By skyforbes The typical workflow is copying email threads into ChatGPT and hoping it understands the nuance. But it often misses key details, like who committed to what, whether something was actually decided or just discussed, sentiment shifts between participants, implicit deadlines.
Figured out a better approach using structured extraction before sending context to ChatGPT. Instead of raw email text, you pre-process the thread to extract the meaningful signals, i.e. decisions, tasks with owners, commitments, sentiment per person, blockers. Then feed that structured data to ChatGPT.
The difference is pretty significant. ChatGPT can now answer questions like "what did the client commit to?" or "what's blocking this deal?" with actual accuracy instead of guessing based on partial context.
Built this as an API that works with the OpenAI API so developers can add it to ChatGPT-based tools. It handles the messy email parsing (thread recursion, participant tracking, intent classification) and returns JSON that gives ChatGPT agents full conversational context.
If anyone's building tools on OpenAI's API and hitting the email context problem, the API is in early access here: https://form.typeform.com/to/zTzKFDsB
The main insight is ChatGPT is incredibly powerful when it has structured context to reason over, not just raw conversational text. Curious if others have found similar approaches or other ways to solve the context problem.