---
title: "Cognitive Load and Slides: What Sweller (1988) Did Not Say"
description: "Sweller (1988) studied problem solving, not slides. This post separates his findings from slide advice, with an invented before-and-after slide."
url: "https://staging.khoi.co.uk/en/blog/tai-nhan-thuc-thiet-ke-thuyet-trinh"
lang: en-GB
author: "Lê Quốc Khôi"
date: 2026-11-02
updated: 2026-11-02
topics: ["The MBA journey"]
tags: ["Cognitive load", "Presentation design", "Instructional design", "Worked examples"]
---

# Cognitive load: designing a talk for the listener

## Key takeaways

- Sweller (1988) argues that solving a problem by working towards a specific goal may use a fairly large share of processing capacity, leaving little for learning to recognise problem types (Sweller, 1988, p. 261). That is how I read cognitive load in the paper.
- The paper backs the idea with a computer simulation, a summary of earlier experiments and a new experiment on 24 Year 10 students solving trigonometry problems (Sweller, 1988, p. 278). The full text contains no slides, lectures or audiences.
- All the slide advice here (ideas per slide, spoken against written words, handouts) is the blog author's suggestion, inferred from the idea that capacity is limited. No source in this post tests it.
- The before-and-after slide for a cold-chain haulier is invented, and so is every figure in it.

Suppose you have 20 minutes in front of a board, and slide three, "Q3 overview", carries nine lines of text. What should the slide carry, and what should you say? This post answers with an idea from John Sweller's 1988 paper, and one boundary: that paper studied how people solve maths problems, not how they listen to talks.

## What does Sweller (1988) claim about cognitive load?

He argues that some ways of solving problems get in the way of learning (Sweller, 1988, p. 257). The method in question is working towards a specific goal by means-ends analysis: the solver compares the current state with the goal, then sets sub-goals to close the gap (Sweller, 1988, p. 258).

His reason is that the solver must hold several things at once: the current state, the goal, the relation between them, the relations between the available operations, and a stack of sub-goals if there is one. The capacity this takes may be so large that little is left for learning, even when the problem is solved (Sweller, 1988, p. 261). What needs learning is a schema: a structure that lets the solver recognise a problem state as belonging to a type that calls for particular moves (Sweller, 1988, p. 259).

The argument rests on three assumptions: capacity is fixed; solving and learning both draw on it; and the two activities use different mechanisms. If all three hold, any extra resource spent on solving must come out of the resource available for learning (Sweller, 1988, p. 275).

What does "cognitive load" mean in the paper? There is no one-sentence definition. This is my reading of pages 261 to 265, not a quotation: the term means how much of the limited processing capacity a method demands. Sweller writes that no direct measure existed, so he compared two methods using a simulation (Sweller, 1988, p. 263). The table below shows the result for a three-step kinematics problem, once with a specific goal and once asking only for as many quantities as possible.

| Measure in the simulation | Specific goal | No specific goal |
|:--|--:|--:|
| Statements in working memory, average | 15.5 | 14 |
| Statements in working memory, peak | 16 | 14 |
| Active "if... then..." rules | 4 | 1 |
| Cycles until solved | 5 | 3 |
| Total conditions matched | 29 | 17 |

These are the figures in Table 2 of the paper. They come from a simulation run on a computer, not from people (Sweller, 1988, p. 272). Table 1 of the paper lists 4 and 1 productions; note 1 says the implemented simulation split them into 10 and 2 (Sweller, 1988, p. 275).

