diff --git a/.github/scripts/auto-label-pr/detectors.js b/.github/scripts/auto-label-pr/detectors.js
index 81bb77843d..4406370a27 100644
--- a/.github/scripts/auto-label-pr/detectors.js
+++ b/.github/scripts/auto-label-pr/detectors.js
@@ -147,19 +147,9 @@ async function detectCoreChanges(changedFiles) {
 }
 
 // Strategy: PR size detection
-async function detectPRSize(prFiles, totalAdditions, totalDeletions, totalChanges, isMegaPR, SMALL_PR_THRESHOLD, MEDIUM_PR_THRESHOLD, TOO_BIG_THRESHOLD) {
+async function detectPRSize(prFiles, totalAdditions, totalDeletions, isMegaPR, SMALL_PR_THRESHOLD, MEDIUM_PR_THRESHOLD, TOO_BIG_THRESHOLD) {
   const labels = new Set();
 
-  if (totalChanges <= SMALL_PR_THRESHOLD) {
-    labels.add('small-pr');
-    return labels;
-  }
-
-  if (totalChanges <= MEDIUM_PR_THRESHOLD) {
-    labels.add('medium-pr');
-    return labels;
-  }
-
   const testAdditions = prFiles
     .filter(file => file.filename.startsWith('tests/'))
     .reduce((sum, file) => sum + (file.additions || 0), 0);
@@ -167,7 +157,24 @@ async function detectPRSize(prFiles, totalAdditions, totalDeletions, totalChange
     .filter(file => file.filename.startsWith('tests/'))
     .reduce((sum, file) => sum + (file.deletions || 0), 0);
 
-  const nonTestChanges = (totalAdditions - testAdditions) - (totalDeletions - testDeletions);
+  const nonTestAdditions = totalAdditions - testAdditions;
+  const nonTestDeletions = totalDeletions - testDeletions;
+
+  // small/medium count churn (additions + deletions) so a balanced refactor isn't undersized.
+  const nonTestChurn = nonTestAdditions + nonTestDeletions;
+
+  if (nonTestChurn <= SMALL_PR_THRESHOLD) {
+    labels.add('small-pr');
+    return labels;
+  }
+
+  if (nonTestChurn <= MEDIUM_PR_THRESHOLD) {
+    labels.add('medium-pr');
+    return labels;
+  }
+
+  // too-big uses net line delta (additions - deletions), matching the review message in reviews.js.
+  const nonTestChanges = nonTestAdditions - nonTestDeletions;
 
   // Don't add too-big if mega-pr label is already present
   if (nonTestChanges > TOO_BIG_THRESHOLD && !isMegaPR) {
diff --git a/.github/scripts/auto-label-pr/index.js b/.github/scripts/auto-label-pr/index.js
index 9769cd8060..c8bdcfb2f3 100644
--- a/.github/scripts/auto-label-pr/index.js
+++ b/.github/scripts/auto-label-pr/index.js
@@ -123,7 +123,7 @@ module.exports = async ({ github, context }) => {
     detectNewComponents(github, context, prFiles),
     detectNewPlatforms(github, context, prFiles, apiData),
     detectCoreChanges(changedFiles),
-    detectPRSize(prFiles, totalAdditions, totalDeletions, totalChanges, isMegaPR, SMALL_PR_THRESHOLD, MEDIUM_PR_THRESHOLD, TOO_BIG_THRESHOLD),
+    detectPRSize(prFiles, totalAdditions, totalDeletions, isMegaPR, SMALL_PR_THRESHOLD, MEDIUM_PR_THRESHOLD, TOO_BIG_THRESHOLD),
     detectDashboardChanges(changedFiles),
     detectGitHubActionsChanges(changedFiles),
     detectCodeOwner(github, context, changedFiles),
diff --git a/.github/scripts/auto-label-pr/tests/detectors.test.js b/.github/scripts/auto-label-pr/tests/detectors.test.js
index 02d69ca95e..aab1827c44 100644
--- a/.github/scripts/auto-label-pr/tests/detectors.test.js
+++ b/.github/scripts/auto-label-pr/tests/detectors.test.js
@@ -1,6 +1,6 @@
 const { describe, it } = require('node:test');
 const assert = require('node:assert/strict');
-const { detectNewPlatforms, detectNewComponents } = require('../detectors');
+const { detectNewPlatforms, detectNewComponents, detectPRSize } = require('../detectors');
 
