How to make a training video (2026): the 8-step method, from the need to the published video

This guide is for L&D, HR and health and safety teams, and for subject experts who have to produce a training video without being video makers. It follows the real order of the work, from the need to the published video, and stays neutral on tools: every step applies with editing software, an agency or an AI generator. The last part says where EducPilot fits, and where it does not.

The short answer

In one sentence: to make a training video, set one observable objective per video, write a short script for the ear from dated sources, pick a one to three minute format, add voice, visuals and captions, have the subject expert review the video itself, then publish it with its date and reference document.

Before you start: is video the right format?

A video builds understanding and makes the essentials stick. It replaces neither the reference document nor practice.

For training required by law, a video usually covers the information, rarely the practice: check what your regulator expects to be done in person.

1. The objective: what the viewer will do after the video

Write the learning objective before anything else, as one sentence with three parts:

1. Who: "new warehouse staff", not "employees".
2. An observable action verb: "go to", "apply", "spot", not "know" or "be aware of", which nobody can check.
3. What, under which conditions: "the assembly point, through the nearest exit, as soon as the alarm sounds".

Which gives: "After the video, new staff go to the assembly point through the nearest exit as soon as the alarm sounds."

Three rules follow:

• One objective, one video. Three objectives make three videos, not one video three times as long.
• Aim right. A short video mostly serves the first three levels of the revised Bloom's taxonomy: remember, understand, apply. Analysing, evaluating and creating are better worked on in a workshop.
• The objective settles arguments. Every scene must serve it; whatever does not stays in the document.

Our learning objective generator writes that sentence with you, from an audience and an action verb.

2. The script: sort the sources, open on a situation, write for the ear

Gather and sort the sources

Start from the documents that are authoritative: the procedure, the regulation, the help page, the classroom deck, the expert's notes. Prefer the official source to an article summarising it, and note its date.

Then sort. Keep what changes behaviour: the rule, the exception that catches people out, the date, the action. Leave history, full definitions and rare cases in the document. The test: if you cut this passage, would anyone act differently? If not, it stays in the document.

Depending on where you start, our converter pages say what actually happens: from a PDF, pasted text, a web page or a PowerPoint deck.

Write the script

An effective training script often has five parts. The examples come from our video on a fake CEO call, which runs 1 min 08 s (translated from French, see the video and its article):

The writing rules fit in four lines:

• One idea per sentence, sentences under 20 words.
• For the ear: numbers as you say them, no brackets, acronyms explained once.
• A source for every fact, noted with its page next to the sentence.
• A word budget: across our finished training videos, pauses included, a minute holds 117 to 137 words. A one to three minute video is therefore about 130 to 380 words. The voice-over duration calculator does the maths for your text.

The guide How to write a training video script covers this step in depth, with a real annotated script.

3. The format: style, frame and length

The style

Drawing has a measured edge for explaining. In a 2014 study of 6.9 million viewing sessions of online courses, tutorials drawn on a tablet held students' attention 1.5 to 2 times as long as slides or code screencasts (Guo, Kim and Rubin, 2014). That measures attention, not learning, and it concerns tutorials: to show a real physical task, film it. The guide Whiteboard explainer videos for training says when the format works and when filming is better.

The frame

• 16:9 for an intranet, a learning platform (LMS) or a meeting room;
• 9:16 for phones, Teams or Slack messages, social feeds.

The same content can exist in both frames, but the layout changes: plan for it at the storyboard stage.

The length

The same study found that median viewing time topped out at 6 minutes, whatever the video's length; its authors recommend cutting lessons into chunks shorter than 6 minutes. Aim for one to three minutes per objective. Beyond that, split: that is the idea behind microlearning.

4. The voice: human or synthetic

Research long favoured the human voice. But a 2017 study compared an older text-to-speech engine, a modern one and a human voice: the modern engine produced better learning transfer than the older one, and it was rated as credible and as helpful for learning as the human voice (Craig and Schroeder, 2017). In other words, the quality of a voice matters more than where it comes from.

Either way:

• a conversational tone, with "you" and "we";
• a calm pace: pauses help people understand, a sped-up voice does not;
• music far below the voice, or none at all;
• a full listen before sign-off, to catch mispronounced words.

See also voice-over in the glossary.

5. The visuals: show what the voice says

The picture does not copy the sentence: it shows it. Richard Mayer's principles of multimedia learning, drawn from his experiments, give the rules:

In practice:

• one visual and one keyword per idea, never the full sentence on screen;
• a comparison when the script contrasts two things, the right way and the wrong way side by side;
• a diagram when it describes a sequence;
• a character who lives the situation and applies the rule;
• your own pictures (photos of the site, screenshots of the software) when you have them, or pictures licensed for commercial use, credited.

Before recording anything, sign off a storyboard: one picture per scene, with its text. That is where corrections cost the least.

6. Accessibility and captions

In one sentence: an accessible training video has synchronised captions, a transcript, enough text contrast and nothing that flashes; if the voice says everything important the picture shows, it needs no audio description.

The reference is the W3C's WCAG 2.2:

In France, the RGAA 4.1.2 carries the same rules (criteria 4.1, 4.3 and 4.5 for media, 3.2 for contrast, 13.7 for flashes); its version 5 is due at the end of 2026.

