to help them find information”

I see that. I use AI for certain searches where I’m starting to feel underwhelmed by traditional search results. I know that if I have a complex question, then Claude, Gemini, or Google’s AI Mode will give a more satisfactory answer.

Only this week I spoke to a client who said “my kids don’t use Google any more”. Which was really interesting because, the very next day I saw a chart from the online trend platform. Similarweb, where they charted AI search traffic over the past 12 months. I’m going to pop the chart in here because it’s a fascinating read.

The takeaways from this data are:

ChatGPT’s dominance in AI search traffic, over the past 12 months, ahas slipped from 77.6% to just 53.7%
Google’s Gemini, as a standalone and power behind AI Mode, and AI Overviews, has increased from 7.27% to 26.7%
Claude was just 1.37% a year ago, behind Deepseek, Grok, and Perplexity, but is now the #3 AI at 7.95%

That’s a lot of change in a year. ChatGPT still accounts for almost twice as much traffic as Gemini, but the gap has narrowed rapidly and we have Google I/O 2026 next week, so what new things will the Mountain View giant reveal?

But back to the official AI guidance from Google.

SEO is Still Best for AI Search

The first big statement from Google is that SEO is still the best grounding for AI search success. In their exact words, we should “apply foundational SEO best practices to generative AI search”. So that’s 25 years’ worth of my SEO experience that’s still useful to you.

But they continue, saying first and foremost about content creation that:

“Creating content that 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’s a massive thing to say and bring a smile to my face. But they actually continue and specifically point out the guidance for “unique, compelling, and useful” content. I won’t go into detail right now, but the headline points they raise are:

Provide a unique point of view,
Create non-commodity content that’s helpful, reliable, and people-first,
Organise content in a way that helps your readers,
Add high-quality images and video,
Focus on what your users want, and avoid overdoing it,
Remember Google’s scaled content abuse spam policy.

They add that it’s actually OK to use generative AI tools for content and that it should meet Search Essentials standards and spam policies

Additionally, Google advise that we create well-structured web pages. Again, they provide a lot of nuance, but in essence they ask that we all continue to:

Meet the Search technical requirements,
Follow crawling best practices,
When it comes to semantic HTML, focus on human readability and don’t worry about perfect code,
If you’re using JavaScript, be sure to follow JavaScript SEO best practice,
Provide a good page experience,
Reduce duplicate content.

And finally, they skim over, but it’s important, that we also ensure that we pay attention to:

Local business and
Ecommerce details.

These are all really sound SEO principles that I recommend and bake into any work I’m doing on your website.

Generative AI Myths

But there’s a bonus round from Google in their AI optimisation guidance, and it’s this – a really handy and revealing list of myths, chiefly:

LLMS.txt files and other “special” markup,
“Chunking” content,
Rewriting content just for AI systems,
Seeking inauthentic “mentions”,
Overfocusing on structured data.

In fact, they preface this section of their guidance by saying the following:

“As generative AI search evolves, so have the theories and practices, and sometimes, the misconceptions, surrounding it. While terms like Answer Engine Optimization (AEO) or Generative Engine Optimization (GEO) are common online, many suggested “hacks” aren’t effective or supported by how Google Search actually works.”

Once more, I’ve not detailed these as you can read them for yourself and I don’t want to simply parrot the guidance. But these last points are really nuanced.

On LLMs.txt, if it’s no skin off your nose to add it, then do it – it really is a 5-minute job and it doesn’t hurt. The caveat here is what if some generative AI tools do use LLMs.txt or will do in the future? There’s no harm in 5 minutes’ work every month to refresh this file.

Chunking is an interesting one, because it’s good to keep words on a page in relevant sections that are easy to read and digest. That’s just good copywriting, But don’t write with chunking in mind, just write. I must admit I have advised one client to do “chunking” but they ignored me anyway,

Rewriting just for AI sounds spammy anyway, so just don’t do it, and the same for mentions – these are just like spammy comments and paid backlinks, not my thing, and you shouldn’t either.

And on the last item, structured data, or schema, they’re not saying don’t do it. Just don’t dwell on it too much.

Last Word

This is all really useful stuff from Google and very welcome too. In effect they’re saying “just keep doing top-drawer SEO” which I totally agree with and have said in the past. However, I will admit that in fear of missing out (FOMO) I have chased and promoted chunking,. LLMs.txt, and schema in the hope that they’d add some benefit in AI search.

And back to that section about AEO/GEO – In a 2,410 word document, they mention:

AEO 4 times,
GEO 4 times,
AI Search 21 times.

And that, boys and girls, is why I focused on AI SEO last year rather than GEO, AEO, or LMNOPO. I did talk about the emerging GEO for another agency I once worked at when I wrote all their digital marketing blog posts, but I was careful not to overegg it. Since evolving my own freelance SEO offerings, I’ve been more cautious. If you chase one definition you run the risk of it being the wrong one. Just because “everyone else” is doing it, doesn’t mean it’s right. That’s why Google guidance is invaluable.

If you want to be included in AI search, as well as continuing to keep up your visibility and rank in “regular search” just let me know. Leave a message via my contact form and I’ll get back to you as soon as I can to discuss SEO, AI search, content, and anything digital for your website.