Same query yields inconsistent results between Perplexity API (sonar) and Web UI: API matches the wrong person, Web UI is correct

General
,
1

Hi Perplexity team,
We’ve encountered a reproducible issue while using the Perplexity API (sonar): the same query produces clearly different conclusions between the API and the Web UI. The Web UI response is more cautious and aligns with what we expect, but the API returns a confident “profile-style” answer that appears to be mapped to a different person with the same/similar name, resulting in an incorrect conclusion.

What we’re seeing

  • API result: The API confidently outputs a person’s “title/experience/background,” but after verification it appears to be a wrong match (possibly due to differences in entity disambiguation, retrieval, or reranking).

  • Web UI result: The Web UI indicates that the exact spelling does not directly match, then more carefully suggests possible name variants/candidates and explains the reasoning based on sources. This outcome is more accurate.

Steps to reproduce

  1. Use a query in the following format:
    "<PersonName> <Region> <Company> LinkedIn"

  2. Call the Perplexity API (sonar) with that query and obtain the answer.

  3. Run the same query in the Perplexity Web UI and compare the result.

Expected behavior

The API result should be consistent with the Web UI, or at least:

  • avoid confidently asserting a single identity when the match is ambiguous; and/or

  • provide a Web-like disambiguation approach (name variants, multiple candidates, and stronger source-grounding).

Actual behavior

The API returns a highly confident answer that appears to be misattributed to the wrong person, while the Web UI correctly handles ambiguity and avoids the wrong mapping.

Questions

  1. Do the Web UI and API use different retrieval/reranking/entity-disambiguation pipelines or different model versions?

  2. Is there any API parameter/mode to make the API behavior closer to the Web UI (e.g., stricter grounding, citations-only output, stronger disambiguation)?

We’d appreciate any guidance on how to make the API results more consistent with the Web UI, or how to prevent the API from producing confident but incorrect entity mappings in ambiguous cases. Thank you!

2

Hi @hellowo, Right now there isn’t enough concrete detail for us to reliably reproduce what you’re seeing. To dig in, could you share:

  • The exact query you’re sending to the API
  • The full API request payload (model, parameters, etc.)
  • A sanitized snippet of the API response that shows the incorrect “profile-style” answer
  • A screenshot or copy of the corresponding Web UI answer for the same query

That will let us check whether there are any differences in configuration, model, or search behavior that explain the mismatch.

Separately, for new integrations we strongly recommend using the Agent API with the web_search tool instead of calling the legacy Sonar models directly.

You can find the Agent API docs here:

If you share a concrete example (request + responses), we can check whether this is a configuration issue, a model behavior difference, or something we should treat as a bug on our side.