Skip to main content

How LinearB Calculates AI Attribution

Learn how LinearB identifies AI-assisted development across your codebase. This article explains how AI attribution is calculated using commit signals and AI tool data, how confidence thresholds work, and how results are applied to commits and pull requests.

Steven Silverstone
Updated by Steven Silverstone

Overview

LinearB identifies and measures AI-assisted development by combining explicit Git signals with AI tool usage signals that are correlated with coding activity in Git.

This article explains, at a high level, how LinearB determines whether code changes show signs of AI assistance and how that attribution is reflected in AI Analytics and related reporting.

AI attribution is designed to balance accuracy and coverage. Some signals provide direct evidence of AI usage, while others are inferred from correlated AI tool activity and Git activity.

AI Attribution Methods

LinearB uses two primary signal types to identify AI-assisted development:

1. Git-Based Attribution

Some AI tools add metadata to commits when developers create code using the tool.

  • How it works: The commit includes a Co-authored-by signature or similar Git metadata added by the AI tool.
  • What this means: When this metadata is present, LinearB can identify the commit as AI-assisted based on the Git signal.
  • Limitations: Not all tools or developer workflows add this metadata automatically.

2. AI Tool Usage Signals

When Git metadata is not available, LinearB uses heuristic signals from connected AI tools.

  • How it works: LinearB correlates AI tool usage API data with coding activity in Git. For each developer and each day, LinearB calculates the ratio of AI-accepted lines to the total lines that developer committed to Git on that same day.
  • Data sources: Supported AI tools such as Claude, GitHub Copilot, Cursor, and Amazon Q.
  • Daily granularity: AI tool providers report usage per user per calendar day (00:00–24:00 UTC), so LinearB evaluates AI tool usage per developer, per day.
  • What this means: LinearB can identify patterns that indicate AI-assisted development even when explicit Git metadata is not present.
  • Timing: AI tool usage data may be processed after the coding activity occurs.
Heuristic attribution is based on correlation between AI tool usage and Git activity. It does not represent a direct measurement of every AI-generated line of code.

Commit-Level Attribution

LinearB evaluates commits using the available AI attribution signals.

  • Git-based attribution is used when explicit AI-related Git metadata is present, such as a Co-authored-by trailer. These commits are always classified as AI-assisted.
  • Heuristic attribution is used when AI tool usage signals correlate with Git activity:
    • By default, if the ratio of AI-accepted lines to committed lines for a developer on a given day is 25% or higher (or the custom Commit AI-Assistance Threshold configured for your organization), all commits attributed to that developer on that day are classified as AI-assisted.
    • If the ratio is below 25% (or the configured custom threshold), all commits from that developer on that day are classified as not AI-assisted, except commits that contain explicit AI-related Git metadata.

The available signals determine whether commit activity is classified as AI-assisted in LinearB reporting.

Because AI tool providers report usage at the day level, heuristic attribution is a day-level approximation. LinearB does not match individual accepted lines to individual commits.

To change the default threshold, see Configuring AI Classification Thresholds.

Pull Request Attribution

LinearB reflects AI attribution at the pull request level because a pull request represents a complete unit of work.

When LinearB detects AI-assisted activity associated with a pull request, that pull request may be classified as AI-assisted in AI Analytics and related reporting.

Data Requirements

  • Git data must be properly synced in LinearB.
  • AI tool integrations must be connected for AI tool usage signals to be available.
  • AI tool users must be mapped to LinearB users so activity can be associated with the correct contributors.
If user mapping is incomplete, AI attribution may be partially missing from user-level or team-level reporting.

Limitations

  • Not all AI tools add Git metadata to commits.
  • AI tool usage signals depend on the data available from each connected AI provider.
  • AI usage data may be delayed depending on the provider and integration path.
  • Code copied manually from an AI chat interface may not be detected unless it is represented in a supported signal.
  • Different AI tools provide different levels of signal coverage.

Where AI Attribution Appears

  • AI Analytics
  • AI Insights
  • AI Adoption
  • AI-related views and filters across LinearB reporting, where supported

FAQ

Is AI attribution always exact?
Some signals, such as explicit Git metadata, provide direct evidence of AI usage. Other signals are heuristic and depend on correlation between AI tool usage and Git activity.

Why is AI usage missing for some developers?
This may occur if AI tools are not connected, usage data is delayed, or AI tool users are not mapped to LinearB users.

Does LinearB detect every AI interaction?
No. LinearB relies on supported Git and AI tool signals. Some workflows, such as manually copying code from an AI chat interface, may not be detected.

Why can AI attribution differ between tools?
Different AI tools expose different types of usage data and metadata, so the available attribution signals may vary by provider.

How did we do?

Configuring AI Classification Thresholds

Improving AI Attribution Accuracy and Coverage

Contact