Google's AI Search Guide: What to Trust, What to Ignore

After months of speculation, conference hints and contradictory advice from every corner of the SEO internet, Google has finally published its first official guide to optimising for generative AI features. If you've been wondering whether to panic about AI Overviews, build out a separate strategy for AI Mode, or pay for one of the new "GEO" services that have popped up overnight, this is the document that's supposed to clear all of it up.

The short version of Google's message is reassuring. Good SEO is good AI search. The same crawlable site, the same useful content, the same advice to focus on humans rather than chasing every algorithmic trend. There's no secret formula. No magic file you need to upload. No new discipline to learn from scratch.

It's a relief, honestly. And most of it is genuinely solid advice.

But anyone who has spent more than five minutes in this industry knows there's usually a gap between what Google says publicly and what their systems actually reward. We've been here before. Plenty of times. Which is why I think this guide deserves a careful read rather than a copy-paste implementation.

This is my honest breakdown of what the guide actually says, what's worth applying, and where I'd take Google's word with a healthy pinch of salt.

What the new Google AI search guide actually says

The guide, titled "Optimising your website for generative AI features on Google Search", was published in May 2026 and covers both AI Overviews (those summary boxes at the top of search results) and AI Mode (the conversational follow-up version that now appears for logged-in users).

The most important thing to understand is how Google's AI features actually work, because it changes how you should think about everything else in the document. There are two mechanisms at play. The first is retrieval-augmented generation, or RAG, which is a fancy way of saying that when someone asks a question, Google pulls relevant pages from its existing search index and uses them to ground the AI's response. The second is something called query fan-out, where Google generates a handful of related questions behind the scenes and runs those searches too, then pulls all of the results together into a single answer.

Both of these mechanisms rely on Google's existing ranking systems. There is no separate AI algorithm sitting underneath, deciding which pages get featured in AI Overviews. There's one ranking system, and the AI features draw from it. Which means the better your site ranks normally, the more likely you are to be cited in an AI response.

AEO and GEO are "still SEO"

The guide is unusually direct on this point. Google defines AEO (answer engine optimisation) and GEO (generative engine optimisation) explicitly, and then states that "from Google Search's perspective, optimising for generative AI search is optimising for the search experience, and thus still SEO."

For Google specifically, that's probably true. There's no separate framework you need to learn. There's no special schema. There's no GEO-specific strategy that's going to outrank good fundamentals.

The caveat I'd add — and Google doesn't, because Google is talking about Google — is that ChatGPT, Perplexity and Claude work differently. They have their own retrieval mechanisms, their own data sources, and their own ways of deciding what to cite. So if visibility in those platforms matters to your business, AEO and GEO still mean something there. Just don't pay anyone to optimise specifically for Google's AI features as if it's a separate discipline. That's the bit that's marketing fluff.

What Google says you don't need to do

This is where the guide gets genuinely useful. Google explicitly names five tactics it says you can ignore. In plain English:

You don't need an llms.txt file. These are special text files that some agencies have been pushing as essential for AI visibility. Google says directly that it doesn't treat them any differently to any other file on your site.

You don't need to "chunk" your content into tiny pieces for AI systems to understand it. Google's models are perfectly capable of reading a normal page and picking out the relevant sections.

You don't need to rewrite your content specifically for AI. The AI systems understand synonyms and meaning. You don't have to capture every variation of how someone might phrase a question.

You don't need to buy or chase inauthentic mentions of your brand across the web. Google's spam systems block this, and the AI features rely on the same ranking signals.

You don't need to obsess over structured data. Schema markup is still worth doing for rich results in normal search, but there's no AI-specific schema you need to add.

A word of caution on Google's "don'ts"

Most of these are probably accurate. I'd happily tell a client to skip llms.txt files and stop worrying about chunking. But it's worth remembering that Google has previously told people not to bother with things that turned out to matter quite a lot. Domain authority, click data, even backlinks at various points, all dismissed at one stage or another, all later revealed (one way or another) to be more important than we'd been led to believe.

The "don'ts" in this guide aren't gospel. They're Google's current public position, which is a slightly different thing.

What Google says to focus on

Strip away the AI vocabulary and the guide's positive advice is essentially classic SEO with a fresh coat of paint.

