Click Coded · AI-operated, human-reviewed
Can AI Assistants Actually Read Your Website?
No jargon, no rubric table for this one, just what actually happens, step by step, when someone asks ChatGPT or Perplexity a question your business could answer. Two real sites from our own benchmark data, one that gets read and one that doesn't, walk through exactly where it goes right or wrong.
The question a real person asks
Someone types "best small SUV under $30k" into an AI assistant. Somewhere out there, ten car makers have a page that answers this exact question. The assistant doesn't know all ten exist. It has to find them, open them, and pull out the facts. That process either works or it silently fails, and almost no site owner ever finds out which.
What the assistant actually does, in order
Step 1
It fetches the page. Not a screenshot, not what a person sees in a browser. The raw response a server sends back, before any JavaScript runs.
Step 2
It checks whether it's even allowed to. A site's robots.txt file can explicitly block an AI crawler by name. Several major sites in our benchmarks do this, sometimes on purpose, sometimes because a default template blocked everything years ago and nobody noticed.
Step 3
It looks for a plain-text summary first. A file at yoursite.com/llms.txt, if one exists, tells the assistant what the site is and what's on it, without having to parse a page at all. Adoption varies a lot by industry: in our own audit of 20 major SaaS homepages, 75% have one — but only 20% of those same sites publish the more complete llms-full.txt, and adoption drops off sharply outside tech-forward industries.
Step 4
Failing that, it reads the raw HTML. If the actual price, model name, or answer is sitting in that raw HTML, it can pull it out directly. If the page is a mostly-empty shell that only fills in with JavaScript after the page loads, the assistant sees the empty shell. Many crawlers don't run JavaScript at all.
Step 5
It looks for structured data (schema.org markup): a specific, labeled tag saying "this number is the price" or "this text is the business name," instead of leaving the assistant to guess from page layout the way a person would.
One site that passes, one that doesn't
From our own automotive industry benchmark, run the same way, same nine checks, against 14 manufacturers: Hyundai USA scored 95 out of 100. Mazda USA scored 50. Same industry, same kind of question a shopper might ask, a 45-point gap in how much of that process above actually completes. Full ranking: the automotive benchmark.
Why "somewhere else" is the real cost
An assistant that can't read your page doesn't show an error message to the person asking. It just answers using whatever it could read, a competitor, a review site, an old forum post, or it says it doesn't know. Nobody sees the page that failed. That's what makes this invisible: there's no broken-link message, no 404, just a business that quietly never comes up.
Check your own site
Same 5 steps, your site The free checker below runs through the same process described above and shows you exactly where it stops, if it stops.
Start here, no signup, results in about a minute. Want the full technical rubric behind this walkthrough: read the methodology. Curious why we check these facts directly instead of asking an AI what it thinks: GEO Is Not Guesswork.
We add new free tools and benchmarks periodically. Get an email when we do, no spam, unsubscribe anytime.
Built by Click Coded, AI-operated and human-reviewed. Contact: alexander.k.eliot@gmail.com