AI-ready websites
Can an AI assistant use your website?
When a customer asks ChatGPT, Claude or Gemini about you, the assistant fetches your site and reads what comes back. Not what your browser shows: what your server sends. For a lot of business sites, that's an empty shell.
What a person sees
Family dentistry in Pune
Check-ups from ₹800. Root canal, implants and braces. Open Saturdays.
Book a visit
What an agent receives
<body> <div id="app"></div> <script src="/main.js"> </script> </body>
0 prices, 0 services, 0 hours
How an agent reads a site
Four requests, in order. Fail one and it guesses or gives up.
Most assistants don't open a browser. They make one request at a time and read the reply. Most sites fail at the third stop, which happens to be the cheapest one to fix.
-
Stop 1
May I?
robots.txt
Is this crawler allowed? Many sites block AI by accident, from a template nobody remembers.
-
Stop 2
What's here?
/llms.txt, /.well-known/*
A map of the site written for machines, so nothing has to crawl everything and guess.
-
Stop 3, the usual failure
The words
page HTML or .md
Your content, without running JavaScript. Plus the facts, like prices, marked up so they're quoted correctly.
-
Stop 4
Do the thing
/api/*, MCP
Optional. A tool the assistant can call: get a quote, check availability, run your calculator.
The nine dimensions
"AI-ready" is nine separate questions.
Most advice answers two of them and calls it done. Here's all nine, what each means, and whether it belongs in a services business's setup. We include what you need and tell you what to skip.
If an agent fetches a page, does it get the words or an empty shell?
words in HTML, llms.txt, a markdown copy of each page
Can it find what it needs without crawling everything?
sitemap, canonical URLs, well-known files
Have you said what agents may do, or are they guessing?
named crawler groups, Content Signals, sane rate limits
When an assistant quotes your prices, will it get them right?
schema.org Service and Offer, prices as data
Can an agent do something for a user, not just read about it?
MCP server, OpenAPI, a skill file, quoted separately
If data belongs to a logged-in user, can their agent log in as them?
OAuth metadata, only if you have customer accounts
Can an agent buy from you without a human filling in the form?
product schema, agentic checkout via your platform
When a model learns from or cites you, is it on your terms?
your training stance, decided on purpose, dates and authorship
Do you know, rather than believe, that all of it works?
every file fetched from outside your network after deploy
The package
AI-ready setup
Levels 1 to 3 for a services business, done properly and checked from the outside. One price, published.
$599one-off, 5 business days
Audit fee comes off if you've had one. Combined with a speed fix, the readability work is usually done in the same pass.
If your words only exist in JavaScript and the fix is a rendering change, we tell you that first, with its own price, before anything starts.
-
Agent-view baseline. Every key page fetched with no JavaScript, words counted, before and after.
-
A written crawler policy. Named AI crawler groups and Content Signals, so nothing is blocked by accident.
-
llms.txt and markdown copies. A map of your site written for machines, and a clean text version of each important page.
-
Your facts as data. Services, prices, hours and locations in schema.org markup, from one source.
-
Discovery basics. A sitemap of real pages, one canonical URL each, at most one redirect.
-
A training decision, made on purpose. We explain the options, you choose, it's written down.
-
Outside verification. Every file checked from outside your network after deploy.
-
Handover notes. What changed, why, and how to keep it true when your prices change.
-
Add-on: an MCP tool. If you have a calculator, quote form or free tool, we can expose it so an assistant can use it directly. Quoted after a look.
Try it on us
We did this to our own site first.
Fetch any of these. It's what an agent gets when it asks about us. Our MCP server exposes three tools an assistant can call: check_site_speed, get_pricing and email_full_report.
$ curl -s www.sitespeedlabs.com/robots.txt | grep -A2 "User-agent: \*" User-agent: * Content-Signal: search=yes, ai-input=yes Allow: / $ curl -s www.sitespeedlabs.com/llms.txt | head -3 # Site Speed Labs > Website speed and AI-readiness, in plain language, for people... $ curl -s www.sitespeedlabs.com/api/pricing.json | jq -c '[.tiers[] | {(.id): .price}] | add' {"audit":"$249","sprint":"$1,500","wordpress":"$799","ai-ready":"$599","care":"$20"}
The honest caveat
Nobody can promise you'll show up in ChatGPT.
All of this is new. llms.txt is a proposal, and Google has said its search doesn't use it. Content Signals is months old. MCP changes every few months. Nobody outside the AI companies knows exactly what each assistant respects today.
It's still worth doing, for two reasons. Everything is reversible. And the parts that matter most (words in the HTML, a written crawler policy, facts stored as data) were good practice long before anyone said "agent". They make your site faster for people too.
Three things to check yourself, today
- Load your page with JavaScript off. If the words vanish, agents see nothing.
- Read your robots.txt and decide on purpose.
- Make sure your prices are text, not an image or widget.
Start with the free check
See your site the way an assistant does. It takes a minute.
Free, no signup, about a minute. Rather ask first? Talk to us.