All articles
6 min readlearning / training

Highlight 3 lines, cut the fun photo: evidence on training slides

I build a deck for every training day I run. For years I built them the way most trainers do: the sentence I was about to say written out on the slide, a photo to wake the room, a diagram with a paragraph of explanation beside it. Sascha Schneider's group pooled 103 studies covering 12,201 learners and found that one cheap change, marking the part of the slide you are speaking about, moved retention by 0.53 standard deviations 1. Two of my three habits were working against that number.

What the research shows.

Signaling means a cue that marks the relevant element: an arrow, a box, a color, a heading that states the point instead of naming the topic. Schneider's meta-analysis pooled 139 retention and 70 transfer outcomes and reported g = 0.53 for retention (95% CI 0.42 to 0.64) and g = 0.33 for transfer (0.22 to 0.43) 1. Reported cognitive load fell. Prior knowledge did not moderate the effect, so the cues paid off for experienced learners too. Alpizar, Adesope and Wong ran a separate meta-analysis over 44 effect sizes from 29 experiments with 2,726 participants and found the same direction 2.

Paul Ginns pooled 43 independent effects on the modality question: the same graphic, explained once in spoken words and once in printed words next to it. Spoken won, d = 0.72 3. The advantage grew with element interactivity, meaning material whose parts have to be held together to make sense, and with system-paced presentation, where the learner cannot stop the lesson to catch up.

Richard Mayer, Julie Heiser and Steven Lonn ran four experiments on a lightning animation with narration. Students who also saw on-screen text duplicating or summarizing that narration scored worse on retention and on transfer than students who watched the animation with narration alone 4. The same four experiments tested the other habit. Adding interesting narration segments to the lesson, or dropping an interesting video clip before it or inside it, lowered transfer scores.

Chao Cheng's team put a size on that second finding in 2026 with a three-level meta-analysis of 177 effect sizes from 50 studies. Seductive details, material that holds attention while sharing no structure with the lesson, cost g = -0.16 on learning overall: -0.19 on comprehension, -0.17 on recall, -0.12 on transfer 5. Their structural model traced the damage to extraneous cognitive load rather than to lost time. An earlier synthesis of 68 studies by Sundararajan and Adesope pointed the same way 6. The effect is small, and it is free to collect: you delete the photo.

Two more results shape how a session runs rather than how a slide looks. Rey and colleagues pooled 56 investigations and 88 pairwise comparisons on segmenting, breaking continuous material into chunks the learner can pace. Retention and transfer both improved by small to medium margins, cognitive load dropped, and learning time went up 7. Ginns, Martin and Marsh pooled rewrites of instructional text into a conversational register, second person instead of third: retention d = 0.30, transfer d = 0.54 8.

Michael Noetel's team reviewed the whole field in 2022, pooling 29 reviews that covered 1,189 studies and 78,177 participants. Eleven design principles produced positive meta-analytic effects on learning, and five improved cognitive load management. The largest gains went to captioning second-language video, to putting related words and pictures close together in time and space, and to signaling 9.

The rule against on-screen text has a boundary.

Adesope and Nesbit extracted data from 57 studies comparing spoken-only, written-only and spoken-written presentations. Learners who got both spoken and written words beat learners who got speech alone 10. The advantage held for system-paced material (g = 0.34) and faded when learners controlled the pace (g = 0.12). It held for low prior knowledge learners and for picture-free material, and it disappeared when the presentation included a diagram or animation that stood on its own.

Mayer's redundancy result and this one describe the same mechanism from two sides. Words printed beside a picture make the eyes choose. Words printed with nothing to look at give a learner who missed the audio a second route in. A slide carrying a schema and a paragraph is the case that hurts.

The second boundary is duration. In Ginns, Martin and Marsh, the conversational-style effect shrank to nothing in studies running longer than 35 minutes 8. Most of this literature tests 10-minute lessons in a lab, on undergraduates, with a quiz an hour later. My training days run 6 hours with people who came in from their own backlog. Treat the numbers as a ranking of design choices rather than as a forecast of what your Thursday cohort will score.

