Blog · Redman Studio, Watford: Ask ChatGPT who to hire. What does it say about you?

Try this before you read on. Open ChatGPT, Claude or Perplexity and ask it to recommend a web designer in your town. Then ask it what your own business does and what it charges.

I run Redman Studio in Watford, and I ask that second question about my clients' sites every week. The answer is usually wrong, vague, or about a different company with a similar name. Not because the assistant is stupid. Because the site gave it nothing to work with.

The assistant can only quote what is written down

An AI assistant does not browse your site the way a person does. It reads the HTML, the same raw text a search crawler gets, and it reads it once, quickly. If your page builder renders the words with JavaScript after the page loads, the assistant may never see them. If your prices are in a PDF or behind a "get a quote" form, it has no price to give. If your name, address and services are spread across three slightly different versions on your site, your Google profile and a directory, the assistant picks one and it is often not the one you want.

Most small business sites fail that test, and most of their owners have no idea, because the site looks fine in a browser.

A one-minute check you can run on your own site

You do not need me for this.

1. Open your homepage, right-click, choose "View page source", and search for a sentence from your own homepage. If it is not there, the assistants are not reading it either.
2. Add /llms.txt to the end of your domain. If you get a page not found, the models have no plain-text map of your business. Mine is at redmanstudio.co.uk/llms.txt.
3. Add /robots.txt. Look for GPTBot, ClaudeBot and PerplexityBot. If they are blocked, or simply not mentioned, you have told the assistants nothing either way. Mine welcomes them by name.
4. Ask an assistant: "What does [your business] do, where is it, and what does it charge?" Compare the answer with the truth.

Step four is the one that stings. I run it on prospective clients' sites and send them the answer. It is free, and occasionally uncomfortable.

What I did to my own site

I sell this as a service, so my site had to pass its own test. Three things.

Every word is in the HTML. The site is built static-first. What you see in the browser is exactly what a crawler gets, with JavaScript switched off. There is no page builder between the words and the reader.

The machine-readable version comes from the same place as the pages. Schema markup is the structured data that tells a machine what things are: a business, a person, a price, an opening time. On most sites it is bolted on by a plugin and drifts out of date. On mine the schema, the llms.txt and the pages are all generated from one set of content, so they cannot disagree. The Companies House number is in the graph, pointing at a public record a machine can check rather than a claim it has to trust.

An assistant can look up my real prices and send me an enquiry. This is the part I have not seen another studio do. I run a small MCP server at redmanstudio.co.uk/mcp. MCP, the Model Context Protocol, is the open standard assistants use to talk to outside tools. Connect the URL to an assistant that supports it and the assistant gets four things it can do:

• read who the studio is, what it builds and what clients say
• look up the published fixed fees, so it quotes £3,500 for a Launch build because that is the figure, not a guess
• list the live client sites with links
• send an enquiry to my inbox, with the person's name and message, only when they ask it to

Enquiries arrive marked with the name of the assistant that sent them. Reads are free and unlimited. Enquiries are rate-limited so nobody can empty a script into my inbox.

Every piece of this is annotated on my proof page, with the live schema and the actual files, so you can check the claims rather than take them.

What this does not do

I want to be straight about the limits, because the AI-search industry mostly is not.

There is no public directory of business MCP servers yet that assistants browse on their own. Someone has to connect mine, and today that is a technical step most people will not take. I am listing it in the MCP Registry so that changes as the assistants add discovery, but right now the server is proof of discipline more than a lead channel.

The things that work today are the boring ones. Words in the HTML. One version of your name and address everywhere. Prices on the page. A plain-text map at llms.txt. Crawlers welcomed by name. None of that needs a server. All of it needs someone to decide the site is for machines as well as people, and to build it that way from the start.

If you want me to run the check

Send me a URL. I will run the four checks above on it and reply with what an assistant currently sees. No charge, no obligation. If the answer is "it is fine", I will tell you that too.

Paul Redman runs Redman Studio, an independent web design and engineering studio in Watford, Hertfordshire. The published price list, the proof page and the MCP server are all on this site.