Help for owners
I want an AI chatbot inside my app or business
Short answer
A useful chatbot knows your own content, says when it does not know, and stays within limits you set. Before building, decide what questions it should answer, what data it may see, what it must refuse and who reviews its answers. Then start with one narrow use and test it on real questions.
Does this sound like your situation?
- Customers ask the same questions all day.
- I want an assistant that answers from my own documents.
- I am worried a bot will say something wrong about my business.
- I do not know what it will cost to run.
- I want it in my app, my website or WhatsApp.
Why does this happen?
Models can sound sure and be wrong
Anthropic's documentation says even advanced models can produce text that is factually incorrect, and that its techniques reduce this without eliminating it.
Confirmed by platform.claude.comTechnical detail: The model's reply breaks JSON.parse
The bot only knows what you give it
Answers about your prices, policies or stock are only reliable when the bot is fed that information and told to use only it.
Confirmed by platform.claude.com
Customer data needs care
A chat can contain personal details. What is sent to an AI provider, and kept, is a privacy decision for your business.
Technical detail: The OpenAI API answers 429: out of credit, or just too fast?
Usage drives cost and limits
Each question uses the provider's service, and providers apply usage limits, so cost and capacity must be planned.
Technical detail: The OpenAI API answers 429: out of credit, or just too fast?
What can I do about it today?
- Pick one narrow jobFor example answering shipping questions or finding a product. A bot with one job is easier to make reliable.
- Collect the real questionsExport or write down 30 to 50 questions customers actually ask, with the correct answers.
- Decide what it must not doRefund promises, medical, legal or price decisions should go to a person.
- Check your privacy policyIf personal data will reach an AI provider, speak to your adviser or regulator about what you must tell customers.
- Plan a human hand-offThe bot should offer a way to reach a real person.
When should I bring someone in?
- The bot will talk to customers about money, health or legal matters.
- You hold personal data under privacy rules.
- You do not know how to judge whether its answers are right.
How do I handle it?
- Define the job and the limitsThe scope, the allowed sources, what it refuses, and when it hands over to a person.
- Ground it in your contentThe bot is set up to answer from your documents and to say when it cannot find an answer.
- Test on your real questionsI check the answers against the correct ones and keep a record of failures.
- Set cost and usage limitsLimits and alerts so a busy day or abuse does not become a surprise.
- Review after launchReal conversations are read, and the bot is corrected as problems appear.
The price depends on what I find, so I quote in writing after a short discovery call. Tell me what is happening and I reply within 24 hours.
What do owners ask about this?
Will the chatbot make things up?
It can. Techniques such as allowing it to say it does not know, and grounding it in your documents, reduce this but do not remove it, so answers should be checked.
Do I need my own AI model?
Usually not. Most business bots use a provider's model with your content added.
Can it work on WhatsApp or my website?
Yes in principle. The channel's own rules and approvals apply, which I check first.
Sources
How was this page made?
I drafted this page with AI assistance from the official documents listed under Sources. Found a mistake? Tell me.