A customer opens ChatGPT and asks for a roofer in Islip. Before it answers, it goes and fetches a few websites. Here is the part nobody tells small business owners: on plenty of perfectly nice sites, what comes back is close to a blank page. Not slow, not ugly. Blank. The hours, the services, the phone number, none of it there.
This is not a design problem and it is not a ranking problem. It is closer to plumbing. The good news is you can check your own site in about two minutes, without buying anything or calling anyone.
What does an AI assistant actually fetch?
When you open a website, two things happen in order. Your browser downloads the page, then it runs the code that fills the page in. On a lot of sites built in the last few years, that second step is doing most of the work. The words you read arrive after the fact, assembled inside your browser by JavaScript.
The bots behind the AI assistants skip the second step entirely. They take the page as it was sent, read it, and move on. Vercel, a web hosting company, published a study with the analytics firm Merj in December 2024 that watched real crawler traffic at scale, including 569 million fetches in a single month by the bot behind ChatGPT, and found that none of the major AI crawlers ran JavaScript at all. They will even download the code files and then never run them. Whatever those files were going to write onto your page never gets written, as far as the machine is concerned.
That study is from the end of 2024, so before writing this I went looking for anything that overturned it. I did not find it. As of August 2026 the behavior is the same: raw page in, answer out.
Why does my site look fine on Google, then?
Google is the exception, and that is exactly why this problem hides so well. Google's crawler has rendered JavaScript for years, so a site that assembles itself in the browser can rank perfectly well in ordinary search results. The reports look healthy. The assistants are a newer set of readers with older habits, and none of them send you a notice when they come back empty handed.
How do I check my own site in two minutes?
Nothing to install, nothing to buy. Do this on a desktop or laptop.
- Open your homepage, right click anywhere on it, and choose View Page Source. A window opens full of code. That is not a special developer view. It is the exact file a crawler is handed.
- Press Command F, or Control F on Windows, and search that window for your phone number. Then your street address. Then a full sentence you can plainly see on your homepage. Then the name of one of your services.
- Read the result the simple way. Whatever you find in there, an AI assistant can read. Whatever you can see on the live page but cannot find in the source is being painted in afterward, and it is not reaching them.
- Run it again on the page that matters most, usually services, pricing, or contact. Homepages are often built by hand while inner pages come off a template, so the two can behave differently.
- Notice the size. If the entire source is a few dozen lines and mostly script tags, you already have your answer.
Two things fail this test regularly, even on otherwise solid sites: booking widgets and review badges dropped in from another company, and any section that only loads once you scroll to it. I walked through a wider version of this check, including the Google side of it, in the note on whether ChatGPT can recommend your business.
Does this matter enough to fix?
That depends on how much of your future you think is answers rather than links. Here is the one number I would put in front of you. SparkToro, analyzing Similarweb clickstream data in 2026, found that 68.01 percent of United States Google searches in the first four months of 2026 ended without a click, up from 60.45 percent two years earlier. Let me be straight about what that measures. It counts Google searches, not AI chats, and it says nothing about your business specifically. What it shows is a direction. People increasingly get their answer without landing on anybody's website, which makes what a machine can read about you more important, not less.
Do llms.txt and schema markup fix this?
There is a file called llms.txt going around as the fix for this, occasionally with a price attached. It is a proposal, not an adopted standard. Google's published guidance on its AI features states that you do not need to create machine readable files, AI text files, or special markup to appear in them, and that there is no particular schema code to add. We still put schema and an llms.txt on the sites we build, because they are free, they are harmless, and labeled facts help a machine parse a page instead of guessing at it. They are not a ranking lever and I will not sell them as one. That same Google page gives the real instruction in a single line: make sure the important content is there in text.
What should I ask whoever builds my site?
You do not need to understand any of the machinery. You need one question answered by whoever builds or maintains your site: is the text on my pages written into the HTML, or is it assembled in the browser after it arrives? A good answer is specific and quick. A vague one tells you something too.
We build the other way, and have since the first site. Ours are static, which means every page is written out as finished HTML before anyone asks for it, so the first thing any reader gets, person or machine, is the whole page. Nothing waiting on a script, no login screen, no assembly. That is not clever engineering. It is an old-fashioned way to build that happens to suit the moment we are in.
It comes with a real trade, and I would rather name it than let you find it later. There is no dashboard for you to log into and change a price at midnight. Changes come through us, usually the same day you ask, and that work is the entire job the monthly plan pays for. That is how our website plan is built, and it is the same reason those sites ship with their facts written for machines as well as people, from day one.
Here is where the evidence stops, so we are square. Being readable does not put you in an AI answer. It is the price of admission and nothing more, and anyone promising you a spot in what ChatGPT says is guessing out loud. These companies could add JavaScript rendering next year, and if they do, part of this note expires and I will come back and say so here. A readable page with nothing worth reading on it stays exactly as invisible as it was. What I can tell you without hedging is that a page a machine cannot read is a page it cannot quote.
Quick answers
Can AI assistants read my website?
Some can read all of it and some can read almost none of it, and the difference is how your site was built. The bots behind ChatGPT, Claude, and Perplexity download your page and read the raw HTML. They do not run the JavaScript that many sites use to fill the page in, so any text added that way is text they never see.
How do I check whether AI can read my website?
Open your homepage on a desktop, right click, and choose View Page Source. That window shows the raw HTML, which is exactly what these bots receive. Search it for your phone number, your address, and a sentence you can see on the page. Anything you cannot find in there is probably invisible to an AI assistant.
Do I need llms.txt or schema markup to show up in AI answers?
No. Google's own guidance for its AI features says plainly that you do not need to create machine readable files, AI text files, or special markup to appear in them, and that there is no particular schema code to add. We ship those files anyway because they cost nothing and help machines parse a page cleanly. Nobody should sell them to you as a ranking lever.
Does a JavaScript-heavy website hurt my Google ranking?
Not by itself. Google's crawler has rendered JavaScript for years, so a site that leans on it can rank perfectly well in ordinary search. That is exactly why this problem hides. Your site can look healthy in every SEO report you have ever been shown and still hand an AI assistant a page with nothing on it.