Lesson 1 of 8 · 7 min read

Does it need a video, and a presenter?

Decide whether your process needs a video at all, then whether it needs anyone on screen, using the image principle and the substitution test.

Start with the format, not the tool

Most requests for a training video arrive with the format already chosen. Someone asks for “a video”, and the next question is which tool makes it. This lesson asks two earlier questions. Does the process need a video at all? If it does, does the video need a presenter on screen?

The answers decide most of the cost. A screenshot guide costs less to make than a video. A narrated screen recording needs no consent forms. An avatar of a real colleague needs consent, disclosure and a plan for when that person leaves, which Lesson 6 covers.

The course project starts here. You pick one process from your own work and choose a format for a reason you can write down. Every later lesson adds a piece to the same video plan.

When a guide beats a video

Some processes are better taught by a step-by-step guide with screenshots. A reader can work at their own pace, find one step again, and keep the page open beside the system. For a software procedure that changes often, a guide may also cost less to maintain.

Update cost depends on the format. Trainn, which sells a competing product, compared two tools. It says Scribe guides update by capturing the steps again, while Guidde videos need manual edits. Scribe’s pricing page confirms that its output is annotated screenshot guides, not video. Read the comparison as a competitor’s view, not a measured result.

A video is worth its cost when the learner needs to see a sequence in motion or hear the reason behind a choice. It also helps when the learner must watch someone handle a judgment call. A guide is worth its cost when the learner needs a reference to follow click by click. Many processes need both: a short video that teaches the decisions, and a one-page guide that sits beside the screen.

A face on screen adds little to learning

If a video is the right format, the next question is whether anyone appears on screen. Richard Mayer’s review of multimedia research gives the clearest answer. Across nine experiments, adding the instructor’s image, as a talking head or a still drawing, did not substantially improve learning. The median effect size was d = 0.26.

Mayer says the evidence does not support an image principle. In plain terms, a face on screen is optional. A narrated screen recording, with the narration describing what the screen shows, can teach a software procedure with nobody in the frame. Lesson 3 covers that narration.

Mayer’s figures come from his own experiments, as reported in his 2014 review chapter. They measured learning on tests, not performance at work. Even so, they are the best evidence available on whether a face helps, and they do not show that it does.

What the avatar studies found

AI avatars are a newer kind of presenter. Torbjørn Netland and colleagues ran a crossover study with 447 online participants. Each watched two human-made and two AI-made videos on production management. Exam results showed no statistically significant difference, and learners enjoyed the human-made videos more.

A 2024 study of 108 undergraduates found the same pattern. Instructor type had no significant effect on test performance. Video engagement was higher with the human instructor. Students in a focus group described the AI instructor as distracting and uncomfortable.

Both studies used short videos and short-term tests. Neither measured what people did at work afterward. So the fair summary is narrow. An avatar is acceptable for basic knowledge transfer: learners scored about the same and enjoyed it less. No study here shows that an avatar improves learning or engagement.

Match the presenter to the content

Training Industry published a guide by an analyst at TechSmith, which sells video software. It suggests an AI voice or avatar for compliance content that changes often, software tutorials, product updates and translation. It suggests a human presenter for leadership messages, culture and sensitive topics. The guide uses only TechSmith’s own data, so read it as one vendor’s advice.

EdzLMS, a learning-platform vendor, adds the substitution test. For each module, ask whether it still works if the learner thinks the presenter is not real. A software walkthrough passes. A leader’s welcome to new staff fails, because the point is that the leader said it.

For modules where the answer is unsure, EdzLMS suggests a pilot in two versions with about 30 learners. Philippa Hardman, who writes about learning design, recommends judging such versions on learning, not on how engaging they feel. Give both versions the same scenario questions from Lesson 2 and compare the scores.

One topic group needs special care. Emeri’s researchers found 14 learner posts that describe lost trust when a course about AI, security or ethics used an AI presenter. One describes a module on the risks of trusting AI, presented by AI actors. Another describes a course on spotting deepfakes with an AI voiceover.

All 14 posts came from Bluesky, where many users lean against generative AI. They show that the problem recurs, not how often it happens. Still, the pattern is consistent. If your topic is AI, security or ethics, an AI presenter contradicts what the course teaches.

