Copilot Code Review: How to Use It in Your Pull Requests with the New Self-Resolving Flow

GitHub Copilot code review now resolves its own comments when a subsequent commit fixes them, and displays the progress of each pull request in a single summary. It also writes the commit message when you accept a batch of its suggestions. All three features became generally available (GA) on September 18, 2026.

There’s also a date to mark on your calendar: as of September 28, 2026, the “Default” setting for the review effort level changes from Lite to Balanced, the deepest and also the most expensive level. We explain what to do about it further down.

What is Copilot code review?

Copilot code review is the AI-powered code reviewer built into GitHub for pull requests. You add Copilot as a reviewer and it leaves inline comments with suggested changes that you can apply in a couple of clicks.

It’s available on all paid Copilot plans; Copilot Free doesn’t include it. It works on:

  • GitHub.com
  • GitHub CLI
  • GitHub Mobile
  • VS Code, Visual Studio, Xcode, and JetBrains IDEs
  • Azure DevOps (in public preview)

Copilot always sends a review of type Comment, never “Approve” or “Request changes”. By default, its review doesn’t count toward required approvals and doesn’t block the merge.

How to use Copilot code review on pull requests?

Open the pull request and, in the right sidebar, under Reviewers, click Request next to Copilot. The review usually arrives in less than 30 seconds.

From there, Copilot’s comments behave like those of a teammate: you can react, reply, resolve them, or hide them. You can also request a review via the REST API by adding copilot-pull-request-reviewer[bot] as a reviewer.

One detail matters for everything that follows: Copilot does not automatically re-review when you push new commits, unless you’ve configured it (we cover that in the automatic reviews section). If not, you request the new review yourself using the button that appears next to Copilot’s name in the Reviewers menu.

What changes in Copilot’s review summary?

The overview comment now shows Copilot’s current assessment of the PR, what effort level it used, and the complete list of findings, grouped into three categories:

  • Open: issues that haven’t been addressed yet. Those introduced by your latest commit carry the new label.
  • Resolved since last review: issues that Copilot confirmed you fixed.
  • Previously missed: issues in code you’d already pushed that Copilot only detected in a later pass. They appear in full in the summary because they’re not published anywhere else in the PR.

Each finding shows its severity and links to its inline comment. Also, each Copilot comment now starts with a brief title, so you can scan the list and decide what to look at first. When you push more commits and request a new review, the summary preserves the history instead of starting from scratch.

How does Copilot resolve its own comments?

When a subsequent commit addresses one of its comments, Copilot resolves that thread in its next review, and pending feedback stays open. This arrived first on September 11; the September 18 update refined it in two ways:

  • Copilot respects your replies. If you reply to a comment asking that the issue stay open, it stays open.
  • Copilot resolves with a reason. Based on your subsequent commits, a thread can close as Won't Fix or Incorrect, instead of disappearing without explanation.

The practical effect: after a couple of review rounds, the open threads in your PR are the ones that really need work. No more clicking “Resolve conversation” on a dozen obsolete comments before merging.

Does Copilot write the commit message when you accept its suggestions?

Yes. When you commit a complete and eligible batch of Copilot suggestions, it generates a commit title and optional description based on what the changes actually do. The batch can also include suggestions from human reviewers. Goodbye to generic “Apply suggestions from code review” in your history.

How much does Copilot code review cost: Lite or Balanced?

Use Lite for routine changes and Balanced for anything complex or risky; the cost difference is real.

  • Lite is GitHub’s standard review: quick and focused feedback on bugs, security vulnerabilities, and style.
  • Balanced sends the PR to a higher-reasoning model for deeper analysis. GitHub recommends it for:
    • security-sensitive code
    • changes that touch multiple services
    • repositories with strict quality standards

Both levels are available on Copilot Pro, Pro+, Max, Business, and Enterprise.

According to GitHub’s documentation at the time of publication, a Lite review typically consumes an estimated USD 0.05 to 1 in AI credits, and a Balanced review, USD 0.25 to 5. Neither range includes GitHub Actions minutes, and GitHub clarifies that consumption grows with PR size and custom instructions.

Lite also became more thorough. Since September 11, it combines findings from an ensemble of agents instead of a single one. According to GitHub’s own experiments, this increased by 47% the average number of addressed comments per review for high-severity findings, and reduced review cost by around 8%. These are GitHub’s internal figures, not independent benchmarks.

Is Copilot code review free?

Not on Copilot Free, which doesn’t include code review. There is a path for organization members without a Copilot license: on Business and Enterprise plans, an admin can enable it, and the credits that those reviews consume are billed to the organization as additional usage.

What changes on September 28?

As of September 28, 2026, any repository or organization with the effort level set to Default will receive Balanced reviews instead of Lite. If you prefer Lite, select it explicitly before that date; GitHub says it will respect that choice and won’t change it.

At the repository level:

  1. Go to Settings.
  2. Under “Code, planning, and automation”, click Copilot and then Code review.
  3. Next to Review effort level, choose Lite or Balanced.

At the organization level:

  1. Go to your organization’s Settings.
  2. Follow the same path: Copilot and then Code review.
  3. Set the Review effort level default value.Repositories that didn’t choose their own level inherit the organization’s. These default values apply to automatic reviews; in a manual request you choose the level in the Reviewers section of the PR, and that choice only applies to that review.

How do I make Copilot review all pull requests automatically?

You configure it with a branch ruleset:

  1. In the repository, go to Settings → Rulesets → New ruleset → New branch ruleset.
  2. Set it to Active and choose the target branches.
  3. In “Branch rules”, select Automatically request Copilot code review.
  4. Check Review new pushes if you want a review on each push.
  5. Optionally, check Review draft pull requests.

Organizations can do the same from Settings → Repository → Rulesets in the organization settings. And if you have Pro, Pro+, or Max, you can enable Automatic Copilot code review from Copilot settings in your profile.

Review new pushes is the option that ties everything together: auto-resolution happens on each new review, so with this option enabled, each push updates the marker without you having to ask for it. Keep in mind that each automatic review also consumes AI credits.

How do you tell Copilot what to review?

Copilot reads your repository’s instructions from three places:

  • .github/copilot-instructions.md for repository-wide rules
  • .github/instructions/**/*.instructions.md for rules that apply to specific paths
  • AGENTS.md in the repository root for conventions shared across AI tools

It reads these files from the PR’s head branch, so you can test a change to your instructions in the same pull request that introduces it. GitHub’s documentation includes an example instruction that’s a good starting point: “When performing a code review, respond in English.”

Devy’s Take

The GA updates aren’t so much a new capability as the end of a tedious task. Anyone who’s used an AI reviewer knows the pattern: comments pile up, half become outdated after the next push, and cleaning up before merge falls to you. A review that carries its own progress and closes its own threads is what actually saves time.

The September 28th change is what you need to sort out this week. Balanced is a better review for code that deserves it, but it shouldn’t become the default for all your PRs just because no one touched a setting. Choose your level intentionally, repository by repository.

If you’re rethinking where code written by agents gets reviewed, we cover it in The PR Is No Longer the Only Place Where an Agent’s Work Gets Reviewed. And if you’re still comparing tools, check out What’s the Best AI for Programming in 2026?.