0
I’m using GPT to produce structured data for an application, and the easy cases work well. The tricky part is a schema with nested objects, arrays, and fields that should appear only when the source information is available.
How dependable is GPT’s JSON output in that situation? I’d like to hear what failure modes people see—such as missing keys, unexpected values, or valid JSON that still breaks the schema—and what validation approach works well before using the response in production.
