Sort every AI review comment into three levels: nit, suggestion or blocker. Fix blockers, consider suggestions, and ignore or batch nits. Without triage, bot comments pile up, authors learn to dismiss all of them, and the one real bug gets dismissed with the rest.
Borrow the labels
Google's public review guidance tells reviewers to mark non-essential points with "Nit:" so authors know they can be ignored, and to label educational comments that are not required the same way. It also says reviewers should not block a change over personal preference and should favor technical facts over opinion.
Those rules work for bots. Ask the tool to prefix its comments, or sort them yourself.
The three levels
Blocker. A defect, a security issue, a broken edge case or a missing test for risky logic. The author must fix it or explain why it is wrong.
Suggestion. A better approach, a simplification or a clearer name. Worth a thought, and the author decides.
Nit. Style, formatting or wording. Do not spend review rounds on it. If a style rule matters, put it in the linter so it is never a comment.
Questions that sort a comment
- Could this cause wrong behavior or harm? Blocker.
- Does it make the code clearly easier to understand? Suggestion.
- Is it a matter of taste or already covered by a style guide? Nit.
- Is it factually wrong? Dismiss it and record it as a wrong comment, so you can track the tool's accuracy.
Check the claim before acting
Vendor documentation is candid on this point. GitHub says Copilot will make mistakes and advises validating its feedback. A comment that sounds confident can be wrong. Verify against the code before changing anything.
Reduce volume at the source
- Move mechanical rules into linters and formatters.
- Configure the bot to skip files you do not want reviewed, if it allows.
- Keep pull requests small so there is less to comment on.
- Write short review instructions in the repository if your tool supports them, listing what matters and what to ignore.
Track the ratio
Over a few weeks, count blockers, suggestions, nits and wrong comments. If wrong comments and nits dominate, tune the tool or reduce its scope. If blockers are rare but real, keep it.
An example in practice
A bot comments that a variable name is vague, that a loop could be a comprehension and that a query is built by string concatenation. The first is a nit, the second a suggestion, and the third a blocker, because it can lead to injection. Same pull request, three levels, and only one of them should hold up the merge.
Quick check
Should a nit block a merge?
No. Google's guidance says to label minor points so the author knows they can be ignored, and not to block over personal preference.
What do you do with a wrong bot comment?
Dismiss it and log it as wrong, so you can measure the tool's accuracy over time.
Sources
Last reviewed: 19 October 2026.
