Lesson 4 of 8 · 7 min read

Storyboard the segments

Split the script into short segments, tie each on-screen word to the narration, and show one full worked example before learners try.

A video that renders is not a designed video

Jewoong Moon, an instructional designer, built a video about the history of instructional design with AI tools and described the process on LinkedIn. Moon’s first cut rendered without errors and looked like a narrated slide deck. The file worked. The teaching did not.

The fix was a storyboard and a review loop. Moon planned the scenes, timed the narration, tied on-screen notes to exact spoken phrases, added cutaways, and reviewed a contact sheet of frames before the final render. This lesson teaches the same steps for a short training video. It is one practitioner’s method, not a study, and the research below supports its main parts.

Plan short segments that learners control

Split the video into segments of about 6 minutes or less. Guo, Kim and Rubin studied 6.9 million viewing sessions in four edX courses from 2012. Median engagement time was at most 6 minutes, whatever the video length. Students often watched less than half of videos longer than 9 minutes.

Guo and colleagues measured engagement, not learning. Their measures were watch time and whether the student tried the next problem. They call engagement necessary but not sufficient for learning. So the 6-minute figure is a planning guide for attention, not a rule about what people learn.

Mayer’s segmenting finding adds the learning side. Learner-paced segments beat one continuous presentation, with a median effect size of d = 0.82 across nine experiments. The effect may be strongest for people new to the topic. Learner-paced means the learner decides when the next segment starts, so put a clear stop at each break.

Plan the break points from the script, not the clock. End each segment at a decision the learner can practice, and put the matching scenario questions right after it. Riley’s segment one ends with all four questions from Lesson 2, because it teaches every decision they test. Segment two ends with a short task: print the label and log the condition for the lamp from Question 1.

Show one full example before learners try

Each segment should walk through one complete, realistic case before it asks the learner to do anything. Novices learn more from studying worked examples than from solving problems unaided. This is the worked-example effect. It can reverse for experienced learners, so a refresher for experts needs fewer examples.

Add pauses that ask for the next step. In a study published in December 2025, 164 adults studied worked examples one step at a time. Those asked to describe each upcoming step before seeing it did better on a test a week later. They also spent longer studying, about 31 minutes against 13.

In a video, that pause is a line such as “What would you check next? Pause here and decide.” The learner commits to an answer, then the next frame shows it.

Tie every screen word to a spoken phrase

Signaling means cues that show the learner what matters. Mayer found that verbal signals, such as headings and outline sentences, helped with a median effect size of d = 0.52. He rated the evidence as moderate. Visual signals such as arrows were not found to help reliably.

In the storyboard, each on-screen key word appears at the moment the narration says it. “30 days from delivery” appears when the voice says “the 30 days start from the delivery date”. The learner hears and sees one idea at once. Moon tied on-screen notes to exact spoken phrases for the same reason.

Decide what each frame shows

For a procedure, most frames are the screen recording itself. Plan a cutaway only where it helps the task, such as a photo of the label printer or the returns shelf. Use a real photo or a real recording from your own workplace.

Never plan generated footage where physical accuracy or readable text matters. One learner reported a safety module in which two generated forklifts collided and their loads merged into each other. The module carried an AI label, and the label did not make the scene correct. Another learner reported garbled text on sample documents in a data-integrity course.

If you use an avatar, place it picture-in-picture over the screen content, not full screen. TechSmith, which sells video software, recommends this from its own viewer study, which did not report a full sample size. For a procedure, TechSmith recommends a screen recording with an AI voice.

Review a contact sheet before the final render

A contact sheet is a page of still frames, one for each storyboard row. Review it before the final render. Check that each frame shows what the narration says at that moment. Check that every key word is spelled right and matches the spoken phrase.

Check also that no frame carries text a learner must read but cannot. Fixing a row on paper costs less than rendering the video again.

Worked example · Larkfield Supply

Riley storyboards segment one

Riley turns the locked segment-one script from Lesson 3 into a storyboard. Each row has four parts: the time, the narration phrase, what the screen shows, and the on-screen key words. Segment one teaches looking up the order, checking the 30-day policy, and choosing refund or replacement.

