Appearance
Script (AI Voiceover) & Subtitles – Best practices for personalised videos
This article is a practical guide to writing scripts (AI voiceover) and subtitles for personalised videos in Studio.
What "good" looks like
Keep videos no longer than 1 minute, preferably closer to 30 seconds, when the goal is conversion.
Start personalisation early, usually with first_name.
Keep the voiceover simple, and use visuals for details.
Use one clear CTA.
Always include subtitles since many viewers watch muted.
Plan the script around personalisation
Personalisation is powerful, but it must feel natural.
Recommended "sweet spot"
Aim for 1–3 personalisations, and use only personal data that is clearly relevant. Avoid sensitive data like full addresses.
Common, high-performing personalisations
Use first_name in the AI voice and on-screen text.
Use place_name or another location field if it supports the CTA.
Use contextual fields only when they truly help the viewer, such as car model, booking date, or appointment time.
Use a simple structure
A reliable structure for short personalised videos is Hook → Context → Proposition → CTA.
Hook (personalised)
Say and show the name early. Example: "Hi, first_name!"
Context
Explain why the viewer is receiving this message.
Proposition (offer or value)
Put the most valuable point early. Do not wait until the end to reveal it.
Call to action (CTA)
Use one clear next step. Example: "Tap the button to redeem your free drink."
Many viewers give only ~12–17 seconds of attention. If the offer comes late, people drop off before they get the point.
Write the voiceover to fit timing and variation
Voiceover is the timing backbone of the video. Everything else should be built around it.
Recommended workflow
Draft and test scripts directly in Studio.
Generate voiceover early to hear cadence and confirm video length.
After the script length is right, add and polish visuals.
Design for "long values"
Personalised fields vary in length.
"Hi first_name" is usually safe, but longer values can cause audio overlap.
Leave an extra pause after personalised segments. Test with your data using the longest expected value.
Prefer showing long details as on-screen text rather than voicing every detail.
Control pronunciation without breaking subtitles
AI pronunciation is sometimes wrong for names and place names.
Approach
Adjust pronunciation in the script using phonetic spelling. Example: Instead of "Jan," write "Yan" for a more aligned pronunciation.
Subtitles do not need to match the voiceover character-by-character in non-personalised parts. You can keep correct spelling in subtitles while using phonetics for the AI voice.
Subtitles
Many people watch with sound off, so subtitles need to convey all the relevant audio.
Best practices
Generate subtitles from the AI voice layer whenever possible.
Edit subtitles to correct spelling for brand terms and place names.
Edit subtitles to improve readability with shorter lines.
Add personalisation tags where needed.
Multiple languages
You can add multiple subtitle languages.
Viewers can switch language in the player.
Subtitles are on by default.
Keep the message "video-first"
Personalised video is usually delivered inside an email or SMS flow.
Do not fully repeat the video message in the email.
Email should tease and motivate the click.
Video should deliver the main message and value.
QA checklist (before publishing)
- [ ] Video length is within target: no longer than 1 minute, preferably closer to 30 seconds.
- [ ] Personalisation tags match data fields exactly.
- [ ] Timing buffer covers the longest personalised values.
- [ ] Pronunciation has been tested for key dynamic fields.
- [ ] CTA is clear and not competing with other CTAs.
Example skeleton
Short personalised offer script
- "Hi,
first_name!" - "It's your birthday, and we're celebrating you."
- "You've got a free gift waiting at
location." - "Click the button to claim it."
Tip: Test with the longest expected values, like long names and long venue names, before you lock timing.