Feature Request: AI-Mediated Bug Reporting and Automated Issue Triage
ChatGPT already has something traditional bug-reporting systems lack: an AI participating directly in the interaction where a problem occurs.
When a user believes something has gone wrong, the conversational model can often distinguish a misunderstanding from behavior that appears genuinely inconsistent. However, even when the model and user identify strong evidence of a probable product issue, there is no direct mechanism for that information to enter a structured internal triage process.
I am not suggesting that conversational models should directly create engineering tickets.
Instead, ChatGPT could submit structured diagnostic reports to a separate automated triage system. That system could:
Separate likely user misunderstandings from credible anomalies.
Group reports describing similar symptoms.
Correlate reports with available backend telemetry.
Measure how many users, platforms, and software versions appear affected.
Remove unnecessary conversation content and personal information.
Combine hundreds of duplicate reports into a single useful incident.
Escalate high-confidence or rapidly growing patterns for human review.
For example, one user reporting that conversation history disappeared may not justify engineering attention. But if hundreds of independent ChatGPT sessions report similar symptoms and backend telemetry shows a common synchronization or revision pattern, the triage system could produce one report containing the scope, timestamps, affected platforms, supporting evidence, and confidence level.
The conversational model would not determine the root cause. It would report what the user observed and what it could independently observe. Backend systems would provide authoritative technical evidence.
This could reduce engineering noise rather than increase it. Engineers would receive aggregated, evidence-backed incidents instead of thousands of individual complaints written with different terminology.
It would also turn ChatGPT’s enormous user base into a distributed product-quality signal. Users are already telling ChatGPT when something behaves incorrectly. At present, much of that information has nowhere structured to go.
In short:
User encounters problem → conversational AI helps determine whether there is a credible anomaly → structured report goes to automated triage → similar reports are clustered and correlated with telemetry → significant patterns are escalated to humans.
This would provide a more accurate picture of product problems while requiring less manual reporting from users and less manual sorting by engineers.