Your website might be invisible to ChatGPT

Either the server sends a finished page — all your words, already written into the file — or it sends a nearly empty file plus a set of instructions, and the visitor's browser builds the page on the spot. The second one is how most modern websites work. Squarespace, Webflow and WordPress mostly send finished pages. Anything built as a single-page app, and a surprising number of "custom" builds, send instructions.

For a human being this distinction is invisible. The page appears, it looks right, everyone is happy.

For a machine that doesn't run the instructions, the page is blank. And most of the machines you now care about don't run them. GPTBot, ClaudeBot, PerplexityBot and the rest of the crawlers feeding AI assistants read the raw file and move on. Google runs the instructions, eventually, with delays and caveats — so classic search is partly covered. The assistants people increasingly ask instead of searching are not.

So you can have a beautiful, thoughtful, expensively written website that says nothing at all to the tools your next client is using to decide who to call.

I checked this on a real site in August. Fetching it returned the title, the description, the share image, and then zero words of actual content. Not the hero. Not the manifesto. Not one service, not one answer, not a single sentence anyone had agonised over. The site was excellent. To a machine it was a title and a shrug.

Why this is about to matter more than it did

The shift is not that people stopped searching. It's that a growing share of the questions that used to become ten blue links now become one answer, assembled by something that read a set of pages and decided which ones were worth quoting.

If your pages aren't in that set, you are not competing badly. You are not competing.

And the fix is not what most people assume. Nobody needs to add more keywords, or write another blog post about "the top 7 trends", or install a plugin. The fix is that the words already on your site — the ones you wrote, the ones that sound like you — need to exist in the file the server sends before any JavaScript runs.

What actually fixes it

There are three honest routes, and which one applies depends on how your site was built.

If you're on a hosted platform — Squarespace, Webflow, Shopify, most WordPress themes — you are probably fine already. Run the test. If the number is healthy, the work here is different: it's about whether the words themselves answer the questions people ask, in language a machine can lift cleanly. More on that below.

If your site is a single-page app, the technical name for the fix is prerendering: at build time, the page is rendered once and the result is baked into the file that gets served. The visitor's browser still takes over the moment it loads, so nothing changes for a person. It is usually a day of work, not a rebuild, and it is the highest-return day of work available to most sites right now.

If you don't know how your site was built, the test above tells you what you need to know, and the person who built it will understand the sentence "the content isn't in the server response."

Then the second half of the job

Getting the words into the file is necessary and not sufficient. Once a machine can read your page, the question becomes whether it can use it.

A few things reliably help, and none of them are tricks.

• Answer the actual question in the first two sentences of a section. Assistants quote passages that stand alone. A paragraph that only makes sense after three others rarely gets picked.
• Put your objections on the page, in your own words. The questions people are too polite to ask you are exactly the ones they type into a chat window. A real FAQ, answered properly, is the most quotable thing most sites have.
• Say who you are and what you do in one plain sentence somewhere near the top. Not a tagline. A sentence a stranger could repeat to a colleague.
• Don't hide the good part behind a click. Anything that only appears when a tab is opened or an accordion expands may not exist to a machine unless the underlying text is in the document.

The part nobody says out loud

There's an uncomfortable second-order effect here, and it's the reason I care about this more than the technical fix.

The sites that get quoted are the ones that say something specific. An assistant summarising ten interchangeable agency websites has no reason to name any of them. It will produce the average of the ten, in its own calm voice, and no one will click anything.

Being readable by machines gets you into the room. Having a point of view is what gets you named in the answer.

Which means the work is not, in the end, technical. Making your site legible is a day. Making it worth quoting is the actual job — and it starts with being willing to write down what you think, in language nobody else could have written.

Run the test. Then go and read your own homepage as though you were a stranger deciding whether to call you.