RAG vs fine-tuning
'Train it on our data' is the usual request, and retrieval is usually the correct answer to it.
The two are routinely confused
Both are described as 'teaching the model your data', which hides the fact that they do fundamentally different things. One shapes behaviour; the other supplies facts at question time.
- Assuming custom training is needed to answer company questions
- Expecting a fine-tuned model to stay current
- Wanting citations from an approach that structurally cannot give them
What each actually does
Fine-tuning continues training on examples, adjusting the weights so the model behaves differently — a different tone, a stricter format, a specialised task. Retrieval leaves the model alone and supplies relevant passages of your content with each question.
Why retrieval wins for facts
Currency, cost, and attribution. Publishing a page and re-indexing takes minutes where retraining takes a cycle; retrieval costs nothing beyond the query; and retrieval can name the source, which a fine-tune cannot because the information is spread through weights with no page to point at.
Currency
Content changes daily; retraining does not.
Attribution
Only retrieval can produce a citation.
Cost
No training run, no specialist, no redeploy.
Where fine-tuning is the right tool
When you need consistent behaviour rather than current facts: a rigid output format, a specialised classification task, or a distinctive voice that instructions alone do not reliably produce. These are behaviour problems, and behaviour is what fine-tuning changes.
Common questions
- Can I do both?
- Yes — behaviour from fine-tuning, facts from retrieval. For most support use cases retrieval alone is sufficient and considerably cheaper to operate.
- Is fine-tuning more accurate?
- Not for facts about your business. It has no mechanism to keep them current and no way to show its source, so an error is both likelier over time and harder to detect.
From the blog
All posts- EvaluatingRAG vs fine-tuning for support: the decision that sticks“Train it on our docs” sounds right and usually isn’t. For website support, RAG wins on currency, citations, and cost; fine-tuning earns its keep for behaviour, not policies.Read
- BuildingPersona tuning without turning the bot into a mascotSet a specific greeting, match your brand’s register, and write hard boundaries. Skip the forced personality that makes refusals feel like a bit.Read
- EvaluatingChatbase alternatives for teams that need cited answers (2026)Matter Chat, Chatbase, Intercom Fin, Tidio Lyro, DocsBot and Chatling compared on citations, refusal behaviour, handoff and what the entry price meters. Checked on the vendors' own pages, with the test to run on all six.Read
Answer honestly. Capture the rest.
Point Matter Chat at your site and see what it can — and can't — answer. It's honest about both.
No credit card. 2 minute setup.
Every answer cites the source it came from. When there isn't one, it says so — and hands the visitor to you.
Installs on the tools you already run.