Welcome!

Writing Scripts That Teach

TL;DR; Narration scripts aren't articles. They're what someone would say out loud to a colleague who's standing next to them. Write them conversationally, keep them short, and read them aloud before you generate the voice. If it sounds weird when you read it, it will sound weird when the AI reads it.

Your script is the slide's speaker notes. When Lupo generates a video, it reads those notes out loud over the slide. Good notes make good videos; clunky notes make clunky videos. This chapter is about the writing craft itself.

Written prose vs. spoken prose

The single biggest mistake people make with narration scripts is writing them like a document. Documents are meant to be read with your eyes, and your eyes can re-read a sentence, skim ahead, and pause when they get confused. Audio is meant to be heard linearly, once. The rules are different.

Examples of things that work in a written article but don't work in narration:

Long sentences with lots of subordinate clauses. Readable on a page, impossible to follow when spoken.
Parenthetical asides — the listener has already forgotten the start of the sentence by the time the parenthetical ends.
Lists introduced by "the following" or "as shown below." There is no "below" in audio. The listener can't skim down.
Numbers read as "approximately 37.4%." Fine on a page. Awkward spoken. Say "a little over a third."

The rule: write the way a smart, friendly colleague would talk. Short sentences. Simple structure. One idea per sentence. Natural phrases, not business-document phrases.

Read it aloud

This is the single most useful habit you can build as a script writer. Before you generate a narration, read the script out loud yourself.

If you stumble, the AI will stumble. If a sentence runs out of breath, it will sound out of breath when narrated. If a transition feels abrupt, it is abrupt. Your mouth is a surprisingly good QA tool.

Most bad AI narrations come from people who wrote a paragraph, typed it into the speaker notes, clicked generate, and were surprised it didn't sound good. The fix is a 30-second read-aloud before you click generate.

Structure: tell them, tell them, tell them

Old public speaking advice that works: "Tell them what you're about to tell them, tell them, then tell them what you told them." For a short narration, it's not quite that explicit, but the shape is similar.

A typical 60-second narration for a slide might be structured like this:

Hook (5-10 seconds): Why should the listener care about this slide?
The main point (30-40 seconds): The actual content.
The takeaway (5-10 seconds): One sentence that summarizes what the listener should walk away with.

You don't need to label these. You just need the shape. Every slide should pull the listener in at the start and leave them with something concrete at the end.

Length

How long should narration be per slide?

A useful rule of thumb: 60 to 120 seconds of narration per slide, max. If a slide needs more than two minutes of narration, it's actually two slides that got glued together. Split them.

You can tell how long the narration is by reading it aloud at a normal pace. A well-paced paragraph is roughly 150 words per minute, so 100 words is about 40 seconds of narration. If your slide's notes run past 200 words, you're approaching the two-minute ceiling.

Long narration isn't a sign that you have more to say; it's usually a sign that you haven't decided what the slide is actually about.

Tone

Match the tone to the audience. A safety training course and a product marketing walkthrough need different voices. But for most workplace training, the goal is warm, conversational, and credible — like a senior colleague explaining something to a new team member.

Things to avoid:

Corporate boilerplate. "In today's fast-paced environment..." Skip it. Start with the actual content.
Marketing hype. Words like "game-changing," "best-in-class," "world-class." They're empty and they make the listener tune out.
Self-deprecating filler. "Now, I'm not an expert, but..." You're narrating a training course. You're the expert for the next three minutes. Own it.

Using emotion tags

Lupo supports inline [[emotion]] tags in speaker notes for Premium voices. These let you shift the voice's tone for a specific line — more on this in Emotions.

A few useful patterns:

Use [[serious]] on a slide about safety, risk, or consequences.
Use [[cheerful]] or [[excited]] on a slide about something positive (a feature announcement, a win).
Use [[empathetic]] on a slide where you're acknowledging something the learner might be struggling with.

Don't overdo it. A narration with three different emotion tags per sentence sounds theatrical and distracting. Usually one tone per slide, set once at the start, is plenty.

Writing speaker notes directly in PowerPoint

Lupo reads speaker notes from the .pptx file itself, so the most natural workflow is:

Build your slides.
For each slide, open the speaker notes panel.
Write the narration directly in the notes.
Save.
Upload to Lupo.

You can also write the scripts in a separate document first and paste them in — useful if you're iterating on the wording before you lock the deck — but eventually they need to live in the notes panel.

What to do if you hate your first draft

The first draft of any narration script is usually bad. That's normal. Two techniques that help:

Cut 30%. Take what you wrote and aggressively delete. If a sentence doesn't earn its place, delete it. Read the shorter version aloud. It's almost always better.
Rewrite as if you were explaining it to one specific person. Pick a real colleague and imagine you're standing at their desk explaining the slide to them. Write what you'd actually say. It will sound much more human than your first draft.

Where to go next

Delivering and Measuring Training — once your scripts are good, how do you roll the course out and know if it worked?
Emotions — the reference for emotion tags.
Go Deeper — the masterclass has a full chapter on scriptwriting with example rewrites.