How to get your business to show up in AI search: what GEO really takes

You do not need a secret new technique to be recommended by ChatGPT, Perplexity or Google's AI Overviews. Google says so itself in its July 2026 guidance: optimising for AI search is still SEO. Three things decide it. The crawlers these systems use must be able to read your site, and it must be indexed. Your content must answer real questions clearly, with knowledge nobody else has. And the facts about your business must match everywhere they appear. An llms.txt file, special markup or text rewritten for AI are not on that list.

What is GEO, and how is it different from SEO?

GEO stands for generative engine optimisation: preparing your content so that generative AI systems such as ChatGPT, Gemini, Perplexity and Google's AI Overviews use it as a source. You will also see AEO (answer engine optimisation), LLMO and "AI SEO". In practice they describe the same goal: appearing in the answer, not just in a list of links.

The difference from classic search lies in how people behave. In a results list, they click a result. In an AI answer, they often read the answer itself; they may click one of the cited sources, or they may not. So the goal has two parts: your name appearing in the answer, and your page being linked as the source.

	Classic SEO	GEO
Goal	A high position in the results list	Being a source or a named brand in the AI answer
What people see	A list of links	A finished answer with a few sources
Measure of success	Rankings, clicks	Impressions, citations, mentions, enquiries
Foundation	Indexing, content, technical set-up	The same foundation

The difference is often overstated. Google's guidance puts it plainly: from Google Search's perspective, optimising for generative AI search is optimising for the search experience, and so it is still SEO. There is no extra requirement for appearing as a supporting link in Google's AI Overviews either: the page must be indexed and eligible to be shown with a snippet. Without the foundations, GEO gains you nothing. With them, GEO means shifting the emphasis: original knowledge, clear answers, consistent business information and new ways of measuring.

Where do ChatGPT, Google and Perplexity get their information?

Every AI assistant reads the web in its own way. If your site is closed to that route, you will not appear in the answer, however good your content is. So the first question is not "What should we write?" but "Can these systems see us at all?"

System	Where the information comes from	What you need
Google AI Overviews and AI Mode	Google's search index	Indexed page, snippets allowed
ChatGPT search	OpenAI's OAI-SearchBot crawler and third-party search providers, including Bing	Allow OAI-SearchBot, be indexed by Bing
Perplexity	PerplexityBot crawler	Allow PerplexityBot
Microsoft Copilot	Bing	Be indexed by Bing, set up Bing Webmaster Tools

AI assistants answer in two ways. The first is what the model learned during training: older knowledge that you cannot influence directly. The second is a web search carried out when the question is asked, and this is where your site can appear. OpenAI keeps the two apart with separate crawlers. GPTBot collects content for training; OAI-SearchBot feeds ChatGPT search. According to OpenAI, sites that opt out of OAI-SearchBot are not shown in ChatGPT's search answers.

Can AI crawlers actually read your website?

Not necessarily, and many businesses do not know. Since 1 July 2025, Cloudflare has blocked AI crawlers by default on newly added domains; anyone setting up a domain is asked whether AI crawlers should be allowed. Cloudflare customers had already been able to block all AI bots with a single click since 2024, and more than a million had done so by July 2025. A site behind Cloudflare, or behind any CDN with bot protection, may therefore have disappeared from ChatGPT search without anyone deciding that it should.

That is why one distinction matters before you block or allow anything:

• Allow search crawlers. Googlebot, Bingbot, OAI-SearchBot and PerplexityBot supply the sources for answers. They should not be blocked in robots.txt, nor by your firewall, CDN or your host's bot protection. Perplexity also recommends allowing the published IP ranges its crawler uses.
• Decide deliberately about training crawlers. GPTBot and similar crawlers collect content for training models. Blocking them is a legitimate business decision and, according to OpenAI, does not cost you visibility in ChatGPT search.
• Check your server logs. Only the access logs show whether a crawler actually gets through. An open robots.txt achieves nothing if the firewall in front of it turns the crawler away.

The usual technical care applies too. Google Search Console and Bing Webmaster Tools should be set up, with your important pages in both indexes. Settings such as nosnippet or max-snippet also limit what Google's AI Overviews may use from your page, so they belong only where they were set on purpose. Your main content should be in the HTML the page first loads, because not every crawler runs JavaScript as reliably as Google. And structured data is not required for AI search, according to Google, but it is worth having as long as it says exactly what the page itself says.

What kind of content do AI systems cite?

In short: content that answers a specific question straight away and contains something that cannot be found elsewhere. Google's guidance explicitly advises against repeating what others have already said or what an AI model could easily produce on its own. AI can answer general questions without you. It needs your page where you know more.

The most widely cited research on the subject is the GEO study published at the KDD conference in 2024. The researchers tested changes to content in a controlled environment. Three changes worked best: citing sources, adding statistics, and including quotations from credible sources. They raised visibility in AI answers by up to 40 per cent. Keyword stuffing, by contrast, did almost nothing. The most interesting finding for smaller businesses: a page ranked fifth in search results gained 115 per cent in visibility by adding citations, while the page ranked first lost 30 per cent. These figures come from a lab setting and will not transfer one to one. But the direction is clear: specific, well-sourced content wins.

For your business, that means:

• Make your customer's question the heading. Not "Our services" but "How long does a loft conversion take?". Put the answer in the first two sentences beneath it.
• Write down your own experience. How long a project takes, what drives the price, which mistake you see most often, which jobs you turn down. You have this knowledge; nobody else does.
• Use numbers, not phrases. Not "quickly" but "within three to five working days". Specific statements get quoted; vague ones get skipped.
• Cite your sources. Legislation, standards, official guidance. A sourced sentence builds trust with readers and machines alike.
• Show who wrote it and when. The author's name, role and the date of the last update should be visible on the page.
• Put comparisons in tables. Tables and numbered lists are easy for people and machines to read.
• Read your service pages the same way. "Quality and reliability" tells an AI nothing. What you do, for whom, how and how long it takes should be spelt out.

