A growing share of people now ask an assistant instead of typing into a search box. They ask ChatGPT for a recommendation, or read the AI summary that sits above Google's results, and act on whatever comes back. Those tools reach a website the same way a search crawler does, but they read it rather differently, and the gap between what a person sees on your website and what an assistant can actually use is wider than most people expect.
None of what follows is a criticism of anybody's website. Almost every website built in the last decade was built for human readers and for classic search, because that is what mattered at the time. These are the ordinary, fixable differences I run into when I look at a website the way a machine does.
They read the page as it arrives, not as it looks
Plenty of modern websites send the browser a small file and then build the visible page afterwards using JavaScript. That is a completely normal way to build, and a visitor never notices, because their browser does the assembly in a fraction of a second.
Not every crawler does that assembly. Google's does. Several of the AI crawlers currently do not, or do it inconsistently. If your services, your location or your contact details only appear after that in-browser step, some assistants are reading a page that is effectively blank.
Anything you would hate to be missed should be present in the page the server sends, not added afterwards. You can check this quickly: open the page, view source, and search the raw code for your phone number or a service you offer. If it is not in there, a crawler that does not run JavaScript will not find it either.
Facts have to be written down, not implied
A person landing on a roofing company's website will work out the service area from the photographs, the accent in the video, and the fact that half the case studies mention a nearby town. A language model has none of that context to lean on. It has the words on the page.
So the things you have never bothered to state plainly, because everyone local already knows them, are usually the things that go missing. Which towns you cover. Whether you work with domestic customers, commercial ones, or both. What your hours actually are. Whether you charge for a callout. Whether you are the person who comes out or you send a team.
Write the obvious things down in plain sentences. "We cover Stoke-on-Trent, Newcastle-under-Lyme and the surrounding villages" is worth more to an assistant than a map graphic, because the assistant can quote the sentence back to somebody who asked.
Structured data does the introductions
Structured data is a small block of machine-readable information that sits in the page code and states the facts about your business directly: the name, the address, the phone number, the opening hours, what kind of business it is. It follows a shared vocabulary called schema.org, which search engines and a number of AI tools both understand.
It does not change how the page looks. It exists purely so that software does not have to guess, and guessing is where wrong answers come from. A wrong phone number in an AI answer is a lost enquiry you never hear about.
Structured data is a one-off job on most websites and it removes a whole category of misunderstanding. If your website already has it, the thing worth checking is whether it still matches reality after any change of number, address or hours.
A plain summary for machines
An emerging convention called llms.txt is worth knowing about. It is a single plain-text file at the root of a website that summarises what the business does and points to the important pages, written for AI assistants rather than for people.
Being straight about this: it is a proposed convention rather than an agreed standard, and support for it varies. It costs very little to add, it cannot harm anything, and it gives an assistant a clean summary to work from instead of an inference drawn from your homepage. I publish one for this website, so you can see what a real one looks like.
Being quoted needs a page worth quoting
Assistants answer questions, so they favour pages that answer questions. A heading phrased the way a customer would ask it, followed by a direct answer in the first sentence or two underneath, is far more quotable than the same information spread across three paragraphs of introduction.
That is also better for the people reading it, which is the useful thing about most of this work. Very little of it is a trick played on a machine. It is mostly writing clearly and making sure the facts are where they can be found.
How to check where you stand
You can do a rough version of this yourself in twenty minutes. View the source of your main pages and search for the details that matter. Ask an assistant what it knows about your business and see what it says, including whether it gets the details right. Search for your business name and read the AI summary rather than the links.
If the answers come back thin or wrong, that is a normal starting position rather than a problem with your website. It is also the sort of thing my SEO Management service reports on every month, alongside the more familiar technical and on-page checks. A related piece worth reading next is why your business details need to match everywhere they appear, since an assistant that finds three different phone numbers for you has the same problem from the other direction.