The protocol.

  1. Write the point of each slide in its title as a full claim. "Retrieval beats rereading" instead of "Retrieval practice". The title is the cue that costs nothing.
  2. Mark 2 or 3 elements per slide with a box, an arrow or one accent color, and mark the ones you speak about. Check it by muting yourself: a stranger scrolling the deck should land on the same spots your voice does.
  3. On any slide holding a diagram, delete the explanatory paragraph and say it out loud instead. That is the modality effect, and it is the single biggest rewrite in most decks.
  4. On a slide holding no picture, keep the words. Bullet text beside your narration helps when the eyes have nothing else to do, above all in a room where some of the audience works in a second language.
  5. Delete the decorative photo, the joke, the war story that shares no structure with the point you are teaching. Keep the story that carries the mechanism.
  6. Cut the long demo into segments of 3 to 5 minutes and put a stop between them: a question to the room, a 60-second exercise, a pause for notes. Hand the pacing over whenever the format allows it.
  7. Rewrite the deck in second person during the final pass. "You paste the transcript, the model returns 5 themes" instead of "the user pastes".
  8. Judge the result on a delayed quiz, never on the room's satisfaction score, for the reason I laid out in the article on feedback forms.

Open the deck you are presenting this week and count the slides where a paragraph of text sits beside a diagram. That count is your first pass.

Sources.

  1. Schneider, S., Beege, M., Nebel, S., & Rey, G. D. (2018). A meta-analysis of how signaling affects learning with media. Educational Research Review, 23, 1-24. doi.org/10.1016/j.edurev.2017.11.001
  2. Alpizar, D., Adesope, O. O., & Wong, R. M. (2020). A meta-analysis of signaling principle in multimedia learning environments. Educational Technology Research and Development, 68(5), 2095-2119. doi.org/10.1007/s11423-020-09748-7
  3. Ginns, P. (2005). Meta-analysis of the modality effect. Learning and Instruction, 15(4), 313-331. doi.org/10.1016/j.learninstruc.2005.07.001
  4. Mayer, R. E., Heiser, J., & Lonn, S. (2001). Cognitive constraints on multimedia learning: When presenting more material results in less understanding. Journal of Educational Psychology, 93(1), 187-198. doi.org/10.1037/0022-0663.93.1.187
  5. Cheng, C., Wu, Y., Wang, R., et al. (2026). Seductive details, cognitive load, and learning outcomes: A multi-level meta-analysis and MASEM. Educational Psychology Review, 38, 28. doi.org/10.1007/s10648-025-10099-z
  6. Sundararajan, N. K., & Adesope, O. O. (2020). Keep it coherent: A meta-analysis of the seductive details effect. Educational Psychology Review, 32, 707-734. doi.org/10.1007/s10648-020-09522-4
  7. Rey, G. D., Beege, M., Nebel, S., Wirzberger, M., Schmitt, T. H., & Schneider, S. (2019). A meta-analysis of the segmenting effect. Educational Psychology Review, 31(2), 389-419. doi.org/10.1007/s10648-018-9456-4
  8. Ginns, P., Martin, A. J., & Marsh, H. W. (2013). Designing instructional text in a conversational style: A meta-analysis. Educational Psychology Review, 25, 445-472. doi.org/10.1007/s10648-013-9228-0
  9. Noetel, M., Griffith, S., Delaney, O., Harris, N. R., Sanders, T., Parker, P., del Pozo Cruz, B., & Lonsdale, C. (2022). Multimedia design for learning: An overview of reviews with meta-meta-analysis. Review of Educational Research, 92(3), 413-454. doi.org/10.3102/00346543211052329
  10. Adesope, O. O., & Nesbit, J. C. (2012). Verbal redundancy in multimedia learning environments: A meta-analysis. Journal of Educational Psychology, 104(1), 250-263. doi.org/10.1037/a0026147

Keep reading