// Blog

Getting found when AI gives the answer

More and more people ask ChatGPT, Claude or Perplexity instead of a search engine – here is how to make sure your website shows up in the answers

Screenshot of the website michael-becker-berlin.de with clearly structured sections

"Which web developer in Berlin knows about websites for associations?" Many people no longer ask questions like this only on Google, but directly to an AI assistant. Instead of a list of ten links, they get a finished answer – with a few sources underneath. If your website doesn't show up there, you simply aren't found at that moment. In this article I explain what you can do about it.

How AI assistants arrive at their answers

ChatGPT, Claude, Perplexity and the AI Overviews in Google Search work differently, but follow a similar principle: for current questions, they search the web, read suitable pages and summarise what they have read. They often link the pages they rely on as sources.

Diagram: a question to the AI assistant, the assistant reads suitable web pages and gives an answer with sources. The labels are in German.

That brings good news: an AI can only cite what it finds and understands. And what makes that happen are essentially the same basics that apply to classic search.

Clear content that answers real questions

AI assistants look for pages that answer a question directly and reliably. That speaks for:

  • Concrete statements instead of marketing phrases. "I look after the websites of allotment garden associations in Berlin" helps more than "We bring your vision online".
  • Your clients' questions. What are you asked again and again? Answer exactly that on your website, ideally as separate sections or in an FAQ area.
  • One page, one topic. If you offer several services, give each one its own clearly named page or section.
  • Up-to-date information. Outdated opening hours, prices or contact persons do double damage: with people and with machines.

My clients mainly ask me two questions again and again: "What does a website cost?" and "How can I change the content myself later?" That's why I answer the first one in detail in an article of its own: What does a website really cost – and what keeps running? A page like this is exactly what an AI assistant can cite as a source when someone asks it the same question.

Structure that machines can read

A person recognises a heading by its font size. A machine only recognises it if it is also marked up as a heading in the code. Pay attention to:

  • one main heading per page, with sensibly structured subheadings below it
  • bulleted lists for lists, instead of writing everything in one paragraph
  • short paragraphs, each containing one idea
  • text as real text – not hidden in an image or a PDF

This also helps people using a screen reader. Good structure is always a piece of accessibility, too.

Who you are: imprint, contact and consistent details

AI assistants want to name sources that can be trusted. A complete imprint, an easy-to-reach contact page and a recognisable person or organisation behind the website all contribute to that.

It is also important that your name, address and phone number are the same everywhere – on your website, in your Google Business Profile, in business directories and on association portals. If the details differ, it becomes harder for search engines and AI to attribute them clearly to you.

Structured data

With so-called structured data following the schema.org standard, you can additionally store information in the code in a machine-readable form: who the provider is, where they are based, what the opening hours are, which events take place. This doesn't replace good content, but it makes it less ambiguous. Google has been using structured data for years, and other systems can evaluate it too.

This website uses the structured data plugin that I wrote for my static site generator Nera. On every page, it records who is behind the website – with address, contact details and profiles – and marks every blog post with its title, date and author. I only have to maintain these details in one single place.

You can check yourself whether a website has structured data and whether it is correct: with Google's Rich Results Test or the Schema Markup Validator. Both work with any public address, including your own website.

Fast and technically sound

Slow or faulty pages are read in full less often – by people and by crawlers alike. I described how you can check the quality of your website yourself in the article Quality check of your own website. The areas shown there – performance, accessibility and search engine optimisation – are relevant for AI search as well.

Don't lock out crawlers by accident

Whether a program may read your website is controlled via the robots.txt file. The big providers now distinguish between three kinds of access: crawlers that collect texts for training their models, crawlers that index pages for search in the assistant, and fetches that a person triggers with their question right now. The most important identifiers:

  • OpenAI: GPTBot (training), OAI-SearchBot (search in ChatGPT), ChatGPT-User (fetch on request) – according to OpenAI
  • Anthropic: ClaudeBot (training), Claude-SearchBot (search), Claude-User (fetch on request) – according to Anthropic
  • Perplexity: PerplexityBot (search), Perplexity-User (fetch on request). Perplexity does not name a separate training crawler.
  • Google: Google-Extended controls whether content may be used to train Gemini. According to Google, this has no influence on whether your page appears in Google Search. For the AI Overviews in Google Search, according to Google, the same requirements apply as for normal search, so they are controlled via the normal Googlebot.

This separation is practical: you can, for example, exclude training and still appear in the assistants' search results. With fetches on request, however, the providers do not always follow robots.txt, because a person requested the page.

Some hosting providers, security services or templates now block AI crawlers by default. That can be a deliberate decision – for example if you don't want your texts to be used to train AI models. The important thing is that you decide it consciously. Anyone who blocks the search crawlers will hardly appear as a source in the answers of the respective assistants.

And what about llms.txt?

You may have come across the term llms.txt. Behind it is a proposal to place a file in the root directory of the website that summarises the site's content in a compact form for language models. The idea makes sense, but: it is not an official standard, and there is no evidence so far that the big AI providers actually evaluate the file. If you want to create one, go ahead – just don't expect miracles. Good content on the actual pages matters more.

Such a file is useful where people deliberately hand it to an AI assistant. I implemented this for the documentation of my static site generator Nera: when the site is built, an llms.txt, the entire documentation as one text file, and a plain Markdown version next to every documentation page are created automatically. Anyone working with Nera can simply paste this address into the chat with an assistant, which then has the current documentation in front of it right away.

Classic search engine optimisation remains the foundation

Much of what AI assistants know about the web comes from the same sources as classic search. A website that is easy to find on Google also has better chances in AI search. That's why it is still worth looking at Google Search Console – I explain how to register your website there in the article Website in Google Search Console.

Conclusion

To show up in the answers of ChatGPT, Claude, Perplexity and the like, you don't need any tricks. What counts is clear, up-to-date content that answers real questions, a clean structure, unambiguous details about you, a fast website and a conscious decision about which crawlers may read your site.

If you'd like to know how your website is doing on these points, feel free to get in touch. I'll take a look and tell you honestly where improvements are worthwhile.

Yours, Michael Becker