In ChatGPT
More and more people ask ChatGPT or Claude about a mortgage, a pension or what cover costs before they ever visit a website. Subcontext puts your products, your answers and your prices inside those assistants, so when someone asks, the answer comes from you. Everything that is said gets recorded, and it takes about five minutes to switch on.
What's changing
A few years ago, someone thinking about a new mortgage or life cover would search Google and click through to a few sites. Today a lot of them just ask ChatGPT. The assistant gives a general answer, and if it can, it names a company or two. Companies that have set themselves up inside these assistants get named and get the enquiry. Companies that haven't get left out of the conversation, and the customer never finds out they existed.
This page explains the four ways that conversation can reach you, and what we do to make sure it does.
Inside ChatGPT
ChatGPT lets companies add themselves as an "app". Once yours is in place, a person can ask ChatGPT about your products and get a real answer from you, right there in the chat, with your name on it. For an insurer that can be a live quote. They can ask follow-up questions, find out whether they're eligible, and book a call with your team without leaving ChatGPT.
Any prices come from your own rates. The answers come from an assistant that we set up to speak for your company, using the things you've told it are true. And every one of these conversations shows up in your Subcontext dashboard, so your team can read it and follow up.
Here's a quote from YourBrand:
Three quick health questions would confirm this price. Want to go ahead?
Inside Claude
Claude is the other assistant a lot of people use, made by a company called Anthropic. It works the same way: you can add your company to it, and then people who ask Claude about your products get your answer. We set it up alongside the ChatGPT one, with nothing extra to build. In your dashboard the two are kept separate, so you can always see whether a customer came to you through ChatGPT or through Claude.
I checked with YourBrand:
Would you like me to check whether critical illness cover could be added?
On your website
Some AI assistants are built into the web browser. When a customer says "get me a quote from YourBrand", the assistant opens your website and tries to fill in the form itself. Today it has to guess what each box means, and it often gets it wrong or gives up.
There is a new standard, called WebMCP, that lets a website tell the assistant plainly what it can do: get a quote, answer a product question, book a callback. Websites and quote forms built on Subcontext support it. So the assistant asks your site for a quote in a way it understands, gets a proper answer, and shows it to the customer. No guessing, no broken forms.
Broker and partner systems
Brokers, comparison sites and other partners are starting to use AI assistants of their own. Rather than a person emailing you for a quote, their software asks your software, and gets an answer in seconds. The standard for this is called A2A, short for "agent to agent".
On Subcontext, each partner you choose to work with gets their own way in. Their requests are answered by the same assistant that speaks for you everywhere else, every request is recorded, and the conversation appears in your dashboard just like one from a customer.
Software asking software, with you in the middle.
Everything is recorded
The natural worry is that a conversation inside ChatGPT is happening somewhere you can't see. With Subcontext it isn't. Whether the question came from ChatGPT, Claude, a browser assistant or a partner's system, we keep the same record we keep for your web chat:
Your team reads these conversations in the same inbox as everything else and can step in whenever they want.
Staying compliant
The assistant only says things you've told it are true about your products. It only offers what you actually sell. Before it answers, and again before the answer goes out, it is checked against the rules you've set, and anything that shouldn't be said isn't. If someone seems vulnerable, they get passed to a person. New leads and callback requests go to a named member of your team. And if you ever want it switched off, it's off the moment you say so.
The compliance page explains each of these in more detail.
Getting started
There's nothing to build. If your products are already on Subcontext, it goes like this.
That gets your team using it straight away. Making the app available to the public in ChatGPT's own directory is a separate step that goes through OpenAI, the company behind ChatGPT. We prepare everything they ask for.
Questions you'll be asked
No. The setup is done from the Subcontext dashboard by whoever manages your products. The only technical-looking step is pasting a link into ChatGPT's settings, and we show exactly where.
Yes. Quotes come from the same rates and rules your website uses. Change a price once and ChatGPT, Claude and your website all quote the new one.
It only says what you've told it is true about your products, every answer is checked against your rules before it goes out, and anything it can't answer gets passed to a person. You can read every conversation afterwards.
Yes, instantly, from the dashboard. It stops answering the moment you do.
Into your Subcontext inbox, like a web chat. ChatGPT or Claude carry the conversation; the quote and the customer's details are recorded on your side.
Once your app is listed in ChatGPT's public directory, yes. Listing is a separate step that goes through OpenAI. Until then you and anyone you share the link with can use it, which is how most firms start.
Book a call and we'll set it up with you on the call.