The Invisible Web: What LLMs Miss (and Expose) on Your Site

Massive thanks to Sam Torres and Tory Gray of Gray Dot Co, without whom I could not have written this article.

We all know that there’s a new frontier in SEO right now: AI search.

ChatGPT, Gemini, Claude, Perplexity—they’re all pulling in live web data, in live fetches. It’s also happening as AI bots crawl the web to scrape information for training materials. But here’s the catch: what they see on your page is not always what you think.

Sam Torres and Tory Gray joined us this month to deliver a training session on JavaScript SEO in AI Search, and as Tory put it: “Sadly, for the most part … LLMs do not see rendered HTML.”

So, if your critical content only appears after JavaScript renders, chances are LLMs are blind to it.

This article breaks down what LLMs actually see, why it matters for your brand, and how to audit your site so you’re not caught out.

Response HTML vs rendered HTML: the blind spot

Let’s start with the basics. Googlebot (and therefore Gemini) renders pages. It can execute JavaScript, build the DOM, and see roughly what a user sees in the browser. (Though it doesn’t do this perfectly, which is why we created our JavaScript SEO training course.)

AI bots from the other LLM companies? Not so much. They usually stop at the first response from the server, which is the raw HTML that comes back before any JavaScript has fired.

Sam Torres summed it up:

“They see the response HTML … If you do a view page source in Chrome … All of that is what an LLM typically sees.”

Sam Torres   Gray Dot Company

So if your key content is baked into the response HTML, you’re fine. But if you’re relying on JavaScript to inject copy or links, LLMs may never know they’re there.

Think of it like sending someone a birthday card. Google (usually) waits for you to finish writing the entire card, sign your name at the bottom, and add the address to the envelope. LLMs grab the half-finished version off your desk and run with it.

They’re impatient little fuckers like that.

When critical content goes missing (or the wrong stuff shows up)

This is where things get messy.

Imagine your homepage headline (H1) only appears after rendering. (According to Sitebulb crawl data, this is a super-common issue by the way.) To an LLM, your page looks like it doesn’t even have a headline.

Tory flagged a few of the biggest risks to watch for:

• If the H1 is only in the rendered HTML

• if the H1 is modified by JavaScript

• if the page contains JavaScript content.

Think product descriptions, pricing tables, or any content users rely on to make decisions, if that copy is injected with JavaScript, it’s invisible in AI search responses.

And then there’s the flip side: oversharing.

LLMs don’t just see your polished copy—they can also scrape developer comments. Oh yeah, they’re watching us alright!

As Sam warned:

“LLMs actually see the content of comments in your code … really think about those are things that can actually pop up … there might be rants, Easter eggs, confidential information.”

Sam Torres   Gray Dot Company

So if your dev team once left an in-joke (“// Richard is an idiot, fix this later”), don’t be mad if it ends up surfaced in an AI-generated summary. Maybe nobody cares about Richard but it’s something to keep in mind.

How to audit for LLM visibility

Knowing what LLMs think your brand does has become super-important in digital marketing. And for those of us working in SEO, this starts with understanding what’s in your response HTML versus what only appears in the rendered HTML.

Luckily for you, Sitebulb makes this easy. Its JavaScript SEO checks flag common issues such as:

• Contains JavaScript links: LLMs won’t follow them.

• H1 only in rendered HTML: Your primary page heading is invisible.

• Critical page copy only in rendered HTML: Product details, CTAs, or landing page copy might not exist in LLM training data.

I like how Tory described it:

“Not all content is created equal … maybe that page is your homepage or maybe that page is a critical landing page that drives a lot of business and revenue for you or it might be ancillary unhelpful content. So, context matters.”

Tory Gray   Gray Dot Co

The key is prioritisation. If it’s disclaimers or footnotes hidden by JavaScript, fine. If it’s your revenue-driving product copy? That’s a red alert.

Visibility isn’t always the goal

Here’s something that doesn’t get said enough though: not every brand wants to appear in LLM results.

Some media companies are understandably concerned that if AI regurgitates their content directly, no one will click through to their site. Others have ethical objections about AI scraping or copyright. And for subscription sites, it can threaten revenue models.

So what’s the takeaway? Treat visibility as a strategic choice. Maybe you’ll fight for it, maybe you’ll avoid it, but don’t drift into it blindly.

The only constant: change

And just to add insult to injury, e ven if you figure this out today, tomorrow could look different.

Gemini already has access to rendered training data thanks to Google. Perplexity is experimenting with browsers. Partnerships, acquisitions, and new features are reshaping what LLMs can see every month.

Tory hit the nail on the head: “We are in the Wild West times y’all, things are changing all the time every day.”

The best defence is a habit of regular auditing and testing. Keep asking: what do LLMs see right now? And use that intel to make smart decisions about where you invest your dev time.

Wrapping up

JavaScript SEO used to be about making sure Googlebot could render and index your content. Now it’s also about making sure LLMs can – or can’t – see what you want them to.

The rule of thumb:

• If it matters, get it into the response HTML.

• If it shouldn’t be public, keep it out of the source entirely.

• And above all, keep auditing.

And if you’re wondering how to check what’s where? That’s exactly what Sitebulb was built for.

TL;DR

• AI bots mostly read response HTML, not rendered HTML.

• That means content loaded via JavaScript is often invisible to them.

• Critical elements at risk: H1s, product copy, navigation, internal links.

• They also pick up things you might not want surfaced, like code comments.

• Use tools like Sitebulb to compare response vs rendered HTML and prioritise fixes.

• Visibility in LLMs isn’t always the right goal. Choose based on your brand strategy.

• The landscape is changing constantly, so ongoing audits are key.