Compare · Building · Research notes
Prompting vs RAG vs fine-tuning — scored
Teams skip the cheap lever. Score the job before you buy a training run.
Last reviewed: 2026-08-15. Re-run the protocol after major model or pricing changes.
How do you pick among the three?
- Name the artifact and the decision.
- If context fits, prompt first.
- If facts live in documents that change, add RAG.
- If style/format is stable and you have a holdout, consider fine-tuning.
What is the scoring rubric?
| Criterion | Weight | How to score |
|---|---|---|
| Fits in context | Gate | If yes, stay on prompts. |
| Corpus freshness | 1–5 | RAG wins when documents move. |
| Holdout quality | Gate | No holdout, no fine-tune. |
| Ops complexity | 1–5 | Each lever adds failure modes. |
Which lever is the default?
| Tool | Best fit | Default | Next lever | Note |
|---|---|---|---|---|
| Prompting | Most jobs | Start here | Add RAG if facts overflow | Cheapest to evaluate. |
| RAG | Document Q&A | Retrieve then answer | Add fine-tune for format lock | Cite chunks. |
| Fine-tuning | Stable behavior | Style/format | Do not use as a knowledge dump | Needs evals. |
Learn the craft, not just the tool
Practice the transferable habit in Fine-tuning vs RAG, then RAG vs fine-tuning, Evals.
FAQ
What should I try first, RAG or fine-tuning?
Prompts, then RAG for changing facts, then fine-tuning if style/format is still wrong.
Can I do all three?
Yes. Prompt + retrieve + a small tune is common. Evaluate each layer.