When you ask Copilot a question, the Thinking mode indicator appears and stays on screen for an unusually long time. Instead of receiving an answer within seconds, you may wait 30 seconds or more. This delay can happen in Microsoft 365 apps like Word and Teams or on the Copilot web interface. The root cause is often a combination of network latency, large context windows, or service-side processing limits. This article explains why Thinking mode stalls and provides specific steps to reduce response time.
Key Takeaways: Speed Up Copilot Thinking Mode
- Copilot pane > Reset conversation: Clears the current context and restarts the session, which often resolves slow thinking.
- Network latency test: Run a speed test to verify your connection has at least 10 Mbps download and low jitter.
- Microsoft 365 admin center > Service health: Check for active incidents or degraded performance on the Copilot service.
Why Copilot Thinking Mode Takes Too Long
Copilot uses a large language model hosted on Microsoft Azure servers. When you submit a query, the following chain must complete before you see a response:
- Your request travels from your device to the nearest Azure data center.
- The service authenticates your Microsoft 365 license and checks usage quotas.
- Copilot retrieves relevant context from Microsoft Graph, including emails, documents, and calendar data.
- The language model processes the prompt and generates the reply.
- The reply is sent back to your device.
Delays occur when any of these steps take longer than expected. The most common causes are:
Network Congestion or High Latency
Copilot requires a stable internet connection with low latency. High jitter or packet loss can cause the service to time out internally, forcing the Thinking mode to spin until the connection recovers.
Large Context Window
If your prompt references a very long email thread, a large document, or multiple files, Copilot must process more tokens. The maximum context window for Copilot is 128,000 tokens. When you approach that limit, processing time increases significantly.
Service-Side Throttling or Degradation
Microsoft applies rate limits per user and per tenant. If many users in your organization submit queries simultaneously, Copilot may queue requests. Additionally, the Copilot service itself may have degraded performance due to an ongoing incident.
Steps to Reduce Copilot Thinking Mode Time
Follow these steps in order. Test after each step to see if the delay is resolved.
- Reset the current conversation
In the Copilot pane, click the Reset conversation button. This clears all previous context and forces a fresh session. Without accumulated context, the model has fewer tokens to process. - Test your network connection
Open a browser and go to fast.com or speedtest.net. Run a test and confirm download speed is above 10 Mbps and latency is below 50 ms. If speeds are lower, restart your router or switch to a wired connection. - Check Microsoft 365 service health
Go to the Microsoft 365 admin center at admin.microsoft.com. Select Health > Service health. Look for any incident labeled Copilot or Microsoft 365 Copilot. If an incident is active, wait for Microsoft to resolve it. - Reduce the scope of your prompt
Instead of asking a broad question like “Summarize all my emails from last month,” narrow it to “Summarize emails from John Smith this week.” A smaller context window reduces processing time. - Close unused Microsoft 365 apps and tabs
Open browser tabs and background apps consume system memory. Copilot in the browser or in a desktop app works better when at least 8 GB of RAM is free. Close unnecessary apps and browser tabs. - Sign out and sign back into Microsoft 365
Click your profile photo in the top-right corner of any Microsoft 365 app. Select Sign out. Wait 30 seconds, then sign in again. This refreshes your authentication token and often clears temporary session issues. - Use a different Microsoft 365 app
If Copilot is slow in Word, try the same prompt in Microsoft Teams or on copilot.microsoft.com. If the response is faster in another app, the problem is specific to the original app. Repair that app via Settings > Apps > Installed apps > Modify. - Clear browser cache and cookies
If you use Copilot in a browser, go to browser settings and clear cached images and cookies. Restart the browser and try again. Stale cache can interfere with real-time communication.
If Copilot Still Has Issues After the Main Fix
Copilot Returns “Something Went Wrong” After a Long Wait
This error usually indicates a timeout. The server could not complete the request within the allowed time. Reduce your prompt length and ensure your network is stable. If the error persists, submit a support ticket through the Microsoft 365 admin center.
Thinking Mode Stops Mid-Response
This happens when the connection drops while the model is generating output. Check your Wi-Fi signal strength. Move closer to the router or use a wired Ethernet connection. If the problem continues, run the Microsoft Support and Recovery Assistant from aka.ms/SaRA.
Copilot Is Slow Only in One Specific Microsoft 365 App
The app may have a corrupted add-in or a misconfigured plugin. Go to the app’s settings and disable third-party add-ins one by one. Restart the app after each disable. When the slow Thinking mode stops, re-enable add-ins except the one that caused the issue.
All Users in Your Organization Experience Long Thinking Times
This indicates a tenant-wide issue. Check the Microsoft 365 admin center for service advisories. If no advisory exists, verify that your Copilot licenses are correctly assigned. Go to Billing > Licenses and confirm each user has an active Copilot for Microsoft 365 license.
Copilot Pro vs Copilot for Microsoft 365: Performance Differences
| Item | Copilot Pro | Copilot for Microsoft 365 |
|---|---|---|
| Maximum context window | 128,000 tokens | 128,000 tokens |
| Priority during peak usage | Standard queue | Higher priority queue for business tenants |
| Integration with Microsoft Graph | Limited to personal Microsoft account data | Full access to tenant data, SharePoint, and Exchange |
| Rate limits per user per minute | 20 requests | 30 requests |
| Service-level agreement for uptime | 99.9% | 99.99% |
Copilot for Microsoft 365 offers higher request limits and better uptime guarantees. If your organization experiences frequent slow Thinking mode, upgrading from Copilot Pro to Copilot for Microsoft 365 may reduce delays. However, the most common cause of slowness is still network or prompt size, not the license tier.
You can now identify the specific bottleneck causing Copilot Thinking mode to take too long. Start by resetting the conversation and testing your network. If the delay persists, reduce your prompt scope and check Microsoft 365 service health. For persistent issues, use the Microsoft Support and Recovery Assistant to run automated diagnostics. Remember that keeping your prompt under 50,000 tokens and using a wired network connection yields the fastest response times.