Can an AI assistant read your website? A ten-minute check

When someone asks ChatGPT or Perplexity for a tiler in Klaipėda, the assistant reads your website instead of looking at it. A crawler fetches the raw HTML of the page, usually without running any scripts, and works with whatever text is in there. If the words are not in the file, you are not in the answer.

You can check this yourself in about ten minutes. Open your site on a laptop, right-click the page and choose View page source, or press Ctrl+U (Cmd+Option+U on a Mac). That plain text is what a machine sees. Then run the nine checks below. If you would rather not open the source, the website check on this site runs the same checks on any address in about a minute.

Words on the page

Real text in the HTML. Pick a sentence from your home page and search for it in the source. If you find it, the page is readable. If the source is a handful of script tags and an empty container, the page is a JavaScript shell: the text appears only after the scripts run, and most AI crawlers do not run them. What to do: ask whoever built the site to render the pages on the server or export them as static HTML. This is the one check a plugin cannot patch. If it fails, the site needs rebuilding before anything else is worth doing.

One main heading. Search the source for h1. You want exactly one on each page, and it should say what the page is and where: roof repairs in Kaunas, not Welcome. Zero headings and the crawler has to guess what the page is about. Three and it cannot tell which one counts. What to do: one h1 per page, everything else as h2.

A title that says what and where. The title tag is the text in the browser tab. It is the first line in a search result and the first line in an assistant’s notes about you. Home, or the company name on its own, says nothing. What to do: service, place, name, in that order, short enough to fit a search result.

A description. The meta description is two sentences the page carries about itself: who you serve, what you do, how to book. When it is missing, the assistant writes its own summary from whatever it finds first. What to do: write those two sentences for every page you care about, starting with the home page.

Facts a machine can quote

Structured data. Search the source for ld+json. You are looking for a block labelled LocalBusiness or Organization with your name, address, phone, opening hours and services. This is the part a machine can quote without guessing, because the fields are named. What to do: most site builders have a setting for it, and a free schema generator makes the block in a few minutes for the rest. Paste it in the head of the page.

An llms.txt file. Open yoursite.lt/llms.txt. It is a plain text file for AI crawlers: who you are, what you do, where, and the pages that matter. Not every crawler reads it yet, but it costs a quarter of an hour and says what you want said. What to do: write it in a text editor and upload it next to your home page.

A sitemap. Open yoursite.lt/sitemap.xml. It lists every page you want found. Without it, crawlers find the pages linked from the home page and stop. What to do: every common CMS produces one with a setting or a plugin. Then submit it in Google Search Console, in your own account.

Doors that are open or shut

robots.txt. Open yoursite.lt/robots.txt. Look for a line starting with User-agent followed by GPTBot, ClaudeBot, PerplexityBot or Google-Extended, with Disallow: / underneath. Look also for User-agent: * with Disallow: / under it, which shuts out everyone. Some hosting packages and security plugins add these lines by default without telling you. A blocked crawler cannot read a single page, however good the pages are. What to do: delete those lines, or change the Disallow to Allow, and check again after the next plugin update.

Loads on a phone. Open the site on your own phone. If the text is tiny and you pinch to zoom, the viewport line is missing and the phone is drawing the desktop page. Your buyers read on a phone, and so does Google when it grades the page. What to do: one line in the head of the page, the viewport meta tag. Every current template has it. Old ones do not.

What to fix first

Text first, because nothing else matters if the page is empty to a crawler. Then the title, the heading and the description, which is an hour of writing. Then structured data and llms.txt. Then robots.txt and the sitemap, five minutes each once you know where to look.

The website check on this site runs these checks on your address, adds canonical address and language checks, pulls speed scores from Google and shows the fix next to every failure. When the page passes, the next question is what an assistant says about you today, which the AI visibility check answers. Book a free audit if you would rather we fix it and show you the before and after.