Searching for sources, not pages: Valyu and Consensus

Most search is page retrieval. A different class of tool retrieves evidence — papers, filings, and publishers — and that changes what the answer can claim.

Nov 16, 2026The avbbc desk
Searching for sources, not pages: Valyu and Consensus
Most search is page retrieval. A different class of tool retrieves evidence — papers, filings, and publishers — and that changes what the answer can claim.

There is a moment in research when you stop looking for a page and start looking for evidence: the question is not 'where does someone discuss this' but 'what does the record say'. General search engines answer the first question well and the second question accidentally.

Two tools in the directory target the second question directly: Valyu, which biases toward sources with provenance, and Consensus, which searches published research.

The short answer

Use evidence search when the answer will be quoted. Use page search when you are orienting yourself. The two produce answers that look the same and carry different obligations.

What provenance changes

A page on the open web might be a vendor's blog, a forum argument, or a press release. A paper has an author, a method, and a venue; a filing has a regulator. When an answer will appear in a document that someone acts on, that difference is the thing that matters.

Valyu makes that its bias: the index favours academic work, filings, and established publishers. That narrows coverage and improves quotability, which is exactly the trade a researcher wants.

Consensus and the 'what does the research say' question

Consensus answers questions over the scientific literature with the papers attached. For a question that has a literature, it is the shortest path from question to citable answer.

The failure mode is the interesting one: for a question where the science is unsettled, the synthesis reads more confident than the evidence. Read the linked papers, not just the summary.

When evidence search is the wrong tool

Question Evidence search Page search
What is this concept? Too narrow Good
What did this company announce? Misses it Good
What does the evidence say about X? Good Noisy
What are people saying this week? No Good
What does the regulation require? Good Unreliable

The third and fifth rows are where provenance pays. The first and second are where a general engine is the honest choice.

The obligation that comes with an evidence answer

A cited paper is not a verified claim. The citation tells you where the sentence came from; it does not tell you that the paper supports the sentence. That is the claim verification step, and it is a separate pass — see verifying claims before publishing.

A worked example

A question arrives: does remote work reduce team innovation? A general search returns opinion pieces and a vendor survey. Consensus returns a set of studies with mixed findings, which is the honest answer — and the synthesis will be smoother than the evidence warrants. The deliverable becomes 'the evidence is mixed, with these two findings', which is more useful than a confident claim and harder to write.

Frequently asked questions

Are these replacements for Exa or Tavily? No. They cover a narrower corpus and a different question type. See search APIs for agents.

Do I still need a general engine? Yes, for orientation and current events.

How do I avoid overclaiming? Read the paper, not the synthesis, for any claim that matters. See search tools.

Last checked: 16 November 2026.

Related: Valyu vs Exa, Consensus vs Perplexity, and search tools.

Related reading

All posts