Force a Perplexity Sonar chat completion to return machine-readable JSON matching a caller-supplied schema instead of free text.

Documented steps

1. Set "response_format": {"type": "json_schema", "json_schema": {...your schema...}} on the chat completions request body (https://docs.perplexity.ai/docs/sonar/features).
2. Keep the schema focused on the facts/structure you need (e.g. answer, confidence, category fields) — do not ask the schema to also contain source URLs.
3. Expect elevated first-token latency (docs cite 10-30 seconds) the first time a brand-new JSON Schema is used, as Perplexity prepares/compiles the schema.
4. Parse choices[0].message.content as a JSON string matching your schema once the response returns.
5. For source links, read the separate citations / search_results response fields instead of asking the schema for URLs — the docs warn requesting links as part of a JSON response may not always work reliably (https://docs.perplexity.ai/docs/sonar/features).
6. Reuse the exact same schema across repeated calls so you only pay the compile-latency cost once per schema, not per request.