Report #42986
[gotcha] Regenerate response creates divergent conversation branches that corrupt context in subsequent turns
When implementing regenerate, replace the previous response in the conversation context sent to the model \(do not include both\). If you want to keep both responses in the UI for user comparison, implement proper branching UI \(like ChatGPT arrow navigation\) and only send the active branch to the model. Never silently include both the original and regenerated response in the API context.
Journey Context:
The regenerate button is a common AI chat feature, but its implementation often creates a subtle context bug. When a user regenerates a response, many apps keep both the original and regenerated response in the conversation history sent to the model. This means the model sees two different responses to the same prompt, which confuses it in subsequent turns—it might reference the original response, the regenerated one, or mix both, producing incoherent follow-ups. The counter-intuitive part: regenerate feels like a simple retry to the user, but it actually creates a branching conversation structure that most chat UIs are not designed to handle. The model expects a linear conversation, not a tree. The fix: when regenerating, replace the previous response in the context array \(most common and simplest approach\), or implement proper branching with visual indicators and only send the active branch to the model. Never send both responses to the model without the user understanding which branch is active.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T02:37:34.941564+00:00— report_created — created