diff --git a/src/ai_typewriter/math_format.py b/src/ai_typewriter/math_format.py
index f02ef4b..797079d 100644
--- a/src/ai_typewriter/math_format.py
+++ b/src/ai_typewriter/math_format.py
@@ -1,7 +1,6 @@
 from __future__ import annotations
 
 import re
-from dataclasses import dataclass
 
 GREEK_AND_SYMBOLS = {
     r"\alpha": "α",
@@ -53,14 +52,6 @@ SUBSCRIPT = str.maketrans({
 })
 
 _SCRIPT_PATTERN = re.compile(r"([_^])\(([^()]+)\)|([_^])\{([^{}]+)\}|([_^])([A-Za-z0-9+\-=])")
-EQUATION_OPEN = "[EQ]"
-EQUATION_CLOSE = "[/EQ]"
-
-
-@dataclass(frozen=True)
-class MathSegment:
-    kind: str
-    value: str
 
 
 def _translate_script(value: str, marker: str) -> str:
@@ -76,11 +67,15 @@ def _replace_script(match: re.Match[str]) -> str:
 
 
 def format_math_text(text: str, mode: str = "plain") -> str:
-    """Optional legacy text-only math formatting for non-Word targets."""
+    """Format the AI answer before key stepping.
+
+    plain keeps the response unchanged, which is the recommended mode for LaTeX
+    markers such as [EQ]\\frac{a}{b}[/EQ].
+    """
     if mode == "plain":
         return text
     if mode not in {"unicode", "unicode_math"}:
-        raise ValueError("math_text_format doit être 'plain', 'unicode' ou 'word_equation'")
+        raise ValueError("math_text_format doit être 'plain' ou 'unicode'")
 
     formatted = text
     for command, replacement in sorted(GREEK_AND_SYMBOLS.items(), key=lambda item: len(item[0]), reverse=True):
@@ -91,43 +86,3 @@ def format_math_text(text: str, mode: str = "plain") -> str:
         previous = formatted
         formatted = _SCRIPT_PATTERN.sub(_replace_script, formatted)
     return formatted
-
-
-def split_word_equation_segments(text: str) -> list[MathSegment]:
-    """Split explicit [EQ]...[/EQ] blocks into Word equation actions.
-
-    No automatic math guessing is done here: the model decides where an equation
-    starts and ends by returning [EQ] before the expression and [/EQ] after it.
-    """
-    segments: list[MathSegment] = []
-    cursor = 0
-
-    while cursor < len(text):
-        start = text.find(EQUATION_OPEN, cursor)
-        if start == -1:
-            if cursor < len(text):
-                segments.append(MathSegment("text", text[cursor:]))
-            break
-
-        if start > cursor:
-            segments.append(MathSegment("text", text[cursor:start]))
-
-        expression_start = start + len(EQUATION_OPEN)
-        end = text.find(EQUATION_CLOSE, expression_start)
-        if end == -1:
-            expression = text[expression_start:].strip()
-            cursor = len(text)
-        else:
-            expression = text[expression_start:end].strip()
-            cursor = end + len(EQUATION_CLOSE)
-
-        if expression:
-            segments.append(MathSegment("equation", expression))
-
-    merged: list[MathSegment] = []
-    for segment in segments:
-        if merged and segment.kind == "text" and merged[-1].kind == "text":
-            merged[-1] = MathSegment("text", merged[-1].value + segment.value)
-        else:
-            merged.append(segment)
-    return merged