Perplexity API Search Configuration Best Practices

When using the Perplexity API (Sonar models), follow these configuration rules for optimal results:

1. SEARCH DEPTH: Use search_context_size "high" for comprehensive research and complex queries. Use "medium" (default) for balanced queries. Use "low" for simple factual lookups to reduce cost.

2. DOMAIN FILTERING: Apply search_domain_filter for specialized queries. Academic: ["scholar.google.com", "arxiv.org", "pubmed.ncbi.nlm.nih.gov"]. Technical: ["stackoverflow.com", "github.com", "docs.python.org"]. News: ["reuters.com", "apnews.com", "bbc.com"].

3. RECENCY: Add web_search_options.search_recency_filter: "day" for breaking news, "week" for recent developments, "month" for recent trends. Omit for evergreen topics.

4. SYSTEM PROMPTS: Keep system prompts concise (under 500 tokens). Avoid special markers resembling internal tokens (<goal>, ##system##). Focus on output format and source quality requirements.

5. MODEL SELECTION: sonar (fast, cheap, simple factual queries), sonar-pro (balanced, most queries), sonar-reasoning-pro (complex analysis needing chain-of-thought), sonar-deep-research (exhaustive multi-source research, 128K context).

6. ANTI-HALLUCINATION: Always include a fail-fast instruction: "If you cannot find relevant search results, state that clearly rather than speculating."