To track AI-generated code across repositories, combine assistant usage telemetry, tool-reported code changes, pull-request activity, and explicit links between changes and agent sessions. These signals answer different questions: a person may use AI without accepting generated code, a tool may count lines without establishing authorship, and a pull request may show workflow activity without identifying which lines came from AI.
Decide what you need to measure
“AI contribution” can mean several different things. Choose the question first, then collect the signal that can actually answer it.
As an Amazon Associate I earn from qualifying purchases.
| Question | Useful signal | What it does not establish |
|---|---|---|
| Who uses an assistant, and how often? | Product usage telemetry, such as users and reported usage over a defined period | Whether suggestions were accepted, retained, or valuable |
| Which changes does the coding product attribute to AI? | Tool-reported code-generation measures, such as added or deleted lines and user- or agent-initiated activity | A complete record of all AI assistance or definitive authorship of every line |
| How is repository work moving? | Pull-request activity and established review or merge-flow measures | How much code in a PR was generated by AI |
| Which agent session produced a change? | Explicit provenance, such as a commit identity and link to a session log | Provenance for tools that do not provide such a trail |
Keep these categories separate in reports. Adoption, attributed output, repository throughput, and provenance can complement one another, but they are not interchangeable.
Free tools Windows power users keep installed
One-click scans. No signup required.
What GitHub Copilot reporting can show
For an organization using GitHub Copilot, GitHub documents usage-metrics dashboards and API or NDJSON reporting at enterprise, organization, repository, and user scopes. The available records differ by scope and purpose, so retain the scope and reporting date with every exported record rather than combining unlike totals. See GitHub Copilot usage metrics and the data available in those metrics.
#1 Best Overall
Usage telemetry: adoption, not authorship
Usage metrics help answer who is using Copilot and how usage changes over time. They do not, by themselves, tell you that generated code was accepted or identify the resulting changes. GitHub says most metrics depend on client-side IDE telemetry; settings and supported IDE or plugin versions affect coverage. Some measures may be unavailable without richer telemetry. Organization and enterprise totals can also differ because of deduplication and attribution timing, so do not compare them as though they were equivalent counts. Details are in GitHub’s usage-metrics documentation.
Code-generation measures: directional product attribution
GitHub’s code-generation dashboard distinguishes user-initiated from agent-initiated changes and reports lines added or deleted. GitHub characterizes its LoC measures as a directional measure of Copilot output across completions, chat, and agent features—not a universal ledger of AI-assisted code or a measure of value. An accepted suggestion, an added line, and a useful contribution are different things; interpret the reported unit exactly as documented. See Lines of Code metrics and the metric definitions.
Pull-request activity: repository workflow
GitHub’s repository-level pull-request reports count daily repository activity and can include pull requests created by Copilot cloud agent or reviewed by Copilot code review. They describe PR activity, not a count of all AI-generated lines. A repository with no activity on the requested day is omitted from that report; absence from the report should not be treated as proof that a repository has no AI use. Consult GitHub’s data-availability definitions.
Session provenance: a trace for a specific workflow
For Copilot cloud agent commits, GitHub documents a more direct provenance trail: Copilot is the author, the person who started the task is listed as co-author, and each commit message links to session logs. Retaining that commit and session link alongside the pull request can help reviewers inspect what happened during that agent run. This is evidence for the documented Copilot cloud-agent workflow, not a general attribution mechanism for every coding assistant. See Managing agent sessions.
Rank #3
Build a portfolio-wide reporting process
For multiple repositories, use a stable repository inventory as the backbone of reporting. Export the available product data or retrieve it through an API, then join records to that inventory while preserving the original scope and date. Avoid treating a repository PR report as though it were a line-level code ledger.
- Set the measurement question. Decide whether the report is about adoption, tool-attributed changes, PR flow, or session provenance. Keep separate measures separate in dashboards and decisions.
- Define the record fields. For each metric, record provider and product surface, reporting window, repository identifier, user or agent attribution, and the unit being counted—such as suggestions, accepted suggestions, added or deleted lines, PRs, or sessions.
- Preserve scope and missing-data context. Label whether a record is enterprise-, organization-, repository-, or user-level. Note relevant telemetry dependencies and whether a report omits repositories with no activity in its window.
- Retain explicit provenance where available. Keep agent identity, commit metadata, and session-log links with the associated PR or change. If a tool offers no equivalent trail, label attribution as unknown or tool-reported rather than inferring it from code style.
- Pair activity with outcomes carefully. If you are evaluating productivity or quality, compare AI activity with measures your team already uses, such as review and merge flow. A relationship between adoption and PR output does not by itself show that adoption caused a change.
GitHub’s impact dashboard relates adoption cohorts to pull-request output, but that relationship is not causal proof. The distinction matters whenever metrics are used to evaluate a team, a repository, or a policy change. See GitHub’s usage-metrics overview.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Can you identify AI-written code from the code itself?
Not reliably enough to replace explicit provenance. A style-based or behavioral classifier may be useful for research, but a classification is not a record of which person or agent produced a particular change. One 2026 study by Taher A. Ghaleb analyzed 33,580 pull requests from five agents and reported a 97.2% F1 score for identifying agents in that dataset. That is a result for the study’s data and method, not a guarantee of accuracy on another organization’s repositories or a verdict on any individual change. See Fingerprinting AI Coding Agents on GitHub.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →For an auditable record, use explicit tool-provided attribution and session evidence where available. If that evidence was never recorded, report the gap; code appearance alone cannot establish universal authorship.
What to compare when choosing tracking tools
- Coverage: Which assistants, agents, IDEs, and repository hosts are included?
- Scope: Can reporting be viewed consistently across users, repositories, organizations, or enterprises?
- Attribution detail: Does the data identify a user, an agent, a session, or only aggregate activity?
- Measured unit: Does it count suggestions, accepted suggestions, lines, commits, PRs, or outcomes?
- Telemetry and gaps: What client telemetry or supported IDE versions are required, and how are unavailable records represented?
- Data access and retention: Are exports or APIs available, and how long can the organization retain records under its applicable product terms?
- Reviewable provenance: Can a reviewer follow a change back to the tool session that produced it?
The documented detail here is strongest for GitHub Copilot. The cited sources do not establish a shared attribution format across GitLab, Bitbucket, Azure DevOps, and all coding-assistant vendors. Evaluate each product’s current data model rather than assuming that a metric or session trail transfers across platforms.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
