How to get your business found in AI search (GEO)

More of your customers now ask an AI engine before they ask Google. They type a real question into ChatGPT, or they read Google's AI Overview at the top of the results, or they ask Perplexity, and they take the answer it gives them. Generative Engine Optimisation, GEO for short, is the work of making sure your business is in that answer. This guide is the plain version of how to do it, written for owners and marketers, not developers.

Everything below is the same method I run for clients at Zivaro, and this page is built using it, so it is also a working example. If you want the deeper background and the reasoning, the two posts linked further down go into more detail.

Why AI search changes the game

AI search changes the game because the customer often gets their answer without ever clicking a website. Classic Google gave you a list of links and the user picked one, so the prize was a top ranking. AI search reads several sources, writes one answer, and names a few of them. The new prize is being one of the sources it names.

That shift has three practical consequences for a South African business:

• Clicks can fall even when nothing is wrong, because the engine answered the question on your behalf.
• Being mentioned by name becomes valuable on its own. If ChatGPT tells someone "a studio like Zivaro can do this," that is a warm lead, with or without a click.
• The competitor who structures their page to be quotable wins, even if your site has more pages and an older domain.

The good news is that most of your competitors have not started. The work is not hard. It is just specific, and almost nobody is doing it yet, which is exactly why it is worth doing now.

Can the AI engines even read your site?

Sometimes they cannot, and the cause is usually one of two things: a robots.txt file that excludes the AI bots, or a security or CDN setting that challenges them before they reach your content. Check this first, because none of the rest matters if the crawlers are locked out at the door.

The four crawlers to let in by name are GPTBot (OpenAI, behind ChatGPT), ClaudeBot (Anthropic), Google-Extended (Google's AI training and AI Overviews signal), and PerplexityBot (Perplexity). Open yoursite.co.za/robots.txt in a browser and make sure none of those four are disallowed. If you use Cloudflare, watch for its "Block AI bots" and managed robots.txt toggles, because those override your own file and shut the crawlers out without you noticing.

One more reading test: your actual words need to be in the page itself, not painted on afterwards by JavaScript. A quick check is to view the page source (right click, then "View Page Source") and search for a sentence you can see on screen. If the sentence is there in the raw HTML, the engines can read it. If the source is mostly empty and the text only appears once the page finishes loading, the crawlers may see nothing. That is a common and fixable problem.

Publish an llms.txt

An llms.txt is a short plain-text file at the root of your domain (yoursite.co.za/llms.txt) that lists your most important pages with a one-line description of each. Think of it as a friendly map you hand to AI assistants so they can find your best content fast.

Be honest about what it is: adoption across the AI engines is still uneven, so I treat it as a cheap hedge rather than a guaranteed win. It takes about half an hour to write, it cannot hurt you, and the possible upside makes it easy to justify. I add one to every project. What it should contain:

• Your homepage and what your business actually does, in one sentence.
• Your key service or product pages, each with a short, plain description.
• Your best guides and answer pages, the ones you would want quoted.

Keep it accurate and keep it short. A bloated llms.txt stuffed with every page is less useful than a tight one pointing at the handful that matter. Do not skip the fundamentals to get this done, but once they are in place, it is half an hour well spent.

Structured data: speak the engines' language

Structured data is a small block of code that tells the engines plainly what a page is: an article, a list of questions and answers, a product, a business. It does not change what your visitors see. It just removes the guesswork for the machine reading the page, and a machine that is not guessing is a machine more likely to quote you correctly.

The trick is to use only the structured data that matches what is genuinely on the page. The two that earn their place most often are:

• Article, for a guide or blog post, naming the author and the publisher so the engines know who stands behind it.
• FAQPage, for a real set of questions and answers, like the visible block at the bottom of this page.

Resist the urge to add markup that overclaims. The engines and Google can compare your code to your visible content, and a mismatch (ten invented FAQ questions nobody asked, a Product tag on a page with no product) reads as manipulation and works against you. Three or four honest questions beat fifteen padded ones. Accurate structured data is a positive signal. Padded structured data is a negative one.

Write answer-first

Writing answer-first means putting the full answer to a question in the first line, then explaining underneath. AI engines rarely quote a whole page. They lift one passage that answers one question cleanly, so your job is to make that passage easy to find and easy to lift.

The pattern is simple and you can see it on every heading above this one:

• Make each heading a real question a customer would actually ask.
• Answer it completely in the first sentence or two, with no need to read anything else first.
• Then add the supporting detail underneath for the reader who wants depth.

A good test: copy any one section out, paste it into a blank document, and read it cold. If it still answers its heading on its own, it will travel well and an engine can quote it whole. If it leans on something you said three sections earlier, rewrite it to stand alone. This means each section repeats a little context, and that mild repetition is the point, not a flaw. Classic SEO writing delays the answer to keep people scrolling. GEO does the opposite: give the answer first, every time.

How to measure it

You measure GEO by checking whether you are the answer, not only whether you got the click. Once a month, ask ChatGPT, Google, and Perplexity the handful of questions your customers actually ask, and write down whether your business is named in the reply. That simple log tells you more than any single dashboard number.

Alongside that monthly check, watch the right signals in your analytics:

• Brand mentions in AI answers, tracked by hand or with a visibility tool. Rising mentions are a real win even if clicks stay flat.
• Referral visits from AI tools, kept separate from your normal Google clicks so you can see them growing.
• The gap between a steady click count and rising brand mentions, which a click-only view wrongly reads as failure when it is actually success.

If a page still is not cited after a quarter, the cause is usually one of three things: the answer is too vague to lift, your brand needs more authority signals before the engines trust it, or a stronger source already owns that answer and you need a sharper angle. Fix them in that order, cheapest first.

If you would rather work from a checklist than from memory, the free GEO Starter Pack pulls all of this together: a step-by-step checklist, ready-to-use templates including an llms.txt starter, and a schema generator so you can produce correct Article and FAQ markup without writing code. Grab it just below.

Want to go deeper? Two companion reads: the gentle introduction in how to show up in AI Overviews and ChatGPT search, and the full operational version in the GEO playbook I use to get a page cited by AI.