Here is a thing that surprises people the first time they see it: the reviews you are proudest of, the ones glowing away in a widget on your homepage, very often do not exist as far as an AI assistant is concerned.
Not hidden. Not ranked low. Simply not there.
What is actually going on?
Your browser does two jobs when it opens a page. First it downloads the HTML, which is the text and structure. Then it runs any JavaScript on the page, which fetches and builds the rest: the review carousel, the booking calendar, the map, sometimes the whole body of the page if the site is built as a single-page app.
The crawlers that feed AI assistants mostly stop after the first job. A joint analysis by Vercel and MERJ of real crawler traffic found that the major AI crawlers, GPTBot, ClaudeBot and PerplexityBot among them, fetch the HTML and do not execute JavaScript. Some of them download the script files and still never run them.
So they read what arrived in the first second, extract it, and move on. Whatever your scripts were about to build never gets built.
Google is the exception worth knowing about: Googlebot does render JavaScript, on a delay. That is why a site can look perfectly healthy in Google Search Console and still be close to blank to an assistant. Two different readers, two different results, and only one of them is being checked by most people.
Which parts of a site does this usually hit?
The parts businesses most want read, as it happens:
- Embedded review widgets. Google reviews pulled in by a third-party script. Your strongest trust signal, and one of the most common things to be invisible.
- Booking, pricing and availability tools loaded inside the page.
- Tabs and accordions where the content only loads when clicked, which is often where the FAQ lives.
- Product and menu listings pulled from a database by script.
- Whole sites built on a framework that renders in the browser, with no server-rendered version.
The pattern is the same in every case. A human sees a full page. A crawler sees a headline, a menu, and a lot of empty containers.
How do you check your own site in two minutes?
You do not need a tool for the first pass.
Open your homepage, right-click, and choose “View page source”. That is the raw HTML, before any scripts run, and it is roughly what a crawler receives. Use your browser’s find function and search for a review quote, a price, an FAQ answer, anything you would want an assistant to know.
If you cannot find the text there, it is not there for them either.
For a second opinion, turn JavaScript off in your browser settings and reload the page. What remains on screen is close to what an AI crawler works with. It is an uncomfortable exercise the first time and a genuinely useful one.
How do you fix it?
Usually with less work than the problem suggests, because the fixes are structural rather than creative.
- Put the important text in the HTML. If reviews matter, publish the text of your best ones as real text on the page, in addition to whatever widget you keep for looks. Same for prices, hours, service areas, and FAQ answers.
- Server-render what you can. If your site runs on a modern framework, most of them have a server-side rendering or static output mode. Turning it on is a build setting, not a rebuild.
- Do not hide answers behind a click. Accordion FAQs are fine when the answer is in the HTML and merely collapsed visually. They are a problem when the answer is fetched on click.
- Check your robots.txt is not refusing the crawlers you want. OpenAI documents three separate user agents, one for training, one for the search index that ChatGPT cites, and one for user-triggered fetches. Blocking them is a legitimate choice, and blocking them by accident while hoping to be cited is not.
- Add structured data. Schema markup for your organisation, services, reviews and FAQs gives the machines an unambiguous version of the facts rather than one they have to infer.
Does this actually change anything?
It changes what you are eligible for.
Being readable does not guarantee you get cited by an assistant. What it does is make you a candidate. A page an assistant cannot read cannot be quoted, cannot be summarised, and cannot answer the question a customer just asked, no matter how good the content on it is. Most of the sites we audit have a real answer to the customer’s question sitting somewhere on the page, in a format nothing can lift.
That is the encouraging part of this. The content is usually already written. The work is making it visible to a reader who arrived early and left quickly.
The wider point
Websites have been built for one reader for twenty-five years: a human with a browser, and Google’s approximation of one. There is a second reader now, and it is less patient and less capable than the first. It does not click, it does not wait, and it does not come back for a second look.
Designing for both is not a competing agenda. It is the same discipline that makes a site fast and clear for people: say the important things in plain text, early, where anything can find them.
If you want to know what your site currently looks like to that second reader, and which of your own pages are effectively blank to it, book a free readiness call. Twenty minutes, no pitch. Worth pairing with reading what AI already says about your business, because the two problems usually turn out to be the same one.