Copilot Chat usage limits: what do the new caps mean for your team's AI training?

Short answer: From 9 November 2026, Microsoft is extending fair use limits to more Copilot Chat features for staff who use it without a Microsoft 365 Copilot licence, and it will not publish the thresholds. That makes training more important, because people who know what to ask, and when, will get far more out of limited capacity than people who learn by trial and error.

What Microsoft has announced

Copilot Chat comes with eligible Microsoft 365 business and enterprise subscriptions, so many UK organisations have given it to staff without buying the separate Microsoft 365 Copilot add-on. Microsoft has already used fair use limits for some features, such as image generation. Now it is extending that approach.

Microsoft's Message Center notice MC1536163, published on 3 November 2026 and shown here in an independent archive, says:

• The change affects users without a Microsoft 365 Copilot licence in commercial tenants.
• It applies to Copilot Chat in the web, desktop and mobile apps.
• New limits start rolling out on 9 November, and limits may expand to more capabilities over time.
• Limits can vary throughout the day. Users who reach one can try again later, and Microsoft says it will show the timeframe in the product.
• Responses already in progress are not interrupted.
• Microsoft sets and manages the limits, and admins cannot view or adjust them.

Microsoft's support article on Copilot Chat limits adds that it does not publish specific numeric limits because they can vary, and that hitting a limit is not an error or malfunction.

The Register and Computerworld both report that the rollout began on 2 November for organisations with 2,000 or fewer Microsoft 365 seats and starts on 9 November for larger ones. The Register also reports that, in most organisations, the limit notice includes an option for the user to request a Microsoft 365 Copilot licence from their admin.

Why this matters for AI training

The analysts Computerworld spoke to focused on one risk: unpredictability. Nhat Bui of Scopic Software told the publication that people only build AI into their daily work once they trust it will be there when they need it. Pareekh Jain of Pareekh Consulting warned that frustrated staff may turn to other AI tools outside the approved environment, which is the shadow AI problem many IT and risk teams already worry about.

That is a training question as much as a technology one. Three points stand out.

Good prompting uses less capacity. Someone who writes a clear brief with context, a format and a goal often gets a usable answer first time. Someone who does not may need several rounds of follow-up. Neither Microsoft nor anyone else has said how limits are counted, so we cannot promise a saving, but fewer wasted attempts is a sensible habit whatever the rules.

Staff need to know what to expect. If people do not know limits exist, a notice in the middle of a task looks like a fault. A short explanation in onboarding or a team briefing heads off help desk tickets and frustration.

Policy matters more when the approved tool is less predictable. If staff hit a limit, what are they allowed to do next? Your AI policy should answer that before they reach for an unapproved tool with company data.

Practical takeaways by role

L&D leaders: Put prompt quality at the centre of Copilot Chat training. Teach people to plan a request before they type it, give context up front, and ask for the output in the format they need. Use real tasks from each role rather than generic demos. Our Microsoft Copilot training is built around this kind of practical, role-based work.

HR leaders: Make sure your employee AI policy says which tools staff may use and what to do if the approved tool is unavailable. Explain the change in plain terms so nobody feels they have done something wrong when a limit appears.

IT leaders: Follow Microsoft's own recommendation and brief your help desk. Decide in advance how your organisation will handle licence requests raised from the limit notice, since Microsoft says admins can approve or deny them. Computerworld reports analyst advice to use internal usage data to see who relies on Copilot Chat most. That same data is useful for deciding where training will make the biggest difference.

Transformation leaders: Treat this as a prompt to look at how Copilot is really being used. Our Copilot rollout training plan sets out how to stage adoption, and our guide to measuring AI training ROI covers how to set a baseline before you decide where to invest.

A short checklist

1. Tell people. Let staff know that Copilot Chat may show a limit notice, and that it is not a fault.
2. Brief the help desk. Give them a short answer for "Copilot stopped working".
3. Update the policy. Say what staff should do if the approved tool is unavailable.
4. Train for better first answers. Focus on clear, well-structured requests for real tasks.
5. Watch usage. Look at who uses Copilot Chat most and where it is barely used, and plan training around both groups.

What this means for your team

Copilot Chat without the add-on is still a useful tool, but it is now one with limits you cannot see in advance. Teams that know how to get a good answer quickly, and know what to do when a feature is paused, will feel the change least. Teams that have been left to work it out alone are the ones most likely to give up on it, or drift to tools you have not approved.

If you are weighing Copilot against other assistants, our guide comparing Copilot, Claude and ChatGPT training may help. You can also read our earlier posts on Copilot Chat skills and the Copilot October 2026 update, or find out more about our corporate AI training and AI training in Leeds. We focus on readiness checks and training, and leave licensing decisions with your IT team.

Talk to us

If you would like help getting your teams confident with Copilot, book a free 60-minute discovery call with Qasim.