How to prepare your website so ChatGPT can find you too

How an AI assistant looks for an answer

When someone asks “who develops custom applications in Brno”, it depends on which service they use and whether the assistant searches the web. Without search, it may answer from what it has learned, which may not reflect what a company currently offers. With search, it can look up further material and link to its sources.

For a website owner, what matters is that search systems can reach their pages and find useful information on them. That is why classic SEO remains important. Google states explicitly that the fundamental SEO best practices apply to its AI Overviews and AI Mode. No special “AI markup” is needed. [1]

ChatGPT, Perplexity and Gemini are not one search engine, though. They do not share the same sources or rules. So the fact that Google knows a page does not mean every AI assistant will use it.

What a customer should find on your website

We recommend starting with the questions people deal with before they send an enquiry. Technical changes will not replace information that is missing from the page.

• What exactly you do and for whom. “We push digital boundaries” can work as a slogan. Next to it, though, you need a concrete description: “We design and develop web applications for companies that need to connect orders, customer accounts and internal systems.”
• Where you operate. Distinguish between where your company is based and the area you serve. If you are based in Brno and work for clients across the whole country, say both.
• Roughly what the service costs. For custom work you do not have to publish a fixed price list. An indicative range, an example brief or an explanation of what affects the price will help. With a “from” price, add what it includes. Give prices to help the customer decide, not as a guaranteed way to get into an AI answer.
• How working together goes. What you need from the client, which steps follow and what they receive at handover. This is also the place for answers to questions about maintenance, timelines and ongoing support.
• What you have already done. Next to the client logo, add the specific brief, your solution and a result you can back up. If you have not measured the business impact, describe what changed in practice. Do not make up numbers.
• Who is behind the company. Give contact details and company information that can be verified. We also recommend checking that they match your company profiles and the other places where you present yourself.

FAQs can make this information easier to follow. But the question and answer format alone does not guarantee that an assistant will use or quote the text.

What to check on the technical side

We recommend making important information available directly in the HTML the server returns. Google can process JavaScript, but that does not hold for all bots. If the main content only loads later or after an interaction, some tools may not get it. [2]

This does not mean removing interactive elements or giving up on design. Service descriptions, references and other essential information should be available even without operating the application.

A sitemap helps with discovering URLs. It does not guarantee, though, that a search engine will crawl or index every page listed in it. [3]

Structured data must match the actual content. For general schema.org checks there is the Schema Markup Validator, while the Rich Results Test checks eligibility for Google rich results. Even a passing test does not confirm that AI will recommend you. [1, 4]

Search and training are not the same thing

Do not set bot access just based on whether a bot has AI in its name. What matters is its purpose.

OpenAI, for example, distinguishes OAI-SearchBot for search in ChatGPT from GPTBot for collecting content that may be used to train models. You can allow the search bot and block the training one. For other providers, go by their own documentation. [5]

Blocking one bot also does not mean a company disappears from all answers. Information about it may be on other websites too.

Treat llms.txt as an optional extra

llms.txt is a proposed format that gives AI tools a brief overview of a website and links to further content. It can serve as a signpost for tools that use it. It is not a universal registration for AI search. [6]

Google states that it does not use llms.txt as a special signal for its search, including generative features, and that adding it does not improve visibility on its own. [7]

If you want to have one, we recommend maintaining it from the same source as the website content. We would give priority, though, to accessible pages, correct information and good references.

Where to start on your own website

We recommend going in this order:

• Go through your main services. Is it clear what you offer, to whom and on what terms?
• Add material that helps people decide. Indicative prices, how working together goes, references and answers to real questions.
• Check technical accessibility. Main content in the HTML, bot access and any obstacles to indexing.
• Align the information about your company. Contact details, your offer and company profiles should not contradict each other.
• Start measuring. Track findability, visits and, above all, relevant enquiries.

This gives you a concrete list of changes. An llms.txt file existing, or a green result in a technical test, does not replace that kind of overview.

What to avoid

• Text hidden to manipulate search. For example, lists of keywords that visitors cannot see. Ordinary expandable answers in an FAQ are a different case: their purpose is to make reading easier, not to hide content from people. [8]
• Piles of pages with no value of their own. If you only swap the city name in the same text, you do not create a useful local offer. A separate page makes sense where you have something specific to say. Among other things, Google penalises content produced at scale mainly to manipulate rankings. [8]
• Guarantees of being recommended in ChatGPT. A supplier can improve content, fix technical obstacles and measure results. They cannot decide which company the assistant chooses.

How to track results

We recommend combining several perspectives:

• Repeat the same customer questions. Test several phrasings and services. Record the date, the tool used, whether it searched the web and which sources it cited. A single answer is not a reliable measurement.
• Check that the information is correct. It is not enough for the assistant to mention your company name. Check whether it describes your offer, area of operation and terms correctly.
• Track visits and enquiries. In analytics, evaluate recognisable visits from AI services and the actions that follow. Keep in mind that the source of every visit may not be identifiable, and a mention without a click will not show up in traffic.
• Use Search Console within its scope. It helps you check performance and indexing in Google. It is not a universal overview of what all AI assistants know about you.

As a practical result, we would count a customer who came with a relevant enquiry and a correct idea of your services.

The foundation is a website from which both a customer and a search system can tell what you do, for whom and why to get in touch with you.

With a new website, it makes sense to deal with content and technical accessibility right from the design stage. With an existing one, we recommend first finding out what is missing and what actually blocks access. Only then decide on the scope of changes.

Want to know what your website needs? Tell us what you are dealing with. You can find how we approach design and development under the Custom websites service.

Sources

• 1. Google: AI features and your website
• 2. Google: JavaScript SEO basics
• 3. Google: What is a sitemap
• 4. Google: structured data testing tools
• 5. OpenAI: Overview of OpenAI Crawlers
• 6. llms.txt proposal
• 7. Google: Optimizing for generative AI features
• 8. Google: Spam policies