Google puts the most weight on what it calls non-commodity content. Its own example is brilliantly clear. A piece called "7 Tips for First-Time Homebuyers" is commodity content; generic, summarisable, easily replicated by anyone with a ChatGPT account. A piece called "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line" is non-commodity, it's specific, personal, and built on actual experience.

The implication is unsubtle. If your content could have been written by anyone, the AI systems have no particular reason to choose yours. If it could only have been written by you, you've got a genuine edge. For the businesses we work with; hospitality, lifestyle, property, service brands, this is excellent news. The first-hand experience is right there in the day-to-day of running the business. Most of you just aren't capturing it on your website yet.

On the technical side, the guide repeats what you'd expect. Your pages need to be crawlable, indexable, fast and well-structured. Semantic HTML is helpful but not essential. Page experience matters. JavaScript-heavy sites need extra care. Duplicate content is a waste of crawl budget.

For local businesses and ecommerce, Google recommends keeping Merchant Center feeds and Google Business Profiles in good shape. AI responses can pull product listings and local business information from these sources, so the optimisation effort you put in there benefits both traditional search and AI features.

None of this is revolutionary. But the framing matters. Google is essentially saying that if your fundamentals are solid, you're already doing AI search optimisation. You don't need to add more. You need to do the basics properly.

Why I wouldn't take any of this as gospel

Here's where I'd encourage some caution.

Google has a long and well-documented pattern of saying one thing publicly while their internal systems do something rather different. This isn't a conspiracy theory. It's a matter of public record at this point.

In May 2024, a vast trove of Google's internal API documentation leaked online. SEO professionals including Rand Fishkin at SparkToro and Mike King at iPullRank spent weeks going through it, and what they found contradicted years of Google's official statements.

There was a feature called siteAuthority sitting in the code, despite Google reps having spent years telling the industry that "domain authority" wasn't a thing they measured.

There was a system called NavBoost that used click data as part of how it ranked results, despite multiple public statements from Google that clicks weren't used for ranking.

There were signals related to author authority, Chrome browser data, and freshness boosts for newer content, all of which had been variously denied or downplayed in official communication over the years.

This is on top of the gap between Google's long-standing "just make great content" rhetoric and the reality that EEAT signals, brand strength, link profile and topical depth are doing significant heavy lifting in practice. Anyone who has actually worked on an SEO campaign knows that great content alone, without the wider trust and authority signals around it, often doesn't move the needle on its own.

So when Google publishes a guide telling you that certain things don't matter, I think the responsible thing is to read it as Google's current public position rather than a complete picture of how their systems work. Some of those "don'ts" will hold up. Some might not. We won't really know which is which until practitioners test them in the real world over the next twelve months.

This isn't a reason to ignore the guide. It's a reason to read it with your eyes open.

So what should small businesses actually do?

Strip the noise away and there are really five practical steps for any small business reading this.

Read the guide and apply the principles. They're directionally right. Make useful content. Keep your site clean. Don't pay for "GEO hacks."

Get the basics right before chasing anything else. If your site isn't fast, crawlable and reasonably well-structured, no amount of AI-specific tactics will help you. The boring fundamentals are still the highest-leverage thing you can fix.

Treat the "don'ts" as low priority rather than gospel. I wouldn't spend money on an llms.txt file. But I also wouldn't aggressively rip structured data out of a site that's already using it. Don't take the absence of an instruction as a reason to make changes.

Test in your own niche. Search the questions your customers are actually asking. See what's appearing in AI Overviews. Look at which sites are being cited. That data is more honest than any official guidance.

Be wary of agencies selling AI-specific services. If someone is trying to sell you "GEO" as a separate discipline from SEO, ask what they're actually doing that's different from standard SEO best practices. In most cases, the honest answer is "not very much."

The bottom line

Google's new AI search guide is useful. It's the clearest public statement we've had about how their AI features work and what they actually reward. There's real signal in it, and most of the advice is good.

But it's one input. It's not the final word, and it's certainly not gospel. The businesses that will win in AI search aren't the ones that follow Google's documentation word-for-word. They're the ones who treat it as a useful reference, apply common sense, and test what actually works for their own audience.

That's how we approach search at Ninety8. We don't lift strategy from Google's blog and hand it to you with a smile. We take the time to understand your business, look at what's actually happening in your niche, and build a search strategy that's specific to you, not recycled from whatever the industry happens to be excited about this month.

If you're tired of generic advice and want someone who will do the work properly, let's have a conversation.