A website assistant that answers from your real information
An assistant grounded in your actual prices, policies and case studies answers the questions buyers ask at 11pm, and hands over to a human with full context.
The problem
How it works by hand
Most chatbots are a worse FAQ page: they deflect, they hallucinate a returns policy you do not have, and the transcript reads like an apology. Meanwhile the questions buyers actually ask, can you do X, what does it cost, have you done this for someone like me, get asked at 11pm when nobody is there to answer them.
A worked example
What a working version looks like
The work is in the knowledge, not the chat window. Your real information, prices, policies, product details, case studies, past proposals, gets pulled into one structured knowledge base with sources tracked. The assistant answers only from that base: every reply is grounded in a specific document, and when the base has no answer it says so instead of inventing one. Questions it cannot or should not answer, pricing edge cases, complaints, anything commercial, trigger a handover that sends a human the full conversation and the visitor’s details, so the follow-up starts informed rather than from scratch. The knowledge base updates when your documents do, so the assistant never quotes last year’s prices.
The exact tools change per business. The shape does not.
What it needs
Honest inputs, nothing exotic
- 01Your real documents: prices, policies, product details, case studies
- 02An hour with whoever knows where the bodies are buried in those documents
- 03Rules for what the assistant must never answer on its own
- 04Somewhere for handovers to land: inbox, CRM or phone
The payoff
What you get back
The difference between useful and embarrassing is whether it runs on your real knowledge. Buyers get straight answers at any hour, the handovers arrive with context instead of cold, and the transcript is something you would happily read back.
Do it yourself
How you would build this yourself
No course, no upsell. This is the order we would build it in, with the tools named, and a prompt to start from.
- 1
Skip the chat widget for now and build the knowledge base: prices, policies, product details, case studies, as clean markdown files. This is the actual work and nobody can do it but you.
- 2
Test it in Claude first: load the documents into a Project and ask it the questions buyers ask. Where it waffles or invents, the documents need work, not the AI.
- 3
Set the refusal rule early: it answers only from the documents, says so when it does not know, and never quotes prices that are not in the list.
- 4
Then add the window: an open-source chat widget plus an API call to Claude with your documents as context gets you live. Log every conversation and read them weekly.
- 5
Wire the handover: anything commercial or unanswerable sends the full transcript and the visitor’s details to your inbox or CRM, so the follow-up starts informed.
Build a website chat assistant that answers only from my documents (attached: prices, policies, product details, case studies). Rules: every answer must come from a specific document and say which one; if the documents do not cover it, say so and offer to take the visitor’s details; never improvise pricing; complaints and commercial questions trigger a handover that emails me the full conversation and the visitor’s contact details. Start with a simple local page I can test before we touch the live site.
Copy it into Claude Code, fill the brackets, and it will plan the build with you before writing a line of code.
We would rather show you how than bill you. The whole ladder of free help, answers, guides and the weekly build-along, is on the do-it-yourself page.