What you’ll be able to do

By the end of this you’ll be able to get ChatGPT, Claude, Gemini or Copilot to actually open your website, tell you what it found, and quote the exact sentences it used to do it. You’ll also know what these tools skip over, why they skip it, and how to fix the gaps so the important stuff (prices, opening hours, what you actually do) doesn’t disappear.

That’s it. No magic, just checking what’s really being read.

Before you start

Here’s the bit most people get wrong straight away: an AI assistant doesn’t “know” your website exists just because it’s live on the internet. When you paste a URL into ChatGPT or Claude, it fetches the page there and then, reads the HTML it’s given, and works from that. Sometimes you need to nudge it, tell it to “open the link” or switch on web search first. Otherwise it might just guess from the URL alone, which is worse than useless.

The reading itself is the tricky part. If your page relies on JavaScript to load in the text (common on a lot of modern sites), the assistant might only see a blank shell. If your content sits behind a login, it can’t get in at all. And if your prices or hours are baked into an image or a PDF rather than typed out as text, most assistants simply won’t see them.

So before you start testing anything, it helps to know roughly what your site is built from. Is it plain HTML, or does it lean heavily on JavaScript to render the page? Do your key facts live in normal text, or are they screenshots and PDFs? You don’t need to be technical to answer this, just have a look and take a guess.

One more thing worth knowing before you start: don’t paste anything private into these tools. We’ll come back to that properly later, but keep it in mind now.

Get set up

There are three things that make the biggest difference to whether an assistant can actually read your site.

First, clear the blocks. Some sites use a robots.txt file or a firewall that stops AI fetchers getting in. This matters because the fetchers used when someone asks an assistant to “read this page” (things like ChatGPT-User, Claude-User and Perplexity-User) are different from the crawlers used to train these models in the first place. Blocking one doesn’t block the other, and a lot of site owners don’t realise they’ve blocked anything at all. If your site sits behind Cloudflare, check the bot settings in the dashboard specifically. Cloudflare blocks many AI crawlers by default, which catches people out constantly.

Second, put the facts in plain text. If your prices, hours, or description of what you do only exist inside an image or a PDF, move them into normal HTML text somewhere on the page too. It doesn’t need to look pretty. It just needs to be there, in words an assistant can actually read.

Third, give assistants a map. A sitemap.xml file (usually sitting at yoursite.com/sitemap.xml) lists your pages in one place, which helps enormously when you want an assistant to read more than one page at a time. You can also add an llms.txt file, a plain-text summary of your site at yoursite.com/llms.txt. It’s worth saying clearly: this is a convention some assistants use, not a standard they all read. Think of it as a helpful extra, not a guaranteed fix.

If your own site is genuinely hard to make readable, perhaps it’s old, perhaps you don’t have access to fix it properly, hilyt.it is worth a look. It hosts a simple AI-readable profile page for a person or business, plain text plus structured data, that assistants can actually fetch and cite. It’s free to browse existing profiles, and creating your own costs a small one-off fee. It won’t fix your main site, but it gives assistants something clean to point to when your own pages are a mess.

Try it yourself

Right, let’s actually do it. Open ChatGPT, Claude, Gemini or Copilot, and paste in your website URL along with a proper question. Don’t just say “what’s this site”. Ask something specific.

Read [URL] and tell me what this business does, its prices and how to contact it. Quote the sentences you used. List anything you could not load.

Swap in your actual URL, obviously. Then watch what comes back. Did it quote real sentences from your page, or did it seem to be filling gaps with guesses? That second part of the prompt, asking what it couldn’t load, is doing a lot of work. It’s the difference between an assistant confidently making things up and an assistant telling you honestly where it got stuck.

If you’ve got a sitemap set up, try this one too:

Here is my sitemap. Read these five pages and summarise each in two lines.

Paste the sitemap URL and name the five pages you actually care about. This is a good way to check whether an assistant can handle more than a single page at once, which matters if your important information is spread across a few different places on your site.

Try this across two or three different assistants, not just one. They don’t all fetch pages the same way, and one might succeed where another fails.

Check the result

Now compare what came back against what your site actually says. This is where it gets interesting, and slightly uncomfortable if your site’s been neglected.

Run the same question through three assistants and line up the answers side by side. Are the prices right? Is the description of what you do accurate, or has it clearly missed something? Ask each one to quote the exact sentence it used, then go and find that sentence on your actual page. If it can’t quote a real sentence, that’s a strong sign it never actually read the page properly.

Try this one directly:

Compare what you can read on my site with this description of my business: [paste a short, accurate description]. What is missing or wrong?

This forces the assistant to be specific about the gap between what’s on your page and what should be there. It’s often more useful than a general “what do you see” question, because it gives the assistant something to check itself against.

Finally, keep a note of which pages none of the assistants could open. That’s your priority list. If three different tools all failed on the same page, the problem’s very likely on your end, not theirs.

If you want a much more thorough version of this check, run across more assistants and more questions automatically, that’s exactly what visibilitytoai.com does. It runs the same set of questions through ChatGPT, Claude, Gemini and Perplexity, and reports back whether each one mentions and cites you, with actual quotes. Useful if you’re doing this properly rather than just spot-checking a few pages yourself. It’s paid and self-serve, so it suits someone who wants the answer once rather than repeating this manual check every month.

If it doesn’t work

Sometimes none of this works, and the assistant just can’t read your page at all. A few things usually explain it: the page only renders properly in a browser (so the assistant sees an empty shell), the site is blocking fetchers without you realising, or the facts you need are locked inside an image or PDF rather than text.

When that happens, there are a few workarounds. You can paste the page text directly into the chat. You can save the page as a PDF and upload that instead. You can add the URL as a source in NotebookLM if you’re using that. And for anything with multiple pages, give the assistant your sitemap and name exactly which pages you want it to read.

Now, the important bit. Do not paste customer lists into any of these tools. Do not paste unpublished pages or admin URLs, even to “test” something quickly. An assistant reading a public page you’ve deliberately shared is fine, that’s the whole point of this guide, but anything you type or paste into a chat may be kept under that assistant’s data policy. Treat it the same way you’d treat sending something in an email you can’t take back.

If you’re unsure whether something’s public or private, don’t paste it. Simple as that.

Keep exploring

If you’re working through the same territory, a few related guides go further than this one. “How to get AI to cite me” looks at getting mentioned with an actual quote attached, rather than just being read. “How to get AI to recommend my business” is about the step after being read, actually being suggested. And if your site is more code than content, “How to get AI to understand my code” covers different ground entirely.

For a general, free primer on all of this, whatisgeo.app is also worth a read.

Sources and review notes

https://developers.openai.com/api/docs/bots https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler https://docs.perplexity.ai/docs/resources/perplexity-crawlers https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots/ https://llmstxt.org/ https://support.google.com/gemininotebook/answer/16215270 https://www.robotstxt.org/

Review date: 2026-09-16

Disclosure: hilyt, visibilitytoai and whatisgeo.app are our products — this site is made by the same studio, Adapt Progress Evolve. We have tried to compare fairly, and the alternatives named are real and worth a look.