This is fascinating, but how are you retrieving this data? Are you running the fan-out prompts separately to see the citations or are you somehow harvesting from OpenAI backend response file?
Yaacov Goldstein ·
Questions · ·
Aug 02, 10:17
·
Sign in with Google to reply.
Directly with the model and official grounding sources: https://developers.openai.com/api/docs/pricing#:~:text=Web%20search
The API metadata returns both selected and unselected grounding sources, which you can't see by scraping web chat.