RAGorWeb search
RAG queries a corpus you chose, split and indexed, and that you answer for. Web search queries a public index where you control neither what exists, nor what comes back, nor what changed since yesterday.
Point by point
| Criterion | Floor 2Floor 2 · The Harnessthe block and its bolted-on plates: what gets added to itRAG | Floor 2Floor 2 · The Harnessthe block and its bolted-on plates: what gets added to itWeb search |
|---|---|---|
| Where documents come from | Yours, chosen | The web, as it is |
| Who answers for the content | You | No one |
| Freshness | That of your indexing | That of the web |
| Confidentiality | The corpus stays with you | The query goes to a third party |
| Upfront work | Splitting, indexing, updating | None |
| What carries authority | Your procedures, your data | Whatever the ranking returns |
On the ground four situations
An internal assistant answers questions on the remote-working policy from the company’s internal memos.
Floor 2 · The Harnessthe block and its bolted-on plates: what gets added to itRAGThe corpus is internal, chosen and maintained. It is what carries authority, and no one else could supply it.
An assistant cites a news article published this morning about a regulatory change.
Floor 2 · The Harnessthe block and its bolted-on plates: what gets added to itWeb searchNo internal corpus holds today’s news. Consulting the outside world is the only route, with the precautions that imposes.
A model answers a question on a piece of legislation from memory, without consulting anything.
Neither one nor the otherThe useful counter-example: this is the dangerous case. Nothing was consulted, the answer is a plausible reconstruction, and it will be phrased with the same confidence.
A legal assistant searches the company’s contracts first, then completes the answer from the official text online.
BothThe right architecture in most cases: the internal corpus carries authority on what belongs to you, the web on what is external and recent, and the answer says where each element comes from.
The test that settles it
Ask who chose the document being cited. If it was you, when the corpus was assembled, it is RAG. If it was a search ranking, at the moment of the query, it is web search.
The trap
Believing that web search removes the need for an internal corpus. On procedures, contracts and in-house data the web knows nothing, and a model that finds nothing will fill the gap rather than abstain.
Check click your answer
Level 1 · Recognise
What fundamentally distinguishes RAG from web search?
Level 2 · Distinguish
An assistant has to answer questions on your company’s leave policy. Which route?
Level 2 · Distinguish
Which risk specific to web search has to be handled explicitly?