diff --git a/extension/extract.js b/extension/extract.js
index 62399a2..067d42f 100644
--- a/extension/extract.js
+++ b/extension/extract.js
@@ -23,11 +23,28 @@
 
   // Albums are noisier free-text than artists, so only TWO precise signals generate album
   // candidates — the Spotify exact-album match (and/or LLM) is still the precision filter:
-  //   1. A quoted string ("Reek of Putrefaction" or 'Scum') — a strong "this is a title" cue.
-  //   2. An album cue phrase ("the album Scum", "Scum LP/EP") naming the next 1-4 words.
+  //   1. A quoted string ("Reek of Putrefaction" or 'Scum') — a strong "this is a title" cue,
+  //      but only when it reads like a title (short, no sentence-ending punctuation mid-span),
+  //      so prose quotes ("We hold these truths…", "lol") don't become candidates.
+  //   2. An album cue phrase ("the album Scum", "Scum LP/EP") naming a title-shaped run: the
+  //      capture starts at a capitalized/quoted token (so a cue followed by a verb/article —
+  //      "the album by Carcass", "the album is great" — yields no candidate) and the LP/EP
+  //      branch is anchored to 1-4 capitalized words instead of grabbing the sentence prefix.
   // Group 1 of each holds the candidate; the `d` flag gives its position for in-place wrapping.
   const QUOTED_RE = /["“]([^"”]{2,80})["”]|['‘]([^'’]{2,80})['’]/gd;
-  const ALBUM_CUE_RE = /\b(?:the )?(?:album|record|ep|lp)\s+(?:called\s+|titled\s+|named\s+)?([A-Za-z0-9][A-Za-z0-9'’.&\- ]{0,79}?)(?=[.,;:!?)\]"”'’]|\s+(?:is|was|by|and|on|from|which|that)\b|$)|\b([A-Za-z0-9][A-Za-z0-9'’.&\- ]{0,79}?)\s+(?:LP|EP)\b/gd;
+  // A title-shaped run for the LP/EP branch: 1-4 tokens starting at a capital/quote, mirroring
+  // CAND_RE's connector set so multi-word titles ("Reign in Blood") survive. Bounded ({0,3}).
+  const TITLE_RUN = `["“'‘]?${TOKEN}(?:[ ]+(?:${TOKEN}|of|the|and|&|in|on|to|for|a|an)){0,3}["”'’]?`;
+  const ALBUM_CUE_RE = new RegExp(
+    // "the album X" — capture must START with a capital or opening quote so a cue immediately
+    // followed by a verb/article ("the album is great", "the album by Carcass") captures nothing.
+    `\\b(?:the )?(?:album|record|ep|lp)\\s+(?:called\\s+|titled\\s+|named\\s+)?([A-Z"'“‘][A-Za-z0-9'’.&\\- ]{0,79}?)(?=[.,;:!?)\\]"”'’]|\\s+(?:is|was|by|and|on|from|which|that)\\b|$)` +
+      // "X LP/EP" — anchored title run (no left-anchor-less prefix grab).
+      `|\\b(${TITLE_RUN})\\s+(?:LP|EP)\\b`,
+    "gd"
+  );
+  // First word of an album-cue capture that, despite a leading capital, is not a title.
+  const ALBUM_LEAD_STOP = new Set(["is", "was", "by", "the", "a", "an"]);
 