Three tips:

• Burned-in captions are enough for WCAG, which lists always-visible captions among its sufficient techniques. A separate file (SRT, VTT) also lets viewers hide them and adds languages.
• Have the voice say what the picture shows: "the assembly point, in the car park, in front of entrance B", rather than "here".
• Publish the transcript under the video: it also helps anyone who wants to check a rule without replaying the video.

Even where the law does not require it of your company, these rules help everyone: part of the audience watches without sound, at a desk or on the move.

7. Review: on the video, by the expert

Three reviewers, three questions:

1. The subject expert: is every fact accurate and current?
2. Someone from the target audience: do they understand without explanation?
3. Legal or compliance, when the video states a rule: does the wording commit the company?

Review in two passes: the script or storyboard first, when changes cost almost nothing; the finished video next. On the video, ask for notes at the exact moment: "At 0:48, this figure is wrong." Faster for the expert, clearer for whoever fixes it.

Before you publish, check:

• one objective, written as one sentence;
• a situation within the first ten seconds;
• every fact tied to a dated source;
• reviewed by the expert, on the video itself;
• captions present and correct;
• length within the planned budget.

If AI wrote the video, this review is not optional: AI assistants can be confidently wrong. Feed them only your sources, and tie every sentence to the page that supports it.

8. Distribution: where, for whom, and when to revisit it

Where. In your LMS, as an MP4 file, or in a SCORM package if you need to track who watched it; on the intranet; in a message to the team concerned; behind a QR code at the workstation, for a task or an instruction.

With what. Under the video: the date, the objective, a link to the reference document and the transcript.

When to send it again. Something seen once is quickly forgotten: that is the forgetting curve. Send the video again as a reminder, a week then a month later, or just before the situation where it matters.

How to measure. The Kirkpatrick model separates reaction, learning, behaviour and results. At the very least, check the objective's verb: one question, or an observation on the job.

When to update it. When the rule changes. Keep a list of your videos with their sources and dates, and fix the scene concerned instead of starting over.

Which tool for which video

Our comparison of the best AI training video tools puts them side by side on the same criteria. For the budget, the training video cost estimator compares in-house, agency and EducPilot production, with sourced assumptions you replace with your own.

Where EducPilot fits, and where it does not

EducPilot turns your documents into drawn, narrated videos that your experts correct by annotating the frame. Step by step:

The limits, honestly:

• No avatar, no shoot, no screen recording. For a physical task or software click by click, film it.
• A synthetic voice, not your trainer and not a cloned voice.
• Three languages: English, French, Chinese.
• MP4 export, without a SCORM package or a separate caption file.
• PDFs, pasted text and web pages: a PowerPoint or Word document goes through PDF, a scan through OCR.
• Human review is still required. The tool makes it faster; it does not remove it.

On price, the free trial lasts 7 days, with 5 minutes of video and a "Made with EducPilot" mark on exports. After that, Pro costs €39 excl. VAT a month for 45 minutes, Team €99 excl. VAT for 130 minutes and Business €249 excl. VAT for 350 minutes. The storyboard and its revisions use no minutes: only the generated video counts. Details are on the pricing page.

By use case

Frequently asked questions

How long should a training video be?

One to three minutes per objective. Across 6.9 million viewing sessions of online courses, median viewing time topped out at 6 minutes, whatever the video's length (Guo, Kim and Rubin, 2014). A longer topic splits into several videos.

How many words per minute of video?

117 to 137 words, from our measurements on finished training videos, pauses included. The voice-over duration calculator gives the length of your text.

Do I need a human voice?

Not necessarily. In a 2017 study, a modern synthetic voice was rated as credible and as helpful for learning as a human voice (Craig and Schroeder). It also has the advantage of being fixed in minutes. A human voice is still better for a personal message from leadership or a testimonial.

Can I make a training video from a PowerPoint?

Yes. PowerPoint itself exports a deck to video, with your voice. For a rewritten, drawn video, export the deck to PDF and start from that PDF: the PowerPoint to training video page details what carries over and what is lost.

Are captions required?

Accessibility rules (WCAG 2.2, success criterion 1.2.2) call for captions on any prerecorded video with sound. Whether the law requires them of your company depends on your sector and country, but they help everyone.

Sources

• Philip J. Guo, Juho Kim, Rob Rubin, How video production affects student engagement: an empirical study of MOOC videos, ACM Learning at Scale, 2014: 6.9 million viewing sessions on edX; median viewing time of 6 minutes at most; tablet-drawn tutorials watched 1.5 to 2 times as long as slides or code screencasts.
• Scotty D. Craig, Noah L. Schroeder, Reconsidering the voice effect when learning from a virtual human, Computers & Education, vol. 114, 2017, pp. 193 to 205.
• Richard E. Mayer, Multimedia Learning, 3rd edition, Cambridge University Press, 2020.
• David R. Krathwohl, A Revision of Bloom's Taxonomy: An Overview, Theory Into Practice, vol. 41, no. 4, 2002.
• W3C, Web Content Accessibility Guidelines (WCAG) 2.2, Recommendation of 12 December 2024 (success criteria 1.2.2, 1.2.3, 1.2.5, 1.4.3 and 2.3.1); Understanding 1.2.2 (always-visible captions, technique G93); Understanding 1.2.5 (no audio description needed when the soundtrack says it all).
• DINUM, RGAA 4.1.2, criteria and tests (in French; version 5 due at the end of 2026).
• Microsoft, Turn your presentation into a video.
• Words per minute: measured on the training videos published on educpilot.org, detailed on the calculator page.
• How EducPilot works, and its prices, as of 23 October 2026.