How to teach your AI assistant with your own information

The first time you try an AI assistant, you are amazed at what it knows. The second time, you are frustrated by what it does not know. You ask for your hours and it makes them up. You ask the price of a service and it answers something reasonable, but false. It is not that it is dumb: it is that you never told it how your business works.

The good news is that teaching it your stuff is easier than it seems, and it almost never requires coding. The key is to understand that an AI assistant is not trained by memorizing: it is trained by giving it access to your information at the right moment.

Why a generic assistant does not know your business

AI models learn from huge amounts of public text: articles, books, web pages. That gives them two natural limits. First, their knowledge has a cutoff date: they do not know what happened after their training. Second, they have no access to your private information: your prices, your calendar, your policies.

There is a third, more treacherous problem: when they do not know something, they tend to sound confident anyway. That is called "hallucinating," inventing an answer with the tone of truth. For a business, a confident false answer is worse than an "I do not know."

An AI assistant without your information is like a brilliant new employee nobody handed the manual to. It has talent, but it answers with guesses.

The solution is called a knowledge base

The most common way to teach an assistant your stuff is known in the industry as RAG: retrieval-augmented generation. It sounds complicated, but the idea is simple. Instead of asking it to memorize everything, you give it a library with your information and, every time it is asked something, it first looks up the answer in that library and then responds based on what it found.

That library is your knowledge base. It can include your price list, your hours, your FAQs, your cancellation policy, service descriptions. The assistant does not invent: it reads your material and answers grounded in it. AWS and IBM describe this approach as "grounding" the answer in an authoritative source, and it is what reduces the made-up replies.

This brings two big advantages for a business. The first is trust: because the assistant answers based on your information, it stops inventing and starts giving real, verifiable facts. The second is control: you decide what is in the library, so you decide what it can and cannot say. If you do not want it to talk about a topic, you simply leave it out; if you change a policy, you change the document. Your knowledge base is, in practice, your assistant's script.

What information to give it, and how

You do not need to write a treatise. You need to gather what your customers ask again and again, and put it clearly. A good starting point:

• Your services or products, with a short description and the price.
• Your business hours and your address or coverage area.
• Your real FAQs, the ones you answer ten times a day.
• Your clear policies: cancellations, warranties, payment methods, deposits.
• The tone you want it to use: friendly, formal, brief, warm.

Format matters less than clarity. An orderly document, with headings and short answers, works much better than an endless text where everything blends together. Think of it this way: each answer should be able to stand on its own.

A trick that works very well is to review your real conversations from the past few weeks. The questions that come up most often are, almost always, the ones your knowledge base should answer first. Do not guess what your customers will want to know: look at what they already asked you. That list is the best possible first draft, because it comes from the reality of your business and not from your imagination.

Keeping it alive: the most common mistake

Here is the trap almost everyone falls into. You train your assistant once, it comes out perfect, and six months later it gives old information: a price that went up, hours that changed, a service you no longer offer. The assistant is not at fault; you gave it an old map.

A huge advantage of the knowledge base is that updating it does not require retraining anything: you change the document and, from then on, the assistant answers with the new version. That is why it pays to review your base every so often, especially after changing prices, hours or promotions. Five minutes a month prevents wrong answers.

Takeaway

A generic AI assistant does not know your business because you never told it how it works. Teaching it is not coding: it is giving it an orderly knowledge base with your prices, hours, FAQs and policies, so it answers grounded in your real information and not in guesses. And since that base updates without retraining, the most important thing is to keep it alive. A well-fed assistant stops inventing and starts sounding like you.

Sources

• IBM — https://www.ibm.com/think/topics/retrieval-augmented-generation
• AWS — https://aws.amazon.com/what-is/retrieval-augmented-generation/
• AWS Prescriptive Guidance — https://docs.aws.amazon.com/prescriptive-guidance/latest/agentic-ai-serverless/grounding-and-rag.html
• IBM Research — https://research.ibm.com/blog/retrieval-augmented-generation-RAG
• MinIO — https://www.min.io/learn/retrieval-augmented-generation