Visible in ChatGPT & AI Overviews: GEO for Your Website

A growing share of your potential customers no longer googles – they ask. ChatGPT, Perplexity, or the AI answer at the top of Google search. The answer arrives; the click on a website increasingly doesn’t. For business websites, that means the rules of visibility have shifted – measurably.

This article explains what is happening, how Generative Engine Optimization (GEO) differs from classic SEO – and which parts of it demonstrably work. Not from theory: we run everything we write about here on our own website and see the effects in our own data.

What has changed – in numbers

That AI answers cost clicks is no longer a hunch; it is documented across several independent studies:

• Ahrefs measured around 35 percent fewer clicks in April 2025 for the top-ranking page on queries with an AI Overview – and 58 percent in its December 2025 update.
• Pew Research observed the actual behavior of 900 users: with an AI answer present, they clicked a result on 8 percent of searches, without one on 15 percent.
• Depending on the study, 60 to over 80 percent of searches with an AI Overview now end without any click at all.

The methods differ; the direction doesn’t: every study finds a decline, none finds an increase. We see the same pattern in our own Search Console data: conversational queries – full questions instead of keywords – rank at positions 2 to 8 for us and still bring hardly any clicks. The answer was given, just not on our website.

What GEO is – and what it isn’t

Generative Engine Optimization means setting up your website so AI systems understand it as a source, represent it correctly, and cite it. The goal shifts from the click to the mention – whoever appears in the AI answer as a source or recommendation is visible, even without a visit.

What GEO is not: a bag of tricks that replaces classic SEO. AI systems draw their knowledge from training data and – in search integrations like AI Overviews or Perplexity – from live retrieval of ordinary search results. If you don’t exist there, you don’t exist in the AI answer either. GEO builds on SEO; it doesn’t replace it.

What AI systems anchor content on

From what can be observed and tested, four patterns stand out:

1. Directly quotable answers. AI systems prefer passages that answer a question completely and self-contained – a clear question as a heading, the answer in two or three sentences right below. Texts that circle the point are hard to quote.

2. Structured data. Schema.org markup – FAQ, organization, breadcrumbs, article – turns prose into machine-readable statements. FAQ snippets have lost value for rich results in classic search; as a clean, verifiable signal for AI citability, they remain worthwhile.

3. Consistent entities. AI systems build an internal picture of who you are: name, location, services, team, founding year. If those facts contradict each other across your website, directories, and profiles, the picture blurs – and blurry entities get recommended less.

4. Technical accessibility. What the crawler can’t read, the AI can’t cite: server-rendered HTML, fast load times, clean semantics. Static sites – built with Astro, for instance – have a structural advantage here: the full content sits in the HTML without a script having to assemble it first.

What we do on our own website – honestly assessed

We use our own website as a test bed. The current state, with an honest effectiveness rating:

• FAQ schema on every relevant page – every FAQ on this website is automatically emitted as structured data. Demonstrably useful: AI answers recognizably adopt phrasings from such question-and-answer blocks.
• A facts page as a “grounding page” – a dedicated page with the company’s core facts as a structured entity (who, where, since when, doing what, with whom). It isn’t aimed at human visitors; it gives AI systems one unambiguous, consistent source for the question “who is this, actually?”.
• An llms.txt – the proposed standard for offering AI systems your most important content in machine-readable form. Honest assessment: no major vendor has confirmed consuming it systematically. It is a one-hour experiment, not a proven lever.
• An agent-ready contact form – our form can be operated by AI agents via WebMCP. That is deliberately a look ahead: when assistants start submitting inquiries on behalf of their users, we want to be the website where that works.
• The fundamentals – server-rendered HTML, load time, clean heading structure, breadcrumb and organization schema. Unspectacular, but the base everything else builds on.

The click paradox – and what it means for your goals

The uncomfortable consequence: good rankings and falling clicks no longer contradict each other. If you only measure visitor numbers today, you see a shrinking curve and may draw the wrong conclusion that the website is getting worse – while it appears in AI answers more often than ever.

Three adjustments that follow:

1. Broaden the metrics. Alongside traffic: do I appear in AI answers on my core topics? Are my brand and services represented correctly? That can be spot-checked – ask ChatGPT, Perplexity, and Google the same questions regularly.
2. Count inquiries, not visits. If fewer people visit the website, but the right ones do, that is no loss. The qualified inquiry is the currency, not the page view.
3. Write content to be cited. Guide content that answers questions completely and carries the brand visibly – then even the answer without a click pays into your reputation.

Action list for your business website

1. Collect your customers’ core questions and answer each one directly and completely on a fitting page – question as the heading, answer in the first sentences.
2. Add structured data: organization, FAQ, breadcrumbs – and verify it is emitted without errors.
3. Make your facts consistent: name, address, services, and figures identical everywhere – website, Google Business Profile, directories, LinkedIn.
4. Be technically accessible: content in the HTML rather than behind scripts, fast load times, clean semantics.
5. Establish spot checks: ask the common AI systems the same five to ten questions monthly and note whether and how you appear.
6. Optional: create an llms.txt – as a documented experiment, not an expectation.

An honest conclusion

GEO is not secret knowledge and no substitute for a solid website – it is the consistent continuation of what good websites have always been about: clear answers, consistent facts, clean technology. What’s new is the goal behind it: being cited, not just being found. Think both together, and AI search doesn’t cost you visibility – it moves it to where your customers now ask their questions.

---

Want to know how visible your website is in AI search? We build websites that do both – rank and get cited – and we’ll tell you honestly where yours stands: drop us a line.