Why does the New Copilot seem not as intelligent as the old Copilot

The new Copilot app is undergoing a unified-app transition with changes to navigation, models, account context, and feature availability, so verify the active account and ground every medical-compliance question in authoritative documents rather than relying on an unqualified chat answer.

Microsoft does not document that the new Copilot is categorically “less intelligent.” However, several documented changes can make it feel less capable or engaging:

• Personal, work, and school experiences are being consolidated into an updated app, while their data remains separate.
• Some features can be temporarily unavailable during rollout.
• Available models, advanced capabilities, and usage limits can vary by subscription.
• Copilot is probabilistic and can produce plausible but inaccurate or ungrounded answers. ^1^ ^2^

Check which experience is answering

1. Open Copilot and select the account/profile control.
2. Confirm whether it shows your personal account or your organization’s work account.
3. If both are present, switch deliberately to the appropriate account before starting the conversation.

Personal and work data are separated by design. A personal-account conversation cannot use the organizational context available to the work experience. Connecting the work account can provide context from permitted email, calendar, cloud storage, and Microsoft 365 content. ^1^

If the change appeared during an app update, a temporary functionality gap is also possible. Windows Insider release notes explicitly identify changing feature availability and temporary loss of capabilities during preview rollouts. ^3^

Make compliance answers source-bound

For medical auditing, do not ask only:

What are the compliance requirements for this workflow?

Instead, upload or reference the applicable policy, regulation, payer guidance, or approved internal procedure and constrain the answer:

Using only the attached 2026 policy and the cited regulatory source, identify the requirements applicable to this workflow. Quote the controlling language, give the page or section for every conclusion, distinguish requirements from recommendations, and state “not established by the supplied sources” when evidence is missing. Do not infer missing rules.

Then ask Copilot to produce:

• requirement;
• evidence quotation;
• source and section;
• workflow control;
• exception or ambiguity requiring human review.

This is grounding: supplying relevant evidence that anchors the response. Microsoft states that grounding can improve relevance, but generated content can still include information absent from the input sources; users should inspect the cited material and verify the output. ^4^

For a personal account, factual conversations may be grounded in web-search results, and Copilot should display links when web grounding is used. Open those links and confirm that they are current, authoritative, and actually support each conclusion. ^2^

Do not enter protected health information or other sensitive audit data unless your organization has approved that specific Copilot account and workflow.

Report the regression with a reproducible example

If the same prompt, account, and sources now produce materially worse answers:

1. In Copilot, select … in the upper-right corner.
2. Select Send feedback.
3. Select Report a problem.
4. Include when the change began, Windows and Copilot versions, account type, a sanitized example prompt, expected behavior, actual response, and a screenshot. ^5^

That distinguishes a product regression from differences caused by account context, rollout state, or insufficient source grounding.

References

1. Copilot app updates for personal, work, and school accounts
2. Transparency Note for Microsoft Copilot (for individuals)
3. Copilot on Windows release notes - Windows Insider Program
4. Application card: Microsoft Copilot (for organizations)
5. Troubleshoot your content in the updated Copilot app