Your AI Employee: Terms & Guidelines

Your chatbot is software that reads the instructions and knowledge base you give it, then writes a reply using Anthropic's Claude AI. It isn't a person, it doesn't have judgement or accountability the way an employee does, and it doesn't "know" your business beyond what you've told it.

Think of it as a very fast, very literal new hire who has read exactly the documents you handed over on day one — nothing more, and nothing it wasn't told.

Deploying it responsibly is a handful of habits, not a compliance department:

• Test before you launch. Use demo mode and try the questions your real customers would ask.
• Keep the knowledge base current. If your hours, prices, or policies change, it's still quoting the old ones until you update it.
• Check in on it early. Read through the first week or two of real conversations for anything off.
• Don't feed it more than it needs. Only give it information a customer is allowed to see — nothing internal or sensitive.
• Keep a human reachable. Every template ships with an escalation message — make sure it points to a real inbox or phone number.

A message goes: your customer's browser → your own backend (the Cloudflare Worker you deployed) → Anthropic's Claude API → and the reply comes straight back the same way. Nothing passes through DIYChatbots — we don't run a server that sees your conversations, because there isn't one.

That also means you control where the data goes: Anthropic's API data-retention terms apply to your own account, and it's worth a quick read of their current policy before you launch. If your customers are in a region with specific rules about automated communication or data collection — GDPR in the EU/UK, for instance — check whether your existing privacy policy needs a line added about the chatbot. That's a question for a local advisor, not this document.

• Always offer an exit to a human for anything urgent, sensitive, or emotionally charged.
• Never let it be the only channel for something time-critical — a complaint, a safety issue, a payment dispute.
• Review, don't set-and-forget. A five-minute skim of recent chats each week catches drift early.
• Turn it off if something's wrong. Deleting the backend URL or taking the page down stops it instantly — there's no notice period.

Being upfront that they're talking to an AI is good practice everywhere, and a legal requirement in a growing number of places. Here's wording you can paste into your welcome message or a small note near the chat window — adjust the bracketed parts for your business:

Don't set your AI employee up to:

• Give final answers on regulated advice — legal, medical, financial, tax — without a licensed person reviewing it
• Make the final call on hiring, credit, insurance, or eligibility decisions about a real person
• Collect payment details, passwords, or highly sensitive personal data through the chat window
• Impersonate a real, named staff member rather than a clearly-AI assistant

The chatbot you deploy runs on your own hosting, under your own API account, saying things to your own customers. DIYChatbots provides the template and setup guide as-is, isn't a party to any conversation your chatbot has, and isn't responsible for what it says once you've deployed it. That responsibility — like hiring any new employee — sits with the business that put it to work.