   // Common capitalized words that are almost never the artist being recommended.
   // Deliberately small: the resolver's exact-name match is the real filter.
@@ -89,24 +106,43 @@
   // album match is what actually confirms it (this only over-generates the candidates).
   function albumMatchesIn(text) {
     const out = [];
+    const seen = new Set(); // dedupe identical (index,length) spans across the two signals
+    const push = (name, index, length) => {
+      const key = index + ":" + length;
+      if (seen.has(key)) return;
+      seen.add(key);
+      out.push({ name, index, length });
+    };
     QUOTED_RE.lastIndex = 0;
     let m;
     while ((m = QUOTED_RE.exec(text))) {
       const gi = m.indices[1] || m.indices[2]; // double- or single-quoted group
       if (!gi) continue;
       const name = text.slice(gi[0], gi[1]).trim();
+      const words = name.split(/\s+/);
       if (name.length < 2 || /^\d+$/.test(name)) continue;
+      // Reject prose quotes: a title doesn't carry sentence-ending punctuation mid-span and
+      // won't run long, and a lone interjection ("lol", "yeah") is a stopword, not a title.
+      if (/[.!?]/.test(name.slice(0, -1)) || words.length > 6) continue;
+      if (words.length === 1 && STOP.has(name[0].toUpperCase() + name.slice(1).toLowerCase())) continue;
       // index/length cover the inner text only (no quotes), so the wrapper links the title.
-      out.push({ name, index: gi[0], length: gi[1] - gi[0] });
+      push(name, gi[0], gi[1] - gi[0]);
     }
     ALBUM_CUE_RE.lastIndex = 0;
     while ((m = ALBUM_CUE_RE.exec(text))) {
       const gi = m.indices[1] || m.indices[2]; // "album X" group or "X LP/EP" group
       if (!gi) continue;
-      const raw = text.slice(gi[0], gi[1]);
+      // Trim a leading opening-quote the title run may have captured, shifting the index so
+      // the emitted span still maps back to the source text; then strip trailing punct/quotes.
+      let start = gi[0];
+      let raw = text.slice(start, gi[1]);
+      const lead = raw.match(/^["“'‘]+/);
+      if (lead) { start += lead[0].length; raw = raw.slice(lead[0].length); }
       const name = raw.replace(TRAILING_PUNCT, "").trim();
       if (name.length < 2 || /^\d+$/.test(name)) continue;
-      out.push({ name, index: gi[0], length: name.length });
+      // A capital-led capture that is actually a verb/article ("Is great", "By Carcass") is no title.
+      if (ALBUM_LEAD_STOP.has(name.split(/\s+/)[0].toLowerCase())) continue;
+      push(name, start, name.length);
     }
     return out;
   }
diff --git a/extension/extract.test.cjs b/extension/extract.test.cjs
index 66eb8b9..4dbf121 100644
--- a/extension/extract.test.cjs
+++ b/extension/extract.test.cjs
@@ -64,6 +64,43 @@ test("albumMatchesIn skips pure-number and too-short quotes", () => {
   assert.ok(!names.includes("X")); // length < 2
 });
 
+test("albumMatchesIn: 'X LP/EP' branch is anchored, not the sentence prefix", () => {
+  // No left anchor used to yield "I think their new" (the prefix before " LP").
+  const ms = albumMatchesIn("I think their new LP is amazing");
+  assert.ok(!ms.some((m) => m.name === "I think their new"));
+  // Either nothing, or only a title-shaped run — never a sentence fragment.
+  for (const m of ms) assert.ok(/^["“'‘]?[A-Z]/.test(m.name));
+});
+
+test("albumMatchesIn: 'the album X' guards a lowercase/verb left edge", () => {
+  // 'the album by Carcass' must not emit 'by Carcass' / 'by' as an album.
+  const a = albumMatchesIn("the album by Carcass").map((m) => m.name);
+  assert.ok(!a.includes("by Carcass"));
+  assert.ok(!a.includes("by"));
+  // 'the album is great…' must emit no album candidate at all.
+  const b = albumMatchesIn("the album is great but I forgot the name");
+  assert.equal(b.length, 0);
+});
+
+test("albumMatchesIn: prose quotes are not album candidates", () => {
+  // A quote with sentence-ending punctuation mid-span, or a lone interjection, is not a title.
+  const a = albumMatchesIn('"We hold these truths to be self-evident."').map((m) => m.name);
+  assert.ok(!a.includes("We hold these truths to be self-evident."));
+  assert.equal(albumMatchesIn('"lol"').length, 0);
+});
+
+test("albumMatchesIn: recall preserved — real titles still emit", () => {
+  assert.ok(albumMatchesIn("the album Reign in Blood").map((m) => m.name).includes("Reign in Blood"));
+  assert.ok(albumMatchesIn("Reign in Blood LP").map((m) => m.name).includes("Reign in Blood"));
+  assert.ok(albumMatchesIn('"Human Jerky"').map((m) => m.name).includes("Human Jerky"));
+});
+
+test("albumMatchesIn: tightened cues still map positions back to the text", () => {
+  for (const text of ["Reign in Blood LP rules", "the album by Carcass", '"Human Jerky" demo']) {
+    for (const m of albumMatchesIn(text)) assert.equal(text.substr(m.index, m.length), m.name);
+  }
+});
+
 test("isSafeHttpUrl accepts http/https and rejects javascript/data/garbage/empty", () => {
   // accepted: the only schemes a resolver link may inject as an href
   assert.ok(isSafeHttpUrl("https://open.spotify.com/artist/abc"));