diff --git a/backend/src/main/java/com/storycove/entity/Story.java b/backend/src/main/java/com/storycove/entity/Story.java
index dde96e5..e87d129 100644
--- a/backend/src/main/java/com/storycove/entity/Story.java
+++ b/backend/src/main/java/com/storycove/entity/Story.java
@@ -8,6 +8,7 @@ import org.hibernate.annotations.UpdateTimestamp;
 import org.jsoup.Jsoup;
 import com.fasterxml.jackson.annotation.JsonManagedReference;
 import com.fasterxml.jackson.annotation.JsonBackReference;
+import com.storycove.util.TextSanitizer;
 
 import java.time.LocalDateTime;
 import java.util.HashSet;
@@ -171,8 +172,8 @@ public class Story {
     }
     
     public void setContentHtml(String contentHtml) {
-        this.contentHtml = contentHtml;
-        this.setContentPlain(Jsoup.parse(contentHtml).text());
+        this.contentHtml = TextSanitizer.sanitize(contentHtml);
+        this.setContentPlain(Jsoup.parse(this.contentHtml).text());
         updateWordCount();
     }
     
diff --git a/backend/src/main/java/com/storycove/util/TextSanitizer.java b/backend/src/main/java/com/storycove/util/TextSanitizer.java
new file mode 100644
index 0000000..2c62e1a
--- /dev/null
+++ b/backend/src/main/java/com/storycove/util/TextSanitizer.java
@@ -0,0 +1,105 @@
+package com.storycove.util;
+
+import org.jsoup.Jsoup;
+import org.jsoup.nodes.Document;
+import org.jsoup.nodes.TextNode;
+
+/**
+ * Normalizes the text nodes of story HTML so that invisible Unicode cruft (commonly left behind by
+ * EPUB/Calibre conversions) can never reach persisted story content. Only text node content is touched;
+ * tag structure and attributes pass through unchanged.
+ */
+public final class TextSanitizer {
+
+    private static final int MIN_COLLAPSIBLE_RUN = 3;
+
+    private TextSanitizer() {
+    }
+
+    /**
+     * Strips Unicode "Cf" (format) characters and collapses runs of 3+ whitespace-like characters
+     * down to a single space, within text nodes only.
+     */
+    public static String sanitize(String html) {
+        if (html == null || html.trim().isEmpty()) {
+            return html;
+        }
+
+        Document doc = Jsoup.parseBodyFragment(html);
+        doc.outputSettings().prettyPrint(false);
+
+        doc.body().traverse((node, depth) -> {
+            // Skip text nodes that are pure structural whitespace (indentation/formatting
+            // between tags, e.g. " ", "\n  "): isBlank() only matches ' ', '\t', '\n', '\r',
+            // '\f', so any node containing real prose, Cf chars, or NBSP spam still gets processed.
+            if (node instanceof TextNode textNode && !textNode.isBlank()) {
+                String original = textNode.getWholeText();
+                String normalized = normalizeText(original);
+                if (!normalized.equals(original)) {
+                    textNode.text(normalized);
+                }
+            }
+        });
+
+        return doc.body().html();
+    }
+
+    static String normalizeText(String text) {
+        String withoutFormatChars = stripFormatCharacters(text);
+        return collapseLongWhitespaceRuns(withoutFormatChars);
+    }
+
+    private static String stripFormatCharacters(String text) {
+        StringBuilder result = new StringBuilder(text.length());
+        int i = 0;
+        while (i < text.length()) {
+            int codePoint = text.codePointAt(i);
+            int charCount = Character.charCount(codePoint);
+            if (Character.getType(codePoint) != Character.FORMAT) {
+                result.appendCodePoint(codePoint);
+            }
+            i += charCount;
+        }
+        return result.toString();
+    }
+
+    private static String collapseLongWhitespaceRuns(String text) {
+        StringBuilder result = new StringBuilder(text.length());
+        int len = text.length();
+        int i = 0;
+        while (i < len) {
+            int codePoint = text.codePointAt(i);
+            if (isSpaceLike(codePoint)) {
+                int runStart = i;
+                int runLength = 0;
+                int j = i;
+                while (j < len) {
+                    int cp = text.codePointAt(j);
+                    if (!isSpaceLike(cp)) {
+                        break;
+                    }
+                    runLength++;
+                    j += Character.charCount(cp);
+                }
+                if (runLength >= MIN_COLLAPSIBLE_RUN) {
+                    result.append(' ');
+                } else {
+                    result.append(text, runStart, j);
+                }
+                i = j;
+            } else {
+                result.appendCodePoint(codePoint);
+                i += Character.charCount(codePoint);
+            }
+        }
+        return result.toString();
+    }
+
+    /**
+     * Mirrors Python's str.isspace(): Unicode space separators (Zs/Zl/Zp, which includes NBSP)
+     * plus the ASCII/control whitespace characters (tab, newline, etc.).
+     */
+    private static boolean isSpaceLike(int codePoint) {
+        return Character.isWhitespace(codePoint) || Character.isSpaceChar(codePoint);
+    }
+}
diff --git a/backend/src/test/java/com/storycove/util/TextSanitizerTest.java b/backend/src/test/java/com/storycove/util/TextSanitizerTest.java
new file mode 100644
index 0000000..44abe45
