diff --git a/app/routes/calls_.record/new.tsx b/app/routes/calls_.record/new.tsx
index 967557943..a46f99235 100644
--- a/app/routes/calls_.record/new.tsx
+++ b/app/routes/calls_.record/new.tsx
@@ -12,6 +12,7 @@ import { H4, Paragraph } from '#app/components/typography.tsx'
 import { type RootLoaderType } from '#app/root.tsx'
 import { CallKentTextToSpeech } from '#app/routes/resources/calls/text-to-speech.tsx'
 import { type KCDHandle } from '#app/types.ts'
+import { formatCallKentTextToSpeechNotes } from '#app/utils/call-kent-text-to-speech.ts'
 import { getEnv } from '#app/utils/env.server.ts'
 import { type SerializeFrom } from '#app/utils/serialize-from.ts'
 import { type Route } from './+types/new'
@@ -156,11 +157,12 @@ export default function RecordScreen({
 							<CallKentTextToSpeech
 								onAcceptAudio={({ audio, questionText }) => {
 									setAudio(audio)
-									const cleanedQuestion = questionText.trim()
+									const formattedNotes =
+										formatCallKentTextToSpeechNotes(questionText)
 									setPrefill({
 										fields: {
 											title: 'Call Kent question',
-											notes: cleanedQuestion,
+											notes: formattedNotes,
 										},
 										errors: {},
 									})
diff --git a/app/utils/__tests__/call-kent-text-to-speech-notes.test.ts b/app/utils/__tests__/call-kent-text-to-speech-notes.test.ts
new file mode 100644
index 000000000..077ef0117
--- /dev/null
+++ b/app/utils/__tests__/call-kent-text-to-speech-notes.test.ts
@@ -0,0 +1,36 @@
+import { expect, test } from 'vitest'
+import {
+	AI_VOICE_DISCLOSURE_PREFIX,
+	formatCallKentTextToSpeechNotes,
+} from '../call-kent-text-to-speech.ts'
+
+test('formats typed-question notes with AI disclosure prefix', () => {
+	const notes = formatCallKentTextToSpeechNotes('  Hello from a typed call.  ')
+	expect(notes).toBe(
+		`${AI_VOICE_DISCLOSURE_PREFIX}\nTyped question: Hello from a typed call.`,
+	)
+})
+
+test('preserves internal newlines in typed question', () => {
+	const question = 'First line\nSecond line'
+	const notes = formatCallKentTextToSpeechNotes(question)
+	expect(notes).toBe(
+		`${AI_VOICE_DISCLOSURE_PREFIX}\nTyped question: ${question}`,
+	)
+})
+
+test('removes duplicate AI disclosure from typed question', () => {
+	const question = `${AI_VOICE_DISCLOSURE_PREFIX} Hello from a typed call.`
+	const notes = formatCallKentTextToSpeechNotes(question)
+	expect(notes).toBe(
+		`${AI_VOICE_DISCLOSURE_PREFIX}\nTyped question: Hello from a typed call.`,
+	)
+})
+
+test('returns empty string for empty input', () => {
+	expect(formatCallKentTextToSpeechNotes('')).toBe('')
+})
+
+test('returns empty string for whitespace-only input', () => {
+	expect(formatCallKentTextToSpeechNotes('   ')).toBe('')
+})
diff --git a/app/utils/call-kent-text-to-speech.ts b/app/utils/call-kent-text-to-speech.ts
index b4f8704f7..9ce363f89 100644
--- a/app/utils/call-kent-text-to-speech.ts
+++ b/app/utils/call-kent-text-to-speech.ts
@@ -10,6 +10,19 @@ export const callKentTextToSpeechConstraints = {
 
 export const AI_VOICE_DISCLOSURE_PREFIX = `This caller's voice was generated by AI.`
 
+export function formatCallKentTextToSpeechNotes(questionText: string) {
+	const cleaned = questionText.trim()
+	if (!cleaned) return cleaned
+
+	// Callers can paste the disclosure prefix into the textarea. Avoid duplicating
+	// it when formatting notes for the call record.
+	const prefixLower = AI_VOICE_DISCLOSURE_PREFIX.toLowerCase()
+	const questionBody = cleaned.toLowerCase().startsWith(prefixLower)
+		? cleaned.slice(AI_VOICE_DISCLOSURE_PREFIX.length).trimStart()
+		: cleaned
+	return `${AI_VOICE_DISCLOSURE_PREFIX}\nTyped question: ${questionBody}`
+}
+
 // @cf/deepgram/aura-2-en supports many speakers; we offer a curated subset.
 const auraSpeakers = [
 	'apollo',
diff --git a/docs/agents/project-context.md b/docs/agents/project-context.md
index 76c00ea42..d99a4902f 100644
--- a/docs/agents/project-context.md
+++ b/docs/agents/project-context.md
@@ -39,3 +39,10 @@ reference:
   It's populated on first request or via `npm run prime-cache:mocks`.
 - Content is filesystem-based: blog posts are MDX files in `content/blog/`.
   Changes to content files are auto-detected by the dev server's file watcher.
+
+## Cloud / headless manual testing
+
+- In cloud VMs, Chrome may block camera/microphone access by default. Visiting
+  `/calls/record/new` can hit the route ErrorBoundary unless `localhost` is
+  allowed mic/camera access in site settings (even if you intend to use the
+  typed question → text-to-speech path).