diff --git a/packages/integrations/src/google.ts b/packages/integrations/src/google.ts
index c94e84256..8ec74ca8b 100644
--- a/packages/integrations/src/google.ts
+++ b/packages/integrations/src/google.ts
@@ -149,27 +149,52 @@ function decodeHeader(value: string): string {
   // RFC 2047: ignore linear whitespace, including folding, between encoded words.
   return value
     .replace(/(\?=)(?:[ \t]|\r\n[ \t])+(?==\?)/g, "$1")
-    .replace(
-      /=\?([^?]+)\?([bq])\?([^?]*)\?=/gi,
-      (original, charset: string, encoding: string, text: string) => {
+    .replace(/(?:=\?[^?]+\?[bq]\?[^?]*\?=)+/gi, (run) => {
+      // §6.2: adjacent encoded words concatenate before decoding, so a header
+      // wrapped at the encoded-word length limit can split a multibyte
+      // character across words; join same-charset bytes before decoding.
+      const word = /=\?([^?]+)\?([bq])\?([^?]*)\?=/gi;
+      let decoded = "";
+      let pending: Buffer | null = null;
+      let pendingCharset = "";
+      let pendingOriginal = "";
+      const flush = () => {
+        if (pending === null) return;
         try {
-          const bytes =
-            encoding.toLowerCase() === "b"
-              ? Buffer.from(text, "base64")
-              : Buffer.from(
-                  text
-                    .replace(/_/g, " ")
-                    .replace(/=([0-9a-f]{2})/gi, (_, code: string) =>
-                      String.fromCharCode(Number.parseInt(code, 16)),
-                    ),
-                  "latin1",
-                );
-          return new TextDecoder(charset.split("*")[0]).decode(bytes);
+          decoded += new TextDecoder(pendingCharset).decode(pending);
         } catch {
-          return original;
+          decoded += pendingOriginal;
         }
-      },
-    );
+        pending = null;
+        pendingOriginal = "";
+      };
+      for (const match of run.matchAll(word)) {
+        const [original, charset, encoding, text] = match;
+        const normalized = charset.split("*")[0].toLowerCase();
+        const bytes =
+          encoding.toLowerCase() === "b"
+            ? Buffer.from(text, "base64")
+            : Buffer.from(
+                text
+                  .replace(/_/g, " ")
+                  .replace(/=([0-9a-f]{2})/gi, (_, code: string) =>
+                    String.fromCharCode(Number.parseInt(code, 16)),
+                  ),
+                "latin1",
+              );
+        if (pending !== null && normalized === pendingCharset) {
+          pending = Buffer.concat([pending, bytes]);
+          pendingOriginal += original;
+        } else {
+          flush();
+          pending = bytes;
+          pendingCharset = normalized;
+          pendingOriginal = original;
+        }
+      }
+      flush();
+      return decoded;
+    });
 }
 function decodeSnippet(value: string): string {
   const entities: Record<string, string> = {
diff --git a/tests/google.test.ts b/tests/google.test.ts
index c5e607b23..5b48a5e02 100644
--- a/tests/google.test.ts
+++ b/tests/google.test.ts
@@ -121,6 +121,35 @@ test("mail decodes folded adjacent encoded words without inserting header whites
   }
 });
 
+test("mail joins adjacent encoded words that split a multibyte character", async () => {
+  // RFC 2047 §6.2: adjacent encoded words are concatenated before decoding, so a
+  // sender that wraps a long header at the encoded-word length limit may break a
+  // multibyte character across two words. "Visit résumé" split inside each é.
+  const word = (bytes: number[]) => `=?UTF-8?B?${Buffer.from(bytes).toString("base64")}?=`;
+  const visitResume = [..."Visit résumé"].flatMap((char) => [...Buffer.from(char)]);
+  const splitAt = visitResume.indexOf(0xc3) + 1;
+  const subjects = [
+    `${word(visitResume.slice(0, splitAt))} ${word(visitResume.slice(splitAt))}`,
+    `${word(visitResume.slice(0, splitAt))}\r\n ${word(visitResume.slice(splitAt))}`,
+    "=?UTF-8?Q?Visit_r=C3?= =?UTF-8?Q?=A9sum=C3=A9?=",
+  ];
+  for (const value of subjects) {
+    const client = clientWith(() =>
+      json({
+        id: "thread1",
+        messages: [
+          {
+            id: "msg1",
+            threadId: "thread1",
+            payload: { headers: [{ name: "Subject", value }] },
+          },
+        ],
+      }),
+    );
+    assert.equal((await client.getThread("thread1"))[0].subject, "Visit résumé", value);
+  }
+});
+
 test("header decoding preserves spaces encoded inside words and next to plain text", async () => {
   const client = clientWith(() =>
     json({