--- /dev/null
+++ b/backend/src/test/java/com/storycove/util/TextSanitizerTest.java
@@ -0,0 +1,138 @@
+package com.storycove.util;
+
+import org.junit.jupiter.api.DisplayName;
+import org.junit.jupiter.api.Test;
+
+import static org.junit.jupiter.api.Assertions.*;
+
+@DisplayName("TextSanitizer Tests")
+class TextSanitizerTest {
+
+    private static final String ZWSP = "​";   // zero width space (Cf)
+    private static final String WORD_JOINER = "⁠"; // Cf
+    private static final String BOM = "";    // zero width no-break space / BOM (Cf)
+    private static final String NBSP = " ";   // non-breaking space (Zs, not Cf)
+
+    @Test
+    @DisplayName("Should strip a large run of zero-width spaces embedded mid-sentence")
+    void shouldStripLargeRunOfZeroWidthSpacesMidSentence() {
+        String garbageRun = ZWSP.repeat(5000);
+        String html = "<p>The hero drew" + garbageRun + "his sword and charged.</p>";
+
+        String sanitized = TextSanitizer.sanitize(html);
+
+        assertFalse(sanitized.contains(ZWSP), "Zero-width spaces should be fully stripped");
+        assertEquals("<p>The hero drewhis sword and charged.</p>", sanitized);
+    }
+
+    @Test
+    @DisplayName("Should strip mixed Cf characters (word joiner, BOM) anywhere in text")
+    void shouldStripMixedFormatCharacters() {
+        String html = "<p>Hello" + BOM + " " + WORD_JOINER + "world.</p>";
+
+        String sanitized = TextSanitizer.sanitize(html);
+
+        assertFalse(sanitized.contains(BOM));
+        assertFalse(sanitized.contains(WORD_JOINER));
+        assertEquals("<p>Hello world.</p>", sanitized);
+    }
+
+    @Test
+    @DisplayName("Should collapse a long run of regular spaces used as a formatting hack")
+    void shouldCollapseLongRunOfRegularSpaces() {
+        String html = "<p>He paused" + " ".repeat(200) + "then spoke.</p>";
+
+        String sanitized = TextSanitizer.sanitize(html);
+
+        assertEquals("<p>He paused then spoke.</p>", sanitized);
+    }
+
+    @Test
+    @DisplayName("Should collapse a long run of NBSP characters down to a single space")
+    void shouldCollapseLongRunOfNbsp() {
+        String html = "<p>Wait" + NBSP.repeat(500) + "for it.</p>";
+
+        String sanitized = TextSanitizer.sanitize(html);
+
+        assertEquals("<p>Wait for it.</p>", sanitized);
+    }
+
+    @Test
+    @DisplayName("Should collapse a mixed run of tabs, NBSP, and regular spaces")
+    void shouldCollapseMixedWhitespaceRun() {
+        String html = "<p>Stop" + " \t" + NBSP + " \t" + NBSP + "go.</p>";
+
+        String sanitized = TextSanitizer.sanitize(html);
+
+        assertEquals("<p>Stop go.</p>", sanitized);
+    }
+
+    @Test
+    @DisplayName("Should leave a single space untouched")
+    void shouldLeaveSingleSpaceUntouched() {
+        String html = "<p>One two three.</p>";
+
+        assertEquals(html, TextSanitizer.sanitize(html));
+    }
+
+    @Test
+    @DisplayName("Should leave a run of exactly two whitespace characters untouched")
+    void shouldLeaveTwoWhitespaceCharsUntouched() {
+        String html = "<p>One  two.</p>"; // two regular spaces
+
+        assertEquals(html, TextSanitizer.sanitize(html));
+    }
+
+    @Test
+    @DisplayName("Should not mangle accented / non-ASCII characters")
+    void shouldPreserveAccentedCharacters() {
+        String html = "<p>Café, naïve, über, 日本語, 😀 stays intact.</p>";
+
+        assertEquals(html, TextSanitizer.sanitize(html));
+    }
+
+    @Test
+    @DisplayName("Should preserve paragraph structure expressed via a double newline")
+    void shouldPreserveDoubleNewlineParagraphBreaks() {
+        String html = "<p>First paragraph.\n\nSecond paragraph.</p>";
+
+        assertEquals(html, TextSanitizer.sanitize(html));
+    }
+
+    @Test
+    @DisplayName("Should preserve paragraph structure expressed via <p> tags")
+    void shouldPreserveParagraphTags() {
+        String html = "<p>First paragraph.</p><p>Second paragraph.</p>";
+
+        assertEquals(html, TextSanitizer.sanitize(html));
+    }
+
+    @Test
+    @DisplayName("Should not alter whitespace used purely for HTML indentation between tags")
+    void shouldNotCollapseInterTagIndentation() {
+        // The "\n  " nodes between <div>/<p> tags are pure structural indentation, not prose,
+        // and must be left completely untouched even though they are 3+ char whitespace runs.
+        String html = "<div>\n  <p>Once upon a time.</p>\n  <p>The end.</p>\n</div>";
+
+        String sanitized = TextSanitizer.sanitize(html);
+
+        assertEquals(html, sanitized);
+    }
+
+    @Test
+    @DisplayName("Control case: realistic story content is unaffected end-to-end")
+    void controlCaseRealisticStoryIsUnaffected() {
+        String html = "<p>The old café on the corner of Rue de Bellévue had seen better days, "
+                + "but Anaïs still loved it.</p>"
+                + "<p>\"We'll meet at dawn,\" she whispered, \"just like we always do.\"</p>";
+
+        assertEquals(html, TextSanitizer.sanitize(html));
+    }
+
+    @Test
+    @DisplayName("Should handle null and blank input gracefully")
+    void shouldHandleNullAndBlankInput() {
+        assertNull(TextSanitizer.sanitize(null));
+        assertEquals("", TextSanitizer.sanitize(""));
+    }
+}