## Does the paper say anything about slides, speech or handouts?

No. Its learners are students working on maths, physics and geometry problems (Sweller, 1988, pp. 262–263). The rest of this post uses the idea that capacity is limited to reason towards advice, and that reasoning belongs to the blog author. The table below keeps the two layers apart.

| The paper (Sweller, 1988) | The blog author's suggestion, not a finding of the paper |
|:--|:--|
| A goal-directed method may use much processing capacity, leaving little for learning (Sweller, 1988, p. 261) | Treat the listener's attention as finite. Do not make people read one text while hearing another. |
| Experts recognise problem types through schemas; novices do not yet (Sweller, 1988, p. 259) | A slide that is easy for a finance director may be heavy for a new hire. Write for the newcomer and put specialist detail in a handout. |
| If conventional problems impose a heavy load without helping learning, they may be replaced by goal-free problems or worked examples (Sweller, 1988, pp. 283–284) | In a workshop, show a worked example before setting an exercise. |
| Not covered | One message per slide. Spoken and written words should not duplicate each other. Reference detail goes in a handout. |

## What does one invented slide look like before and after the edit?

**A hypothetical example.** Rạch Mây Cold Chain is a firm that hauls chilled seafood in Cần Thơ. The firm's name and every figure below are my own inventions, to show the method; they are not data from any company. The presenter asks the board to approve two new refrigerated trucks at VND 3.2bn.

The "before" slide is titled "Cold-chain overview, Q3" and has nine lines: revenue of VND 12.4bn, up 9 per cent on Q2; fuel cost up 14 per cent; 38 per cent of trips running empty on the return leg; the 2019 trucks off the road for 11 days in total; three large orders delivered late; Customer A cutting orders by 20 per cent; a proposal to buy two trucks; estimated payback of 31 months; see appendix.

Nine lines are nine separate ideas, and the listener cannot tell which one to remember. The "after" slide keeps one: the old trucks were off the road for long stretches and made orders late.

> **Figure:** (compare) An invented slide with made-up figures, for illustration only. It is not research evidence or data from any company.

```json
{
  "type": "compare",
  "columns": ["Before", "After"],
  "rows": [
    { "label": "Title", "cells": ["Cold-chain overview, Q3", "The 2019 trucks were off the road for 11 days in Q3, making three large orders late"] },
    { "label": "Text on the slide", "cells": ["Nine lines, a different figure on each", "Nothing beyond the title"] },
    { "label": "Picture", "cells": ["None", "Two-bar chart: days off the road for the 2019 fleet (11) and the 2023 fleet (0)"] },
    { "label": "Spoken", "cells": ["Reads each line aloud", "Tells it in order: which truck failed when, which orders slipped, how the customer reacted. Fuel and empty running only if someone asks"] },
    { "label": "Handout", "cells": ["Appendix at the end of the deck", "One page: revenue, fuel, empty running, the 31-month payback assumption"] },
    { "label": "Next slide", "cells": ["None", "The ask: two trucks, VND 3.2bn"] }
  ]
}
```

Two things changed. The title is now a sentence with a subject and a verb, so the listener knows what to take away. The lines that were dropped moved into speech or the handout. This is a design choice nobody has measured, and the idea behind it is untested too: that listeners should not have to read nine lines and listen to you at the same time.

## What goes on the slide, in the talk, or in the handout?

The table below is the blog author's suggestion. None of its thresholds comes from research.

| Type of content | Put it in | Works if | Does not work if |
|:--|:--|:--|:--|
| The message, one sentence | The slide title | Each slide has one thing the listener must remember | The slide is built to be read later |
| One piece of evidence | A picture or one large figure | The evidence answers the title directly | You need all twelve figures to persuade |
| The argument, the order of telling | Speech | You can say it in a few sentences | The listener must look back at a number while you speak |
| Reference detail | Handout | The listener only needs it if they ask for more | The decision hangs on a figure that appears only there |

On the number of ideas: a suggested starting limit is one message and one piece of evidence per slide, and if you need more, split the slide.

On audiences of mixed expertise: explain a term aloud the first time you use it, and add a line to the handout such as "if you already know this, skip page 2". The underlying idea is that experts and novices recognise a problem differently (Sweller, 1988, p. 259); applying it to talks is the blog author's inference.

## Should workshop exercises have a goal or not?

This is the place where the paper draws its clearest practical implication, and even so it stretches beyond what the paper tested. Sweller proposes that if conventional problems impose a heavy load without helping learning, they may be replaced by goal-free problems or worked examples, with conventional problems kept for tests and perhaps as motivation (Sweller, 1988, pp. 283–284). In the earlier experiments he summarises, "no specific goal" meant swapping a question such as "what is the racing car's acceleration?" for an instruction to calculate as many quantities as possible (Sweller, 1988, p. 263).