 // Minimal GitHub API mock — only repos.getContent is called by detectNewPlatforms/detectNewComponents
 // to check for CONFIG_SCHEMA in newly added files.
@@ -145,3 +145,79 @@ describe('detectNewComponents', () => {
     assert.equal(result.labels.size, 0);
   });
 });
+
+// ---------------------------------------------------------------------------
+// detectPRSize
+// ---------------------------------------------------------------------------
+
+describe('detectPRSize', () => {
+  const SMALL = 30;
+  const MEDIUM = 100;
+  const TOO_BIG = 1000;
+
+  function size(prFiles, isMegaPR = false) {
+    const totalAdditions = prFiles.reduce((sum, file) => sum + (file.additions || 0), 0);
+    const totalDeletions = prFiles.reduce((sum, file) => sum + (file.deletions || 0), 0);
+    return detectPRSize(prFiles, totalAdditions, totalDeletions, isMegaPR, SMALL, MEDIUM, TOO_BIG);
+  }
+
+  it('counts only non-test changes toward small-pr', async () => {
+    // 10 source + 5000 test lines -> non-test churn of 10 is still small.
+    const labels = await size([
+      { filename: 'esphome/components/foo/foo.cpp', additions: 10, deletions: 0 },
+      { filename: 'tests/components/foo/test.esp32-idf.yaml', additions: 5000, deletions: 0 },
+    ]);
+    assert.ok(labels.has('small-pr'));
+    assert.equal(labels.size, 1);
+  });
+
+  it('counts additions and deletions as churn (not net delta)', async () => {
+    // A balanced refactor (40 added, 40 removed) is 80 lines of churn -> medium, not small.
+    const labels = await size([
+      { filename: 'esphome/components/foo/foo.cpp', additions: 40, deletions: 40 },
+    ]);
+    assert.ok(labels.has('medium-pr'));
+    assert.equal(labels.size, 1);
+  });
+
+  it('labels medium-pr when non-test changes exceed small threshold', async () => {
+    const labels = await size([
+      { filename: 'esphome/components/foo/foo.cpp', additions: 60, deletions: 0 },
+      { filename: 'tests/components/foo/test.esp32-idf.yaml', additions: 5000, deletions: 0 },
+    ]);
+    assert.ok(labels.has('medium-pr'));
+    assert.equal(labels.size, 1);
+  });
+
+  it('uses net delta (not churn) for too-big', async () => {
+    // 600 added + 600 removed: 1200 churn (above too-big) but 0 net delta -> not too-big.
+    const labels = await size([
+      { filename: 'esphome/components/foo/foo.cpp', additions: 600, deletions: 600 },
+    ]);
+    assert.equal(labels.size, 0);
+  });
+
+  it('labels too-big when non-test changes exceed the big threshold', async () => {
+    const labels = await size([
+      { filename: 'esphome/components/foo/foo.cpp', additions: 2000, deletions: 0 },
+      { filename: 'tests/components/foo/test.esp32-idf.yaml', additions: 5000, deletions: 0 },
+    ]);
+    assert.ok(labels.has('too-big'));
+    assert.equal(labels.size, 1);
+  });
+
+  it('does not label too-big when mega-pr is set', async () => {
+    const labels = await size([
+      { filename: 'esphome/components/foo/foo.cpp', additions: 2000, deletions: 0 },
+    ], true);
+    assert.equal(labels.size, 0);
+  });
+
+  it('produces no size label for a large mega-pr in the gap above medium', async () => {
+    // Non-test changes land between MEDIUM and TOO_BIG: not small/medium, and mega-pr suppresses too-big.
+    const labels = await size([
+      { filename: 'esphome/components/foo/foo.cpp', additions: 500, deletions: 0 },
+    ], true);
+    assert.equal(labels.size, 0);
+  });
+});