GitHub announced on 5 November 2026 that the Copilot usage metrics API now includes a copilot_code_review object on each repository-level daily row, for both enterprise and organisation reports.
What the changelog says
- Comments evaluated and comments addressed are now counted. Addressed means the developer acted on a comment in any way: applying a suggestion, asking Copilot for the fix, editing the code, or replying that the fix will come later. Divide one by the other and you have an addressed rate.
- Counts are split by severity (high, medium, low), review tier (lite, balanced, max) and comment type (bug, security, or custom rules from your own guidelines).
- A separate count shows merged pull requests reviewed by Copilot where no person requested changes afterwards.
- Comments are counted on the day the pull request merges, and Copilot usually does not evaluate comments on pull requests opened by bots.
- Access is limited to enterprise and organisation owners, billing managers, and roles with the View Copilot Metrics permission. The usage metrics policy must be enabled.
The interpretation
For a year, the question buyers asked about AI-assisted development was “how much code does the AI write?” This release measures the other half: how much of the AI’s review feedback anyone acts on. An addressed rate near zero means a reviewer nobody listens to. A very high rate with no human change requests could mean careful teams, or rubber-stamping. The number alone does not tell you which.
It also shows that review is becoming a measured, billed, tiered product, not an invisible background feature.
So what
If you commission software from a supplier that uses AI review, ask what they track and who reads it. Metrics only help when a named human owns the response. BuildApps uses AI-assisted workflows with human review on every change. See AI-assisted development, custom software development, or start a project.