Why your website is invisible to AI assistants, and how to fix it

Something has changed in how people find businesses, and most websites haven't noticed. A growing share of customers no longer type "IT support London" into Google and scan ten blue links. They ask an assistant: "Which IT support companies in London would you recommend for a 30-person charity?" The assistant reads whatever it can, weighs it, and returns three names. There is no page two. If your business isn't one of the three, the customer never sees you, and you never know it happened.

This article explains what makes a website readable to AI assistants, what quietly makes one invisible, and what we changed on our own site to go from a visibility score of 21 to 60 in a single evening. If you'd rather see your own number first, our free AI visibility check takes about a minute.

How do AI assistants decide who to recommend?

Assistants draw on two sources. The first is training data, the vast snapshot of the web the model learned from, which explains why they recommend Amazon or KPMG from memory. The second, and the one that matters for a small business, is live browsing: when a user asks, assistants like Perplexity, ChatGPT with search and Google's AI overviews fetch pages in real time and read them. What they can read, they can describe. What they can't read, they skip.

"Read" is the operative word. An assistant is not a person looking at your design. It's a machine parsing text and data. Your beautiful hero video, your prices in a PDF, your opening hours in an image of a sign: all invisible. This is the whole game, and it's why sites that look wonderful to humans can be blank to machines.

What makes a website invisible to AI?

When we built our visibility check, we reduced it to eight things, weighted by how much each one costs you when it's missing.

• AI crawlers are blocked. Many sites, often without knowing it, block GPTBot, ClaudeBot and PerplexityBot in robots.txt, sometimes because a plugin did it by default. A blocked crawler cannot read anything at all.
• No structured data. Schema.org markup (Organization, LocalBusiness, Service, FAQPage) is the machine-readable version of your business: name, address, hours, services, prices. Without it, an assistant is guessing from prose, and often guesses wrong.
• Key facts aren't text. Prices, opening hours and service areas that live only in images, PDFs or videos cannot be compared, quoted or matched to a customer's question.
• Unclear identity. If the homepage never says plainly what you are, for whom, and where, an assistant can't match you to "IT support for charities in London" even if you are exactly that.
• No llms.txt. An emerging convention: a plain-text file at /llms.txt that tells assistants what your business is and which pages matter. Cheap to add, increasingly read.
• No answerable content. Assistants lift answers from pages that pose and answer a question. "How much does IT support cost?" followed by a real answer gets quoted. Brochure prose gets skipped.
• Hard to crawl or stale. No sitemap, slow pages, login walls on key information.
• Missing sector essentials. The one fact a customer in your sector asks first: a menu with prices, care home availability, a trade's call-out charge, an IT firm's retainer price and response times.

A worked example: our own site, 21 to 60

We ran the check on elated.consulting before writing this. Technical readiness: 21 out of 60. Not a great advert for a company about to sell this as a service, which is precisely why we're showing you.

The site had no structured data at all. No llms.txt. Prices existed in agreements but not on any page. The homepage headline, "Your technology department", is a fine line for a human and says nothing a machine can use. And the questions people actually ask, cost, coverage, "do you do CCTV", were answered nowhere.

What we changed, in order of impact:

1. Organization and Service schema on every page, generated from one source of truth so it can't drift from the site: name, address, phone, hours, area served, audience, Companies House number, and each of our six services with a description and a "from" price.
2. An FAQ page written the way people ask: "How much does managed IT support cost?", "Which areas do you cover?", "Do you install CCTV?", each answered in two specific sentences, with FAQPage schema.
3. A plain identity sentence on the homepage: "Elated Consulting is an independent IT consultancy in London that acts as the technology department for organisations of 5 to 250 people."
4. llms.txt, generated from the same data as the schema.
5. Explicit permission for AI crawlers in robots.txt, rather than relying on the default.

Result: 60 out of 60 on technical readiness. Every check green. As far as a website can make a business readable to AI, ours now is.

The honest part: readability isn't reputation

Our assistant-presence score, the part that measures whether assistants actually recommend us when asked, was 0 out of 40 before the changes, and it was still 0 the same evening. That's not a failure of the fix; it's how this works. Technical readiness is what you control and can finish in a day. Presence is built over weeks, as browsing assistants re-crawl a site they can now read, and as evidence accumulates off-site: a complete Google Business Profile, consistent listings, reviews that describe the work, links from clients and partners, and content that answers specific questions with numbers.

Anyone who promises to get you "ranked in ChatGPT" in a week is selling something that doesn't exist. What can be promised is this: a site that assistants can read and quote, and a measured view of whether they do, updated as it changes.

What should you do first?

Run the free check; it tells you which of the eight you're missing and what each one costs you. Then fix them in the order above: schema, an FAQ, the identity sentence, llms.txt, crawler permissions. Most businesses can close every technical gap in two to three weeks with their web developer, or we can do it as a fixed-price piece of work. Then start on the off-site evidence, and measure monthly rather than hoping.

The businesses that win the next few years will publish in both formats: a site that persuades a human, and data a machine can rely on. Neither alone is enough anymore.