Worked example · Larkfield Supply

Riley chooses the format for the returns video

Riley, the training coordinator at Larkfield Supply, must teach 25 staff the new returns process in the order system before it goes live on December 1. The staff work in the warehouse and the office. The process has five steps. Staff look up the order and check the return reason against the 30-day policy. They choose refund or replacement, print the return label, and log the item’s condition.

  1. Riley first asks whether a guide would do. Four of the five steps are clicks in the order system. A screenshot guide would teach those well, and Riley could capture it again when the screens change.
  2. The choice between refund and replacement is different. It is a judgment under the 30-day policy, and staff need to hear why each choice is right. Riley decides on a short video for the decisions and a one-page screenshot guide for the clicks.
    What would you do next?Riley’s manager suggests an avatar of Morgan, the warehouse manager, to present the video. What should Riley check first?Think it through, then show the answer

    Whether the video needs a presenter at all. The image principle says a face adds little to learning. The video shows screens, so narration can carry it with nobody in the frame.

  3. Riley runs the substitution test. If staff know the voice is not a real person, does the module still work? It does. The content is a software procedure, and nothing in it depends on trusting one speaker.
  4. Riley checks the topic group. Returns are not about AI, security or ethics, so an AI voice does not contradict the message.
  5. Riley weighs Morgan’s avatar against what it would add. It would need Morgan’s written consent, a disclosure, and a plan for when Morgan leaves. It would add a face, which the research says does not improve learning. Riley declines the avatar for this video, and Lesson 6 explains the consent work it would have needed.That answers the question you guessed at: Guess: does putting the instructor’s face on screen make a training video teach better?
    What would you do next?Why an AI voice rather than Riley’s own recorded voice?Think it through, then show the answer

    The screens may change before and after go-live. With an AI voice, Riley edits the script and generates the narration again, instead of booking another recording. The Training Industry guide lists software tutorials and product updates as good uses.

  6. Riley plans two segments of about 4 minutes each. Segment one covers looking up the order, checking the policy, and choosing refund or replacement. Segment two covers the label and the condition log.

Riley writes the decision at the top of the video plan: a narrated screen recording with an AI voice, no avatar, in two segments of about 4 minutes. A one-page screenshot guide sits beside it for the clicks. Lesson 2 adds the next part of the plan, the questions.

Starter prompts

  • Format interview

    I need to teach a process at work: [describe the process, its steps, and who must learn it]. Interview me, one question at a time, to decide the right format. Ask how often the process changes, which steps are clicks and which are judgment calls, where people will use what they learn, and whether the topic involves AI, security or ethics. Then recommend one of these: a step-by-step screenshot guide, a narrated screen recording, a video with a presenter, or a guide plus a short video. Give the reason in two sentences. Do not assume facts I have not given you. If you need a fact, ask for it.

  • Run the substitution test

    Here is a description of a training module: [paste the topic, the audience and an outline]. Apply the substitution test: would the module still work if every learner believed the presenter was not a real person? Answer yes, no or unsure, with one reason. Then say whether the topic involves AI, security, ethics or a sensitive people issue, where an AI presenter can cost trust. If the answer is unsure, describe a two-version pilot with about 30 learners and the same test questions for both versions. Do not recommend a vendor.

When not to use AI

  • A message where the speaker is the point, such as a leader’s message to staff or a sensitive people topic. Use the real person.
  • A course about AI, security or ethics. An AI presenter there contradicts what the course teaches.
  • Making the format decision for you. An assistant can list options, but only you know how often the process changes and who will use it.
Recall from memoryWhat did Mayer’s review find about the instructor’s image on screen, and what can teach a software procedure instead?Answer it without looking back, then show the answer

Across nine experiments, adding the instructor’s image did not substantially improve learning. A narrated screen recording, with the narration describing what the screen shows, can teach a software procedure with nobody in the frame.

Exercise · your project

Choose the format for one process you need to teach.

  1. Pick one real process from your work, and list its steps.
  2. Mark each step as a click or a judgment call, and note how often the process changes.
  3. Decide between a guide, a video, or both, and write the reason in two sentences.
  4. If you chose video, run the substitution test and check the topic group before you decide on a presenter.
  5. Write the decision at the top of your video plan. Every later lesson adds to this plan.

Sources