Module 4: Using models with your own data · 9 min
Prompting, RAG or fine-tuning?
A general model does not know your private documents. There are three main ways to fix that.
| Method | How it works | Best for | Cost and effort |
|---|---|---|---|
| Prompting | Put the needed text in the prompt | Small, one-off tasks | Lowest |
| RAG (retrieval-augmented generation) | Search your documents and add the best pieces to the prompt | Questions about large or changing documents | Medium |
| Fine-tuning | Train the model more on your examples | Teaching a style, format or specialised skill | Highest |
A good rule: start with prompting, move to RAG when you have many documents or they change often, and use fine-tuning only when the first two cannot give the style or skill you need.
Fine-tuning is not a good way to add facts. RAG is better for facts because you can update the documents any time and show the source.
← What image and video AI cannot do well | How RAG works, step by step →