Measurement library
Do Perplexity Sonar and Sonar Pro return different sources?
Yes. Sonar and Sonar Pro are different models with different retrieval behaviour, so the same question sent to both can return different source lists, a different number of sources, and a different ordering. This matters for measurement because a citation result is only comparable against another result from the same model. A domain that is cited by one and not the other has not changed; the measurement has.
Published 2026-08-03. Reviewed 2026-08-03. Published by ShiftNode Digital s.r.o.
Why the model changes the source list
An answer engine does two separable things: it retrieves candidate documents, then it writes an answer that may reference some of them. The retrieval step is where citations are decided. Two models in the same product family can differ in how many candidates they retrieve and how many they surface in the returned source list.
That means the denominator of any citation measurement is model-specific. If you measure a domain on one model in June and a different model in July, the difference between the two numbers contains a model change and a visibility change, and you cannot separate them after the fact.
Where this shows up in a GetSounding audit
Guest and Free audits run Perplexity Sonar. Paid audits run Perplexity Sonar Pro and add four further services. That is a deliberate coverage difference, not a quality difference in the counting rule: a citation requires the audited hostname to appear in a returned source URL on either model.
The practical consequence is that a Free result and a paid result are not directly comparable as a time series. Compare Free to Free, or paid to paid.
How to test it yourself
- Fix the question set. Use the same wording for both models, with no brand name in the questions.
- Run both models on the same day. Retrieval indexes change, so a week between runs contaminates the comparison.
- Repeat each question. A single pass cannot distinguish a model difference from ordinary run-to-run variance.
- Record the full returned source list, not just whether you were cited. The interesting result is usually which domains appeared instead.
- Compare like with like. Report the citation rate for each model separately and never average them into one number.
What this page does not establish
- This page explains why the source list can differ. It does not claim a fixed ratio between the two models, because that ratio depends on the question set, the domain and the date.
- Model availability and behaviour change. A result is only evidence about the model label recorded with it.
Sources
Run a free audit to see this measurement format applied to your own domain, or read the full measurement method.