What matters beyond your own website?

AI does not get to know a business from its own website alone. If the information about you online is scattered or contradictory, the assistant either leaves you out or mentions you with the wrong details. The work beyond your website is making sure your business is described the same way everywhere.

• Keep your business details identical. Company name, address, email, services and opening hours should read exactly the same on your site, your Google Business Profile, LinkedIn and industry directories. An old address in one place is one of the most common reasons for AI getting the facts wrong.
• Earn genuine mentions, not bought ones. Customer reviews, an interview in a trade publication, a conference talk, a partner's case study are all valuable. Google's guidance specifically notes that artificially obtained mentions are not as helpful as they may seem. Promotional posts left in forums and comment sections fall into that category.
• Keep product data consistent. If the price on your site differs from the price you send to Google Merchant Center, products get rejected, and the wrong figure can find its way into AI answers too.

Which GEO services are a waste of money?

Google's July 2026 guidance has a whole section on what you do not need to do. A good share of what is sold as a GEO package appears right there. The same guidance recommends prioritising legitimate SEO work over so-called "AEO/GEO hacks".

Service on offer	What Google's guidance says
Setting up llms.txt	You do not need new machine-readable files, AI text files or special markup to appear in Google Search.
Splitting content into small chunks	There is no need to break content into tiny pieces for AI to understand it.
Rewriting text "for AI"	Write naturally; the systems understand synonyms and meaning.
Buying mentions and links	Inauthentic mentions are not as helpful as they may seem.
Focusing everything on structured data	Not required for generative AI search; useful for rich results.

Two more warning signs belong on this list. The first is a ranking guarantee. AI answers have no fixed order; the answer depends on the person, the question and the moment. Anyone promising "number one in ChatGPT" is selling something they cannot measure. The second is figures without a source. Claims such as "a large majority of AI citations come from earned media" or "sites with llms.txt get more AI traffic" appear in many guides, almost never with a verifiable measurement behind them. A figure without a source is no basis for a proposal.

How do you measure AI search visibility?

Until a year ago there was hardly any official way to do it. Now there is, and it is free. Google and Microsoft both show in their own dashboards how often your site appears in AI answers. Paid tools can add to this data but do not replace it.

• Google Search Console. Visits from Google's AI Overviews and AI Mode are already counted in the Performance report under the "Web" search type. Since 2026, Search Console has also shown visibility in AI Overviews and AI Mode in a report of its own; it has been available to every site since August 2026.
• Bing Webmaster Tools, AI Performance. Launched in February 2026, the report shows how often your pages are cited as a source in Microsoft Copilot and Bing's AI summaries, which pages are cited, and which phrases the AI used when it searched for content.
• Google Analytics. Visits from chatgpt.com or perplexity.ai show up as separate referring sites. The numbers are often small, but these visitors are usually close to a decision.
• A manual question test. Write down ten to twenty questions your customers would ask. Once a month, put them to ChatGPT, Gemini and Perplexity and note whether you are mentioned and which source appears. Because answers vary, look at the trend over months, not at a single result.
• The real measure: enquiries. Ask in your contact form or first call how people heard about you. "ChatGPT recommended you" tells you more than any report.

What should you ask a GEO agency?

Ask these questions in writing before you accept a GEO proposal. The answers show whether you are buying real work or a new label.

1. Which AI crawlers are allowed in robots.txt, the firewall and the CDN, and how did you check?
2. Is our site indexed by Bing, and are Bing Webmaster Tools set up?
3. Will you show us the Search Console AI report and Bing's AI Performance report regularly?
4. Which questions will we track, and how did you choose them?
5. Where will our experience and our numbers come from in the content you write?
6. If you recommend llms.txt or special markup, which official source shows that a platform uses it?
7. Do you guarantee a result? If so, on what basis?
8. In whose name will the accounts, reports and content sit when we stop working together?

The last question looks minor but settles a lot. If Search Console, Bing Webmaster Tools and Analytics are in your name, your data stays with you, whichever agency you work with.

What have we done on our own website?

On makkdigital.com we apply the same principles to ourselves. As of 24 September 2026:

• Authors are people. Every blog article is published under the name of the person who wrote it; in the structured data, too, the author is a person, not the company.
• Answer first. Every article opens with a short introduction that answers its question directly. Automated tests check its length before publication; an article that breaks the rule cannot go live.
• FAQs from one source. The frequently asked questions are visible on the page, and the structured data is generated from the same source. What visitors read and what machines read cannot drift apart.
• Markup matches the screen. In the structured data on our service pages, each service description is word for word what the page shows. If they differ, an automated test stops the release.
• Open to crawlers. Our robots.txt does not block AI crawlers.
• Every language researched on its own. German, Turkish and English texts are not translations; each follows the terms people actually search for in that language. Why that matters is explained in our guide to multilingual websites.

Some things we have deliberately not done. There is no llms.txt. We have not split our texts into chunks for AI, nor written a separate AI version. Google's guidance says none of this is needed, and for us it would have no measurable benefit.

To be straightforward: the site is only now going live, and we have no measured results yet. We will add the Search Console and Bing data from the first three months to this article, along with what worked and what did not.

If you would like to know where your business stands in AI answers today, this assessment is exactly where our SEO and performance marketing service begins: which questions you are named in, which ones your competitors are named in, and what is missing in between.