Lesson 2 of 7
Give it knowledge
6 min read
Your assistant is confident about your refund policy — and completely wrong, because it's guessing from the whole internet. What if it read your policy first?
From guessing to grounded
By default an assistant answers from fuzzy general memory — often close, sometimes confidently wrong. Give it knowledge: attach your documents, and it looks them up before answering. This is called RAG — Retrieval-Augmented Generation: it retrieves the right passage from your files, then writes the answer from that. Grounded in your facts, not the internet's.
Grounding an assistant in your own documents turns a confident guess into a checked, sourced answer.
It shows its receipts
The payoff isn't just a better answer — it's a traceable one. A grounded assistant can point to the exact document it used, so you can check it. Attach a handbook, a product catalogue, a set of FAQs, and it answers new questions from your material, citing where each fact came from.
Grounded answers come with a source you can open. If it can't find the fact in your docs, a good assistant says so instead of inventing it.
Attaching documents narrows the assistant to your facts, but it doesn't make it perfect — it can still misread a messy file. Skim the source it cites before you trust an important answer.
The shape of it
- —By default an assistant guesses from general memory — sometimes confidently wrong.
- —Attach your documents (RAG) and it answers from them, grounded in your facts.
- —A grounded answer cites its source, so you can check where it came from.
You want your support bot to answer questions about your 14-day return window — correctly, every time. What do you set up?
Continue in the app
Take the whole No-code AI Agents & Chatbots course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.