diff --git a/tests/tools/test_xlsx_phonetic_text.py b/tests/tools/test_xlsx_phonetic_text.py
index 8e926c3cd6..483bcb1d79 100644
--- a/tests/tools/test_xlsx_phonetic_text.py
+++ b/tests/tools/test_xlsx_phonetic_text.py
@@ -1,4 +1,4 @@
-"""XLSX phonetic guides annotate base text; they are not part of the cell value.
+"""Phonetic guides (XLSX ``rPh``, DOCX ``w:rt`` ruby) annotate base text; they are not the value.
 
 See https://learn.microsoft.com/en-us/dotnet/api/documentformat.openxml.spreadsheet.phoneticrun
 (rPh is permitted under both si and is).
@@ -13,6 +13,7 @@ from tools.read_extract import extract_document_text
 from tools.registry import registry
 
 S = "http://schemas.openxmlformats.org/spreadsheetml/2006/main"
+W = "http://schemas.openxmlformats.org/wordprocessingml/2006/main"
 R = "http://schemas.openxmlformats.org/officeDocument/2006/relationships"
 P = "http://schemas.openxmlformats.org/package/2006/relationships"
 
@@ -33,20 +34,39 @@ def _workbook(path, storage, text):
     return path
 
 
-@pytest.mark.parametrize("storage", ["shared", "inline"])
-@pytest.mark.parametrize("rich", [False, True])
-def test_read_file_excludes_phonetic_guides(tmp_path, monkeypatch, storage, rich):
-    base = '<r><rPr><b/></rPr><t>東</t></r><r><t>京</t></r>' if rich else '<t>東京</t>'
-    phonetics = '<rPh sb="0" eb="1"><t>トウ</t></rPh><rPh sb="1" eb="2"><t>キョウ</t></rPh>'
-    path = _workbook(tmp_path / "cities.xlsx", storage, base + phonetics)
+def _docx(path, paragraph_xml):
+    with zipfile.ZipFile(path, "w") as package:
+        package.writestr("[Content_Types].xml", "<Types/>")
+        package.writestr("word/document.xml",
+                         f'<w:document xmlns:w="{W}"><w:body><w:p>{paragraph_xml}</w:p></w:body></w:document>')
+    return path
+
+
+_XLSX_PHONETICS = '<rPh sb="0" eb="1"><t>トウ</t></rPh><rPh sb="1" eb="2"><t>キョウ</t></rPh>'
+_DOCX_RUBY = ('<w:r><w:ruby><w:rt><w:r><w:t>トウキョウ</w:t></w:r></w:rt>'
+              '<w:rubyBase><w:r><w:t>東京</w:t></w:r></w:rubyBase></w:ruby></w:r><w:r><w:tab/><w:t>sentinel</w:t></w:r>')
+
+
+def _document(tmp_path, kind):
+    if kind == "docx-ruby":
+        return _docx(tmp_path / "cities.docx", _DOCX_RUBY)
+    storage, rich = kind.split("-")
+    base = '<r><rPr><b/></rPr><t>東</t></r><r><t>京</t></r>' if rich == "rich" else '<t>東京</t>'
+    return _workbook(tmp_path / "cities.xlsx", storage, base + _XLSX_PHONETICS)
+
+
+@pytest.mark.parametrize("kind", ["shared-plain", "shared-rich", "inline-plain", "inline-rich", "docx-ruby"])
+def test_read_file_excludes_phonetic_guides(tmp_path, monkeypatch, kind):
+    path = _document(tmp_path, kind)
     monkeypatch.setenv("TERMINAL_ENV", "local")
     monkeypatch.setenv("TERMINAL_CWD", str(tmp_path))
-    task_id = f"xlsx-phonetic-{storage}-{rich}"
+    task_id = f"phonetic-{kind}"
     try:
         result = json.loads(registry.dispatch("read_file", {"path": str(path)}, task_id=task_id))
         assert not result.get("error"), result
         assert result["extracted_document"] is True
-        assert result["content"].splitlines()[1] == "2|東京\tsentinel"
+        row = result["content"].splitlines()[0 if kind == "docx-ruby" else 1]  # XLSX line 1 is the sheet header
+        assert row.split("|", 1)[1] == "東京\tsentinel"
     finally:
         file_tools.clear_file_ops_cache(task_id)
 
diff --git a/tools/read_extract.py b/tools/read_extract.py
index 9ec26aec8e..4fe9fbfe71 100644
--- a/tools/read_extract.py
+++ b/tools/read_extract.py
@@ -468,8 +468,11 @@ def _extract_docx(path: str) -> str:
     breaks = {f"{w}tab": "\t", f"{w}br": "\n", f"{w}cr": "\n"}
     lines: list[str] = []
     for para in root.iter(f"{w}p"):
+        # w:rt is the ruby (phonetic) guide over w:rubyBase; it annotates the text, it is not text.
+        guide = {n for rt in para.iter(f"{w}rt") for n in rt.iter()}
         text = "".join(
-            (n.text or "") if n.tag == f"{w}t" else breaks.get(n.tag, "") for n in para.iter())
+            (n.text or "") if n.tag == f"{w}t" else breaks.get(n.tag, "")
+            for n in para.iter() if n not in guide)
         lines.extend(text.split("\n"))
     return _joined(lines, "DOCX contains no extractable text")