  1. Row 1, 0:00 to 0:10, holds the disclosure, which Lesson 6 writes. Row 2, 0:10 to 0:30: “In this part, we process one return together, from lookup to decision.” The screen shows the order system’s home page. Key words: “Lookup to decision”.
  2. Row 3, 0:30 to 1:00: “Here’s our return: an Aubriot desk lamp, product code L-K-S 20417.” The screen shows the order number typed into the returns search box. Key words: “LKS-20417”.
  3. Row 4, 1:00 to 1:40: “Look at the delivery date. The 30 days start here, not at the order date.” The screen zooms on the delivery date, December 9. Key words: “30 days from delivery”.
    What would you do next?The delivery date is December 9 and today is January 6. What should the next row show?Think it through, then show the answer

    The count. Row 5 says, “That’s 28 days, so this return is inside the window,” with the key words “Day 28: inside”. The full example shows the check before learners do it alone.

  4. Row 6, 2:10 to 2:50: “Now ask the customer: refund or replacement? It’s their choice.” The screen shows the decision panel with both options. Key words: “Customer chooses”.
  5. Row 7, 2:50 to 3:30: “They want a replacement, so check stock first. If it’s out of stock, offer a refund instead.” The screen shows the stock panel with 14 lamps in stock. Key words: “Out of stock? Offer refund”.
  6. Row 8, 3:30 to 4:00: “Your turn. Pause here and answer the four questions below.” The screen shows an end card with a continue button, then scenario questions 1 to 4 from Lesson 2.
    What would you do next?Riley’s manager asks for a generated clip of a customer handing a lamp back at a counter, to open the segment. Should Riley add it?Think it through, then show the answer

    No. It teaches nothing the scenario questions need, so the coherence finding argues against it. Generated people and objects can also come out wrong, as the forklift report shows.

The storyboard for segment one has eight rows and runs about 4 minutes. Riley makes a contact sheet from a test recording and finds one problem: the key words in row 4 appear two seconds before the narration says them. Riley moves them. Lesson 5 chooses the voice and the recording tool for this plan.That answers the question you guessed at: Guess: when should an on-screen key word such as “30 days from delivery” appear?

Starter prompts

  • Script to storyboard

    Here is a locked narration script for one segment of a training video: [paste the script]. Turn it into a storyboard table with these columns: row number, start and end time, narration phrase, what the screen shows, on-screen key words, and notes. Rules: on-screen key words must repeat a phrase the narration says in that row, and appear when it is said. Keep key words to five words or fewer. Show one complete worked example before any question for the learner. Add a pause row where the learner answers questions. Mark any frame that would need generated footage of people, objects or readable text, and suggest a screen recording or a real photo instead. Do not change the narration. If the script does not say what the screen shows, write "GAP".

  • Plan the segments

    Here is my locked script and my learning objective: [paste both]. Suggest where to split the script into segments of about 6 minutes or less. Each segment should end at a natural decision point, followed by scenario questions. Give the break points and a one-line purpose for each segment. Do not add or change content.

  • Contact-sheet review

    Here are still frames from my draft video, one per storyboard row, with the narration for each: [attach the frames and paste the narration]. For each frame, check: does it show what the narration describes; are the key words spelled correctly and the same words the narration uses; is any text too small or garbled to read; does any generated image show something physically wrong. List problems by row number. Do not suggest new content.

When not to use AI

  • Generating footage of people, machines or documents. Physical errors and garbled text teach the wrong thing. Use screen recordings or real photos.
  • Signing off the contact sheet. An assistant can flag likely problems, but you check each frame against the process yourself.
Recall from memoryWhen should an on-screen key word appear, and what do you review before the final render to check it?Answer it without looking back, then show the answer

At the moment the narration says it, so the learner hears and sees one idea at once. Before the final render, review a contact sheet with one still frame for each storyboard row.

Exercise · your project

Storyboard the first segment of your video.

  1. Split your locked script into segments of about 6 minutes or less.
  2. Build a storyboard table for the first segment with the first prompt.
  3. Check that one full worked example comes before any question.
  4. Replace any planned generated footage of people, objects or text with a screen recording or a real photo.
  5. Add the storyboard to your video plan.

Sources