diff --git a/coderd/x/chatd/quickgen.go b/coderd/x/chatd/quickgen.go
index c93bc9b7259..aea63d849f2 100644
--- a/coderd/x/chatd/quickgen.go
+++ b/coderd/x/chatd/quickgen.go
@@ -551,9 +551,6 @@ func validateGeneratedTitle(title string) error {
 	if title == "" {
 		return xerrors.New("generated title was empty")
 	}
-	if len(strings.Fields(title)) > 8 {
-		return xerrors.New("generated title exceeded 8 words")
-	}
 	return nil
 }
 
@@ -652,11 +649,20 @@ func titlePasteText(
 	return pasteText, nil
 }
 
+// titleMaxWords caps generated titles at the prompt's stated 2-8 word
+// budget. Quickgen pins temperature, so a model that overshoots the
+// budget for a given conversation keeps overshooting on retry; keeping
+// the first words of an otherwise good title beats failing generation.
+const titleMaxWords = 8
+
 func normalizeTitleOutput(title string) string {
 	title = normalizeShortTextOutput(title)
 	if title == "" {
 		return ""
 	}
+	if words := strings.Fields(title); len(words) > titleMaxWords {
+		title = strings.Join(words[:titleMaxWords], " ")
+	}
 	return truncateRunes(title, 80)
 }
 
diff --git a/coderd/x/chatd/quickgen_internal_test.go b/coderd/x/chatd/quickgen_internal_test.go
index fe67c2da4a4..a6a8123191b 100644
--- a/coderd/x/chatd/quickgen_internal_test.go
+++ b/coderd/x/chatd/quickgen_internal_test.go
@@ -861,6 +861,32 @@ func TestNormalizeTurnStatusLabel(t *testing.T) {
 	}
 }
 
+func Test_normalizeTitleOutput(t *testing.T) {
+	t.Parallel()
+
+	cases := []struct {
+		name  string
+		input string
+		want  string
+	}{
+		{name: "empty after normalization", input: " \"\" ", want: ""},
+		{name: "collapses whitespace and strips quotes", input: "\"Fix   pq  duplicate\tkey\"", want: "Fix pq duplicate key"},
+		{name: "keeps titles within the word budget", input: "Review risky changes in acme/webapp PR #123", want: "Review risky changes in acme/webapp PR #123"},
+		{
+			name:  "truncates overlong titles to the word budget",
+			input: "Re-capture Tasks-enabled evidence for the Coder pull request feature flag",
+			want:  "Re-capture Tasks-enabled evidence for the Coder pull request",
+		},
+		{name: "truncates to 80 runes", input: strings.Repeat("a", 100), want: strings.Repeat("a", 80)},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			require.Equal(t, tc.want, normalizeTitleOutput(tc.input))
+		})
+	}
+}
+
 func TestFallbackTurnStatusLabel(t *testing.T) {
 	t.Parallel()
 
@@ -946,6 +972,31 @@ func TestGenerateStructuredTitleWithUsage_DropsRejectedTemperature(t *testing.T)
 		"generation should retry without temperature after the model rejects it")
 }
 
+func TestGenerateStructuredTitleWithUsage_TruncatesOverlongTitle(t *testing.T) {
+	t.Parallel()
+
+	model := &chattest.FakeModel{
+		GenerateObjectFn: func(_ context.Context, _ fantasy.ObjectCall) (*fantasy.ObjectResponse, error) {
+			return &fantasy.ObjectResponse{
+				Object: map[string]any{
+					"title": "Re-capture Tasks-enabled evidence for the Coder pull request feature flag",
+				},
+			}, nil
+		},
+	}
+
+	title, _, err := generateStructuredTitleWithUsage(
+		t.Context(),
+		model,
+		titleObjectCall(resolvedModelCall{}),
+		titleGenerationPrompt,
+		"re-capture UI evidence for a Coder pull request",
+	)
+	require.NoError(t, err)
+	require.Equal(t, "Re-capture Tasks-enabled evidence for the Coder pull request", title,
+		"an overlong generated title should be truncated to the word budget, not rejected")
+}
+
 func newOpenAICompatStructuredOutputServer(
 	t *testing.T,
 	toolName string,