eLearning Script Writing: How to Write Narration That Sounds Human

An eLearning script is the narration for a lesson, written to be heard. The quickest way to a good one: write it from the objective, in plain conversational language addressed to the learner; let the words explain what’s on screen rather than read it out; plan about 150 words a minute and keep each segment short; build in moments for learners to do something; then read it aloud, and fix every line you trip over, before anyone records it.

Most guides to eLearning script writing agree on the advice to “be conversational” and “read it aloud”. We agree too. What they leave out is how much evidence sits behind that advice, where it runs out, and the decisions a script quietly makes for you: how the words and the screen share the work, how long the lesson runs, and whether a synthetic voice will do. This guide covers all of those, in eight steps.

What a script is, and where it fits

A script holds the words. A storyboard holds the words alongside the visuals, the interactions and the navigation for each screen. For a narrated video or a screencast, a script with a column of visual notes is often all you need. For an interactive module with branching and feedback, the script lives inside the storyboard’s narration column.

To keep things concrete, we’ll follow one example: a five-minute screencast that teaches new volunteers at a community food bank to log a shift in the rostering app. It’s an invented example, but it’s the kind of short, practical lesson many course builders script.

Step 1: Start from the objective and the practice, not the content

Before you write a word of narration, write down two things: what learners should be able to do at the end, and what they’ll do in the lesson to practise it. Our guide to writing learning objectives covers the first.

For the volunteer screencast, the objective is “log a completed shift, including a swap with another volunteer, in under two minutes”. The practice is logging a sample shift in the app’s training mode, straight after the demonstration.

Now every line of the script has to help learners get there. This is the best defence against the commonest script problem: a lesson written from the source material outwards, which ends up as a spoken policy document. If a sentence doesn’t move learners towards the task, it can go.

Step 2: Decide what kind of script you’re writing

Scripts for different formats need different amounts of detail:

Whichever format you choose, a simple template keeps the script and the screen together. This is the one we’d use for the screencast:

The “Learner does” column is the one most templates leave out, and it’s the one that keeps a script from becoming a lecture.

Step 3: Write for the ear

People can’t re-read speech. So a script needs shorter sentences, one idea at a time, familiar words and the main point early in each sentence. It also needs to sound like a person talking to one other person.

That last part has better evidence behind it than most writing advice. A meta-analysis of studies that rewrote instructional material in a conversational style, for instance by addressing the learner directly as “you”, found statistically reliable average gains on retention (d = 0.30) and on transfer, applying the material to new problems (d = 0.54) (Ginns, Martin & Marsh, 2013). The effect size, d, is a standard measure of how far apart the two groups’ average scores were. Mayer’s summary of his own group’s experiments puts the personalisation principle (conversational rather than formal wording) at an effect size of 1.00 across 15 tests (Mayer, 2024).

The meta-analysis also found a limit worth designing around:

That’s one more reason to keep each segment short. Here’s what the change looks like in the volunteer script.

Formal: “Utilisation of the shift-logging function is mandatory for all volunteers upon completion of each rostered shift. Non-compliance may result in inaccurate attendance records.”

Conversational: “At the end of every shift, you’ll log it in the app. It takes about a minute. If you skip it, the roster thinks you weren’t there.”

Same information, half the effort to follow. Some habits that get you there:

• Use “you” and “we”, and contractions (“you’ll”, “it’s”, “don’t”).
• One idea per sentence. If a sentence has two commas and an “and”, split it.
• Put the action first: “Tap My Shifts”, not “The My Shifts button should now be selected”.
• Explain a term the first time you use it, then use the same term every time.
• Write numbers the way you’d say them: “about a minute”, not “approximately 60 seconds”.

Conversational doesn’t mean jokey or chatty. It means plain, direct and polite: the way a good colleague explains something at your desk.

One thing we do at Avondale University when creating scripts to film for online content is to transcribe previous lectures, so the academic can see their exact words in class, and see how their own cadence of real speech looks written out on page. Academics are used to writing in a muscular way, using intellectual words, terms and acronyms. So it’s important to show them what plain, direct language looks like.

We also allocate studio time while filming to edit the lecture scripts on the fly, allowing academics to quickly change a phrase or paragraph that sounds too wordy, or doesn’t come across in a simple, easy to digest way.

Step 4: Make the words and the screen share the work

A script is never heard on its own. It plays against whatever is on screen, and several of Mayer’s design principles are really decisions about how the two divide the work.

Conversational wording had an effect size of 1.00 across 15 tests

Effect sizes Mayer reports for six principles a narration script settles, from his research group's experiments · effect size

Turn them into four questions to ask of every line:

Does the narration explain the picture, rather than a paragraph beside it? (Modality.) When the screen shows a graphic or a demonstration, put the explanation in the narration, so learners can look and listen at once.

Is the narration reading the screen aloud? (Redundancy.) In Mayer’s summary, adding printed text that repeats graphics and narration showed almost no benefit: an effect size of 0.10 across 12 tests. Keep on-screen text to labels, short key terms and instructions like “Your turn”. Captions that learners can switch on are a separate matter, covered in Step 8.

Does every line serve the objective? (Coherence.) The history of the rostering app, the anecdote about the volunteer who never logged a shift, the background music: each has to earn its place. Cutting extraneous material had an effect size of 0.86 across 19 tests.

