Google’s AI Search Guide: What Business Owners Should Actually Do

For about two years, business owners have been told that AI search needs a whole new playbook. New file formats. New ways of writing. New agency services with three-letter names.

Now Google has put its position in writing. Its Search Central documentation includes a guide called Optimizing for AI features in Google Search, and a large part of it is a list of things you can stop worrying about.

We offer an AI Visibility (GEO) service, so you might expect us to downplay that. We’d rather do the opposite. We read the guide line by line, compared it with what we’d published on our own blog, and corrected the places where we’d overstated things. This article is the plain-English version of what we found, and what it means if you run a business in Cape Town, Johannesburg, Austin or anywhere else.

What does Google’s AI optimisation guide actually say?

The guide covers Google’s AI features in Search, mainly AI Overviews and AI Mode. Its core argument is that these features are built on the same foundations as normal search results. If your pages are in Google’s index and can appear with a snippet, they can be used in an AI answer. If they can’t, no amount of “AI optimisation” will change that.

Google frames it in a way that’s worth repeating word for word: optimising for generative AI search is still SEO. The guide tells site owners to focus on what visitors would enjoy, find helpful and feel satisfied with.

That sounds obvious. It isn’t how much of the industry has been talking.

The myths Google has now addressed directly

Here’s what the guide says you don’t need to do, with our honest take on each.

1. You don’t need an llms.txt file

llms.txt is a proposed text file that summarises a website for large language models. It spread quickly in SEO circles, and some providers started charging to set one up. Google’s guide says Google Search doesn’t use it. It won’t help or hurt your rankings.

Could another AI company read one someday? Possibly. But if someone is selling it to you as a way into Google’s AI Overviews, that’s not what Google says.

2. You don’t need to “chunk” your content

A popular tactic has been to break pages into short, self-contained fragments so AI systems can supposedly lift answers more easily. Google says its systems can understand the nuance of multiple topics on a page. You don’t need to chop a good guide into disconnected pieces.

This doesn’t mean structure is useless. Clear headings, a short summary at the top and readable paragraphs help people. They just aren’t a secret AI lever.

3. You don’t need to rewrite your copy for machines

Google’s systems understand synonyms and general meaning. Writing awkward, keyword-heavy sentences “so the AI understands” tends to make pages worse for the people reading them, and Google says it doesn’t need it.

4. You don’t need a page for every way a question can be asked

Imagine a plumbing business in Durbanville that publishes separate pages for “geyser burst what to do”, “burst geyser help”, “geyser leaking emergency” and so on. That’s the pattern Google warns against. The guide says not to create separate content for every variation of a query, and that a high number of pages doesn’t make a site higher quality. One thorough, genuinely useful page on burst geysers will do more.

5. Structured data isn’t required for AI features

Schema markup has been sold as the foundation of AI visibility. Google’s guide says structured data isn’t required for generative AI search.

We still add schema for clients, because it removes doubt about business details and can make pages eligible for rich results. We’ve written about that in our schema markup guide for business owners. But it’s a supporting detail, not the reason anyone gets cited.

6. Inauthentic mentions don’t work

Paying for mentions on low-quality sites, seeding your name in forums, or spinning up fake “best of” lists won’t get you into AI answers, according to Google. Genuine reviews and real coverage are a different thing entirely, and still worth earning.

What Google says you should do instead

Area	What the guide recommends	What that looks like for a small business
Indexing	Make sure pages are indexed and eligible for a snippet	Check key service pages in Search Console. Fix “noindex” tags, blocked pages and broken sitemaps.
Crawling	Follow crawling best practices; keep content publicly accessible	Don’t hide prices or service details behind logins, PDFs or images of text.
Content	Unique, non-commodity, people-first content	Write what only you can: local pricing, real process, honest limits, examples from your work.
Media	Add good images and video where they help	Real photos of your team, premises and finished jobs instead of generic stock.
Page experience	Good experience across devices; semantic HTML; reduce duplicates	A fast mobile site matters even more on South African data costs.
Business data	Use Merchant Center and Google Business Profile; keep details accurate	Correct hours, categories, services and product feeds.
Measurement	Use the generative AI performance report in Search Console	Check it monthly alongside your normal clicks and impressions.

Not one of those rows is new. That’s rather the point.

Why “non-commodity content” is the part to take seriously

If there’s one phrase in the guide worth sticking on the wall, it’s this idea of content that goes beyond common knowledge.

AI Overviews summarise. If your page says exactly what twenty other pages say, there’s very little reason for Google to point anyone to you specifically. Generic “10 tips for choosing an accountant” articles were already struggling. They’ll struggle more now.

Compare two pages from a Cape Town accounting firm.

The first explains what a tax practitioner does, in general terms. It’s accurate, and it’s on thousands of other websites.

The second explains what the firm actually sees when small businesses miss provisional tax deadlines, what SARS correspondence typically looks like afterwards, and how they handle it, with a realistic fee range. Nobody else can write that page, because nobody else has that experience.

The second page is what Google is describing. It’s also the page a prospective client is more likely to trust.

