Retrieval and citation are separate evidence layers
Retrieval-augmented generation combines a generator with retrieved external passages. The original RAG paper demonstrates that retrieval can improve knowledge-intensive generation, but it does not describe the proprietary ranking or citation logic of every commercial answer engine.
OpenAI documents linked sources in ChatGPT search, Perplexity documents search results and cited answers, and Google documents links in its AI Search features. None publishes a shared citation formula that applies across products.
Make attribution easy to verify
- State the answer before commentary and keep each important claim close to its supporting source.
- Link to the original documentation or dataset when it directly supports the claim.
- Name the responsible organisation, publication date, method and material limitations.
- Use stable canonical URLs and keep cited evidence accessible in crawlable HTML.
- Correct changed or unsupported claims publicly rather than silently preserving stale authority signals.
Measure more than the presence of a link
- Record the prompt, engine, model or product label, language, locale and collection date.
- Separate brand mention, linked citation, source position and factual attribution accuracy.
- Open every cited source and test whether it supports the answer.
- Repeat the panel before treating a single citation as durable visibility.