A hypothetical case for a workshop on reading financial statements (the table and figures are invented): instead of "Calculate the break-even point of product A", give a four-row table of costs and sales and ask "Work out as many figures as you can from this table". This works if the aim of the session is for participants to recognise problem types. It does not work if you need to train decisions against one specific question: the paper keeps goal-directed problems for testing and says nothing about training decisions.

## Going deeper: what did the paper test, and where are its limits?

Sweller builds on research into experts and novices, from chess to physics problems (Sweller, 1988, pp. 258–259). The studies he cites, such as Owen and Sweller (1985), I know only through his summary in this paper; by that summary, the group solving conventional problems made four to six times as many mathematical errors per side calculated (Sweller, 1988, p. 276).

The paper's own experiment used 24 Year 10 students at a Sydney school, 12 per group, solving six trigonometry problems. After each problem, students had to redraw the previous problem's diagram and its solution (Sweller, 1988, p. 278). The two groups did not differ in total solution time or in mathematical errors, although the goal-free group was slightly faster on all six problems (Sweller, 1988, pp. 279–280). The goal-free group made significantly fewer errors when reproducing angle positions, side positions, side values and the solution, but did not differ on segment labels or angle values (Sweller, 1988, p. 281). Sweller notes that the result on the main task did not replicate earlier studies (Sweller, 1988, p. 282).

The author states his own limits: the model is deliberately minimal and not meant as a full description of problem-solving behaviour (Sweller, 1988, p. 264). He also says the emphasis on conventional problem solving rests on tradition more than on research ("based more on tradition than on research findings") (Sweller, 1988, p. 283).

Carrying the idea over to talks, the limits are plainer: a small sample, school students, a maths task, and nobody sitting and listening to another person speak. Not checked: research after 1988 and any literature on slides. This post rests on one source.

This site's glossary entry for [cognitive load](https://staging.khoi.co.uk/en/glossary#cognitive-load) uses Sweller's 1988 sense, as this post does.

The source was read in full through a PDF copy of the Wiley version, not through the publisher's own page. Page numbers are the ones printed on each page of that copy. The PDF's text layer has recognition errors, so the post uses only figures that read clearly.

## Frequently asked questions

### Does Sweller (1988) say anything about how to build slides?

No. The paper is about students solving maths, physics and geometry problems, and mathematics is the stated focus (Sweller, 1988, pp. 262–263). The full text never mentions slides, lectures, audiences or handouts. Every piece of slide advice in this post is the blog author's inference from the idea that processing capacity is limited.

### How many ideas should a slide carry?

The paper gives no number for slides. The author's suggestion is one message and one piece of evidence per slide, with everything else moved to speech or a handout. That is a working convention, not a threshold drawn from research.

### Whom should I write the slides for when the audience varies in expertise?

Sweller writes that experts recognise problem types through schemas, while novices lack them and fall back on general strategies (Sweller, 1988, p. 259). The author's inference: write the slide for the newcomer, and put specialist detail in a handout.

### How strong is the evidence in the paper?

The paper's own experiment had 24 Year 10 students in two groups of 12, solving six trigonometry problems (Sweller, 1988, p. 278). On the main task, the results did not replicate earlier studies (Sweller, 1988, p. 282). The word 'tentatively' attaches only to the idea of using computational models as measuring devices; the second conclusion, he says, may be put more strongly, and the closing conclusions use 'may' repeatedly (Sweller, 1988, pp. 283–284).

## References

- Sweller, J. (1988) ‘Cognitive load during problem solving: effects on learning’, *Cognitive Science*, 12(2), pp. 257–285. https://doi.org/10.1207/s15516709cog1202_4

---

Canonical HTML version: https://staging.khoi.co.uk/en/blog/tai-nhan-thuc-thiet-ke-thuyet-trinh