The call
Start on request in a few repositories, tune .github/copilot-instructions.md until comments lead to changes, and enable automatic review only in the repositories that earn it.
Decide who pays for each review before you extend it to members without a license, and keep a person accountable for every approval.
Every Copilot code review, on GitHub.com or in an IDE, now costs AI credits plus GitHub Actions minutes, and an untuned reviewer posts comments people learn to ignore. Roll it out in an order that ties that spend to comments engineers act on.
When you need this
- Reviewers are asking whether to act on, triage, or ignore Copilot's comments.
- Contribution volume has outgrown your reviewers and you want a first pass before a person looks.
- Copilot code review is about to be enabled for members without a Copilot license.
- You are weighing Copilot code review against a dedicated review product.
Evaluation criteria
Start on request
Which repositories try Copilot review first, and who requests it?
Look forAsk reviewers to assign Copilot manually in a few repositories, which is the default behavior. After the pilot, organization owners can enable automatic review for selected repositories or all of them.
Tune the instructions until comments lead to changes
What should Copilot comment on, and what should it leave to CI and linters?
Look forWrite repository context and review priorities in .github/copilot-instructions.md. State what CI checks and when Copilot should stay silent. Add path-specific rules where needed. The goose maintainers reported about one real catch in five comments before they added a confidence threshold and a skip list.
Choose the effort level
Which repositories need longer analysis, and which need fast feedback?
Look forUse Balanced, the default, for security-sensitive code and changes across services. GitHub routes Balanced reviews to a higher-reasoning model and estimates US$0.25 to US$5 in AI credits per review. Use Lite for routine changes at US$0.05 to US$1. Organization owners set the automatic-review default, and repository admins can override it.
Decide what triggers a review
Should Copilot review once, on every push, or on drafts?
Look forEnable Review new pushes and Review draft pull requests only where the extra reviews justify the credits. Otherwise, Copilot reviews once, and reviewers can request another pass manually.
Set the billing source
Is each review billed to the author, the requester, or the organization?
Look forAutomatic reviews bill the pull request author by default. Manual reviews bill the requester. You can bill the organization instead. Enable Only allow Copilot code review to be triggered by authorized users to prevent outside licenses from triggering reviews.
Extend to members without a license on purpose
Do contributors without a Copilot seat get reviews, and who pays?
Look forEnable AI credits paid usage first. Then enable Allow members without a Copilot license to use Copilot code review in GitHub.com, which defaults to off. These reviews bill the organization as paid usage outside every user's budget.
Plan the runners
Where do the review's Actions minutes run?
Look forUse standard GitHub-hosted runners by default. Larger runners charge a higher per-minute rate. Self-hosted runners consume no Actions minutes. Disabling GitHub-hosted runners falls back to limited review without full project context.
Keep a person accountable for approval
Can Copilot's approval satisfy your required-approval rule?
Look forDecide whether Copilot approvals should satisfy required-approval rules. The feature is in public preview and off by default. Enable it in repository and organization settings, plus enterprise settings. An enabled approval counts like a teammate's and expires when new commits arrive.
Protect the review instructions
Can an author change the instructions Copilot uses on their own pull request?
Look forRequire code-owner review of instruction files and agent skills. Copilot reads both from the pull request's head branch.
Questions to ask vendors
- Which models review our pull requests? GitHub picks them automatically and keeps them private. Its code review page says only that reviews use "a carefully tuned mix of models, prompts, and system behaviors", which can include models your organization has not enabled. If your model policy matters for compliance, ask your GitHub account team to name the models in writing.
- What did code review cost us last month in AI credits and in Actions minutes, by repository?
- Do our content exclusion rules cover every path that should stay out of code review?
- How do Copilot approvals interact with our branch protection rules and CODEOWNERS?
Red flags
- Automatic review on every push across all repositories before anyone has tuned the instructions.
- Reviewers dismissing most Copilot comments while the instructions stay unchanged.
- Copilot approvals counting toward required approvals where a person must own the merge.
- Code review enabled for members without a license and no organization budget behind it.
Products in this piece
Sources
- Official documentation
- 7
- Public source
- 2
- GitHub Copilot docs: Copilot code review
- GitHub Copilot docs: Repository custom instructions
- GitHub Copilot docs: Models and pricing
- GitHub Copilot docs: Build an optimized review process with Copilot
- GitHub Copilot docs: Using Copilot code review across the pull request lifecycle
- GitHub Copilot docs: Content exclusion
- GitHub Copilot docs: Budgets for usage-based billing
- Angie Jones: How I taught GitHub Copilot code review to think like a maintainer (Nov 10, 2025)
- Terence Tham: 38 issues, a code review agent showdown (DEV Community, Feb 24, 2026)