Does the narration point at what matters? (Signalling.) Spoken cues such as “the key step here is the swap button” or “there are two things to check” help learners find the essential material.

Mayer is clear that each principle has boundary conditions: for whom, for what kind of lesson, under what circumstances. Treat them as strong defaults, then check them against what happens when real learners use the lesson. Our library entries on cognitive load and dual coding explain why these principles tend to hold.

Step 5: Get the length right

Script length is the easiest thing to get wrong and the easiest to check. Count the words, then time a read-through.

As a starting point, plan about 150 words a minute, so a six-minute video needs about 900 words. That’s a planning figure, not a target. In an analysis of 862 videos from four edX courses, Guo, Kim and Rubin found instructors spoke at anywhere from 48 to 254 words a minute, with an average of 156 (Guo, Kim & Rubin, 2014). Students generally stayed longer with videos where instructors spoke faster, but the authors read speaking rate as a sign of enthusiasm rather than something to aim for in itself, and they caution against drawing conclusions about learning from their engagement data alone. Their practical advice was to invest in planning before recording and to segment videos into chunks shorter than six minutes.

So time whoever will record the script, or the synthetic voice you’ll use, reading a full page of it. Then adjust the word count to that pace, not to a rule of thumb. If a lesson runs long, split it at a natural break and give each part its own small task. Our entry on microlearning covers when short lessons help and when they don’t.

Step 6: Write the practice into the script

A script that only explains is a lecture. The best ones stop and hand the work to the learner, and the words for those moments need writing as carefully as the explanations.

• Ask before you tell. “Before I show you, where would you look for the swap option?” A question asked before the answer gives learners a moment to think, rather than just wait.
• Ask why. “Why does the app ask for the other volunteer’s name?” Prompts like this ask learners to explain the reason behind a step. Our entry on elaborative interrogation covers what the research says about them, and their limits.
• Hand over clearly. “Now you try. Log the sample shift on your screen, then press Continue.” Say exactly what to do and how learners will know they’ve finished.
• Script the feedback. If the lesson checks the practice, write what learners hear or see after a right and a wrong attempt. Our guide to writing multiple-choice questions covers feedback that helps.

Step 7: Choose the voice, then write for it

For years the advice was clear: use a human voice. Mayer’s voice principle, that people learn better when narration is spoken in a friendly human voice rather than a machine voice, has an effect size of 0.74 across 7 tests in his summary. But synthetic voices have changed a great deal, and newer experiments tell a different story.

In a 2019 study, Craig and Schroeder compared narration by an older text-to-speech engine, a modern one and a recorded human voice (Craig & Schroeder, 2019):

They concluded that software may now be able to deliver narration credibly and effectively. A later experiment found that making a text-to-speech voice sound more enthusiastic didn’t improve learners’ scores on immediate retention and transfer tests, though the authors note those tests may have been too blunt to show deeper effects (Liew et al., 2023). These are single studies with short lessons, so we’d treat the evidence as encouraging rather than settled.

In practice, we’d decide on these grounds:

• Use your own voice, or a voice artist, when the lesson relies on your expertise and personality, or when learners will see you on camera.
• Consider a synthetic voice when the content changes often (a software walkthrough you’ll update every quarter), when you need several languages, or when you have no quiet place to record.
• Whichever you pick, write for it. For a synthetic voice, spell out abbreviations and numbers the way they should be said, add pronunciation notes for names and jargon, and listen to the whole lesson in that voice. A human narrator needs breath marks and a note on any word to stress.

We’ve never used synthetic voices in a lecture script at Avondale University, as we prioritise human contact and connection between lecturer and student. However, we are experimenting with using AI tools to transcribe readings, so students have the opportunity to listen to an academic paper in the same way they would listen to a podcast.

Step 8: Read it aloud, test it, and turn it into captions

Read it aloud yourself. Every place you stumble, a narrator will too. Long sentences, unclear references (“this”, “it” with nothing obvious to point to) and tongue-twisters show up at once.

Record a scratch track. A rough recording on your phone, played against the visuals, shows where narration and screen fall out of step: the cursor still moving when the narration has moved on, or a long silence while a screen loads.

Test it with a few learners. Play the scratch version to two or three people like your learners, and ask them to do the practice task at the end. If they can’t, the script needs work, and you’ve found out before the final recording.

Turn the final script into captions and a transcript. Your script is already most of the accessibility work. Prerecorded video with sound needs captions, as our storyboard guide explains. For audio-only content, such as a podcast-style lesson, WCAG 2.2 success criterion 1.2.1 (Level A) asks for an alternative that presents equivalent information, such as a transcript (W3C). Update the script to match what was actually recorded, then use it for both. Requirements vary by country and sector, so check which standard your organisation or client works to.

Five common mistakes

1. Scripting the source material. If the script follows the order of the policy or the manual rather than the path to the task, go back to Step 1.
2. Writing for the eye. Long sentences, passive voice and stacked clauses read fine and sound terrible. Read every line aloud.
3. Reading the screen. If the narration and the on-screen text say the same thing, cut the text back to a label.
4. No room for the learner. A script with no questions, pauses or tasks is a lecture, however short.
5. Guessing the length. Count the words, time a read-through, and split anything that runs long.

What to do next

Take the script for your next lesson, or the first page of it, and read it aloud with a pencil in your hand. Mark every place you stumble, every sentence longer than a breath and every line that reads out what’s on screen. Fix those before you record anything.