Does this mean GEO is a scam?

No, but it does mean some of what’s sold under that name is.

Here’s how we’d split it. For Google’s own AI features, Google is clear: do good SEO. There’s no hidden layer. Anyone selling a separate “Google AI Overviews optimisation” package built on llms.txt files, chunking and schema bundles is selling things Google has said it doesn’t use or need.

ChatGPT, Perplexity, Claude and other assistants are different companies with their own systems. Google’s guide doesn’t govern them, and none of them publishes a full rulebook. What we can say from testing is that the same foundations tend to carry over: clear information about what you do and where, consistent business details across the web, and genuine reviews and coverage.

So the useful part of AI visibility work isn’t a set of tricks. It’s two things. First, solid SEO, done properly. Second, regularly checking what these AI tools actually say about your business, and fixing the inaccurate or missing information at the source. That second part is what ranking reports don’t show you, and it’s where our AI Visibility service spends most of its time.

What we changed on our own site after reading it

We’d be in no position to write this if we hadn’t checked our own work first.

Going back through our blog, we found a few places where we’d leaned too hard on the “structure your content for AI extraction” idea, and where we’d described schema as central to AI visibility. Those weren’t outrageous claims at the time, but they don’t match Google’s published position. We’ve updated them, including our articles on staying visible as AI Overviews take clicks and getting mentioned by ChatGPT.

We’d suggest every business do the same with its own site and with any agency proposal on the table. If a plan leans on things Google has explicitly ruled out, ask why.

A practical checklist for the next 30 days

1. Confirm your important pages are indexed. In Google Search Console, inspect your homepage, main service pages and top blog posts. If one isn’t indexed, nothing else on this list matters for that page.
2. Look at the generative AI performance report. It sits under Performance in Search Console and shows how often your pages appeared in AI Overviews and AI Mode, by page, country and device. It counts impressions only, not clicks, and newer or smaller sites may see no data until they pass Google’s threshold.
3. Find your commodity pages. Pick the five pages that could have been written by anyone. Add what only you know: prices, process, local detail, mistakes you see customers make.
4. Merge near-duplicates. If you have several thin pages chasing variations of one question, combine them into one strong page and redirect the rest.
5. Fix your Google Business Profile. Hours, categories, services, photos. Our Google Business Profile optimisation service covers this if you’d rather hand it over.
6. Check speed on mobile. Run your site through our free foundation checker and fix the biggest problem first.
7. Ask the AI tools about yourself. Put your customers’ real questions to Google, ChatGPT, Gemini and Perplexity. Write down what’s wrong or missing, then correct it on your site and profiles.
8. Cancel anything built on myths. llms.txt setup fees, chunking rewrites, mass page generation. Google says these don’t help in its search.

Common mistakes we expect to keep seeing

Treating AI search as separate from SEO. Businesses that split budgets between “SEO” and “GEO” often end up paying twice for overlapping work, with neither done well.

Publishing more instead of better. AI writing tools make it cheap to publish hundreds of pages. Google allows AI-assisted content as long as it meets its spam policies and is genuinely helpful, but volume on its own does nothing for quality.

Ignoring the basics because they’re boring. A noindex tag left over from a website rebuild can quietly remove a page from both normal results and AI Overviews. We see this more often than you’d think.

Blocking Google by accident. Security plugins, aggressive bot rules and misconfigured robots.txt files can stop Googlebot reaching pages. If Google can’t crawl it, it can’t use it.

Where this leaves South African and US businesses

For local businesses in South Africa, the guide is good news. Many SA sites still have basic problems: slow mobile pages, thin service descriptions, unclaimed or outdated Business Profiles. Fixing those is affordable, and it now counts for AI features as well as standard results.

In the US, competition for AI visibility is fiercer and the temptation to buy shortcuts is stronger. The guide gives you a clear filter for agency pitches: if a tactic depends on something Google says it doesn’t use, it isn’t a strategy for Google.

Either way, the businesses that come out ahead will be the ones whose websites are actually the most useful in their field. That was true before AI Overviews. It’s still true.

Frequently asked questions

Does Google use llms.txt?

No. Google’s AI optimisation guide says Google Search doesn’t use llms.txt or similar AI text files, and they don’t help or hurt rankings.

Do I need schema markup to appear in AI Overviews?

No. Google says structured data isn’t required for its generative AI features. It’s still useful for rich results and for stating business details clearly.

Is GEO different from SEO?

For Google’s AI features, Google says it’s the same thing: SEO. Other AI assistants run their own systems, so it’s still worth checking what they say about your business, but the foundations are shared.

Should I break my content into smaller chunks for AI?

No. Google says its systems understand pages that cover several topics. Write and structure pages for readers instead.

How do I see if my site appears in AI Overviews?

Google recommends the generative AI performance report in Search Console. You can also search your customers’ real questions and see which sources are shown.

Can AI-written content rank in AI Overviews?

Google allows content made with the help of AI tools if it meets Search Essentials and spam policies and is genuinely helpful. Mass-produced, low-value pages won’t help.