Contextual help in MS Copilot needs a simple refresh button

Copilot can’t hard‑flush its grounding ctx… a tiny layer always stays, so for a clean run just open a new chat from the pane; if it still sticks, reload the host app once.

Copilot experiences already include ways to clear or reduce prior conversational context, which address the need to “start over” and avoid earlier prompts influencing new answers.

Two relevant patterns from existing Microsoft Copilot integrations:

1. Explicit “clear chat / start over” controls
   • In Power BI, the clear chat button removes the contents of the current conversation and any previous context set throughout the conversation. This is effectively the same as starting a new chat and is recommended when switching topics to avoid overloading Copilot with unrelated prior context and to ensure focused responses.
   • In Dynamics 365 Business Central, selecting Start a new Copilot chat session clears Copilot’s memory of the last few messages. This is specifically recommended after a lengthy conversation so Copilot can deliver more accurate answers on a new topic.
   These are direct precedents for the “simple refresh button” concept: a visible control that resets conversational context without changing underlying data.
2. UX guidance for collaborative, controllable AI
   • Microsoft’s UX guidance for generative AI emphasizes collaborative UX, where users guide the copilot via a continuous feedback loop between inputs and outputs.
   • Recommended output design patterns include:
     • Showing inputs and outputs together so users understand how prompts shape responses.
     • Keeping a history of prompts and outputs so users can explore different directions without losing earlier results.
     • Adding “appropriate friction” at key moments (save, share, copy, paste) and clearly signaling that AI-generated content may be incorrect, encouraging users to review and fact-check.

Applied to Microsoft 365 Copilot contextual help, a “refresh” or “start over” button is consistent with these patterns:

• It gives users explicit control to drop prior context and explore alternative lines of inquiry, helping mitigate echo-chamber or confirmation-bias effects.
• It can coexist with a visible history, allowing users to branch into a new conversation while still being able to refer back to previous answers if needed.
• It aligns with the guidance to make AI behavior understandable and steerable, rather than opaque.

From an implementation perspective for Copilot-based apps or extensions:

• Provide a clearly labeled control (for example, Clear chat, Start over, or New session) in the Copilot pane.
• On activation, clear the in-session conversation state used for grounding responses, without refreshing or altering the underlying data or model.
• Optionally maintain a separate, user-visible history (e.g., a timeline of prior sessions) so users can revisit earlier outputs without them influencing the new session.
• Combine this with existing UX best practices: AI notices/disclaimers, encouragement to review and edit content, and support for trying alternative prompts.

This pattern is already validated in products like Power BI and Business Central and can be adopted in Microsoft 365 Copilot-based solutions to give users a simple way to reset context and explore different perspectives.

References:

• Copilot for Power BI overview
• Chat with Copilot (preview)
• Creating a dynamic UX: guidance for generative AI applications