sonar-pro

sonar-pro is the model Perplexity's AI search service uses to answer questions. If a typical chatbot is like a librarian who answers only from what it already learned, sonar-pro is more like an errand runner who, the moment you ask something, goes out onto the internet, digs up relevant pages, and summarizes them into an answer.

The problem is what sources this errand runner brings back. If it grabs a flyer from just any shop, that flyer ends up folded into the answer regardless of whether it's actually trustworthy. A recent investigation found that both sonar and sonar-pro, when answering software recommendation questions, frequently pulled citations from obscure sites with almost no visitors — even from blog posts by companies unrelated to the market in question.

The article explains that researchers "accessed Perplexity's two models, sonar and sonar-pro, via OpenRouter" and asked them about 380 software categories. sonar-pro is the model Perplexity actually uses to search the web and retrieve source documents when generating answers, and the investigation pointed out that many of the citations it pulled came from obscure, low-traffic sites. This doesn't mean sonar-pro fabricated falsehoods itself — rather, it means the quality of the material it actually retrieved from the web was low, and that distinction shouldn't be missed.

Ask Perplexity for a software or product recommendation, then click through each source link shown below the answer.

Example prompt: "Recommend the 5 best [category] software tools and tell me the sources behind each recommendation."

Check for yourself whether the sites cited as sources are ones you've actually heard of, and whether they're genuinely relevant to that field. Doing this gives you a feel for how to critically filter AI search results.