How Document Analyst answer relevance works

The problem this solves

When a user uploads a large document, Document Analyst reads content from the top down. If the answer to their question sits near the end of a long file, it may fall outside the word limit and get cut off entirely.

Improve answer relevance fixes this by reordering the document's content before the agent reads it, so the most relevant parts are always prioritized regardless of where they appear in the file.

How re-ranking works

When Improve answer relevance is on, a semantic re-ranker (powered by Cohere) scores each chunk of the uploaded file against the user's question. Chunks are reordered from most to least relevant. The agent then reads from the top of that reordered list, within its usual word limit.

The word limit itself does not change. Only what gets prioritized within it.

When it makes a difference

Re-ranking has the most impact when:

• The uploaded file exceeds your agent's word limit
• The relevant content is buried later in the document- contracts, reports, long PDFs

Example: A user uploads a 50-page contract and asks about the termination clause. Without re-ranking, the agent reads the first 3,000 words regardless of where the clause appears. With re-ranking, the termination clause chunks are scored as highly relevant and read first.

When it makes no difference

If the uploaded file is within your agent's word limit, it is read in full. Re-ranking has no effect in this case.

Default behavior

Improve answer relevance is on by default for all new agents. For existing agents it is off, to avoid changing established workflows. It is recommended to enable it.

To enable or disable this setting, see 👉 [Configure Document Analyst].

Related articles

• Configure Document Analyst
• Document Analyst Overview
• Document Analyst Best Practices
• Document Analyst limits