Abstract

The same information can be easy or hard to use depending on how it is laid out. This article reviews four bodies of evidence. First, diagrams organise information by location, making relations explicit that text leaves implicit, and learning with node-link diagrams such as concept maps is associated with better retention (Larkin & Simon, 1987; Nesbit & Adesope, 2006; Schroeder et al., 2018). Second, people read some graphical encodings more accurately than others, and graphs suit spatial tasks such as seeing trends while tables suit tasks that need exact values (Cleveland & McGill, 1984; Vessey, 1991). Third, working memory is limited, and placing related words and pictures together and adding cues reduce load and support learning (Mayer & Moreno, 2003; Ginns, 2006; Xie et al., 2017). Fourth, an overview map and a route presentation support different kinds of spatial knowledge (Münzer et al., 2012). On this evidence, Mars draws flowcharts, mind maps, sequence diagrams, timelines and data charts as the answer streams, sets key deck metrics as big-number headlines with charts and tables beside them, and lays multi-day trips on an interactive map. We also state the limits: the evidence comes mostly from learning studies, effects depend on the reader’s expertise, decoration can distract, and it supports the design direction without being a test of Mars itself.

Keywords: diagrammatic representation · graphical perception · cognitive load · concept maps · cognitive fit · signaling · spatial knowledge · AI charts · data-driven slides

1. Introduction

Two representations can hold exactly the same information and still differ in how much work it takes to use them. Larkin and Simon showed this formally: when a diagram and a text are informationally equivalent, their computational efficiency still differs, because a diagram indexes information by location in a plane and often makes explicit what a sequence of sentences leaves to be inferred (Larkin & Simon, 1987). The cognitive theory of multimedia learning adds that people process pictorial and verbal material in separate channels, each limited in how much it can handle at one time (Mayer & Moreno, 2003).

Much everyday work is still delivered as long text: an answer that explains a process in paragraphs, a report that lists its figures as bullet points. This article asks when a diagram, a chart or a map helps people understand and use information, when it does not, and how those answers translate into the design of an assistant that answers questions and builds decks.

Method: this is a narrative literature review. We favoured peer-reviewed work accessible through a DOI, preferred meta-analyses and controlled experiments, and included theoretical analyses only where they define a widely used concept. Every factual statement carries its source, and the full reference list with original links follows the article.

2. Diagrams make relations visible

In Larkin and Simon’s analysis, the advantage of a diagram lies mainly in search and explicitness: information needed for an inference is often present at a single location, and cues to the next step sit at an adjacent location, so problem solving can proceed as a smooth traversal of the diagram with little search (Larkin & Simon, 1987). A later review of visual-spatial displays reaches a measured conclusion: cognitive science has characterised performance well for well-defined tasks with relatively simple displays, while many challenges remain for complex displays used in ill-defined tasks (Hegarty, 2011).

Meta-analyses of node-link diagrams give outcome-level evidence. Across 55 studies with 5,818 participants from Grade 4 to postsecondary, constructing, modifying or viewing concept maps was associated with increased knowledge retention, with effects ranging from small to large and significant heterogeneity (Nesbit & Adesope, 2006). A later synthesis of 142 independent effect sizes (n = 11,814) found a moderate overall effect (g = 0.58); creating concept maps (g = 0.72) was associated with greater benefit than studying them (g = 0.43), and the effect held in STEM and non-STEM domains (Schroeder et al., 2018).

Based on this evidence, when an answer is about structure, steps or order, Mars does not leave it as a wall of text: data charts and structured diagrams (flowcharts, mind maps, sequence diagrams and timelines) draw while the answer streams, with key insights and follow-up questions, and re-flow to a vertical layout on narrow screens so they read just as well on a phone (Larkin & Simon, 1987; Nesbit & Adesope, 2006).

3. Some charts are read more accurately than others

Cleveland and McGill identified a set of elementary perceptual tasks people carry out when they extract quantities from graphs and ordered those tasks by how accurately people perform them; their guideline is that graphs should use tasks as high in that ordering as possible, and on this basis they argued that popular forms, pie charts among them, needed “radical surgery” (Cleveland & McGill, 1984). Heer and Bostock replicated earlier spatial-encoding experiments with crowdsourced participants and reported that such perception experiments are viable at scale (Heer & Bostock, 2010).

The verdict on particular chart types is less settled than design guidelines often suggest. In a series of experiments on judging proportions, Spence and Lewandowsky found little to choose between pie and bar charts, with pie charts slightly ahead for more complicated judgments, and both forms superior to a table; they concluded that the traditional prejudice against the pie chart is misguided (Spence & Lewandowsky, 1991). Cognitive fit theory explains why neither graphs nor tables win across the board: graphs emphasise spatial information and tables emphasise symbolic information, and performance improves when the representation matches the task, graphs for spatial tasks and tables for symbolic ones (Vessey, 1991).

Following cognitive fit, Mars separates the two jobs when it builds a deck: it draws monthly figures as a bar, line or pie chart in the Mars chart style, and puts values that must stay exact into a table (Vessey, 1991; Spence & Lewandowsky, 1991). Every number comes from what you gave it or records it read; Mars never invents figures to fill a page.

4. Less load, clearer cues

Cognitive load theory begins from the observation that some activities take up so much processing capacity that little remains for building the knowledge structures, or schemas, that learning requires (Sweller, 1988). Mayer and Moreno define cognitive overload as intended processing that exceeds the learner’s available capacity, describe five overload scenarios, and test theory-based ways to reduce load in each (Mayer & Moreno, 2003).

Two of those remedies have meta-analytic support. Across 50 independent studies, reducing split attention by placing related elements of information close together in space or time produced learning gains, substantial ones for complex material in particular (Ginns, 2006). A meta-analysis of cueing, signals that direct attention to the key elements of multimedia material, found that cues slightly lowered subjective cognitive load (d = −0.11) and improved retention (d = 0.27) and transfer (d = 0.34), and the more cues reduced load, the better learners performed (Xie et al., 2017).

Mars applies both ideas. In a deck, it sets two to four headline metrics in large type on one slide, with the charts and tables beside them, so the key number works as a cue and its evidence sits next to it (Xie et al., 2017; Ginns, 2006). In a conversation, a chart arrives inside the answer together with its key insights rather than on a separate page (Ginns, 2006). Turning a meeting recording into a report deck, Mars lays out the meeting summary, key decisions, action items, each side’s views, risks and open questions, and leaves out a section with nothing to say rather than padding it (Rey, 2012).

5. Maps for the whole route

Map research distinguishes knowing a route from knowing the layout of an area. In two experiments in which high-school students took an assisted tour of an unfamiliar university campus (n = 84 and n = 48), Münzer and colleagues found a trade-off: presentations that gave comprehensive configural information supported learning the layout at the cost of accurate wayfinding, while a route presentation supported wayfinding at the cost of layout knowledge (Münzer et al., 2012).

Relying on GPS guidance may also have a cost when it becomes a habit. Among 50 regular drivers, those with more lifetime GPS experience had worse spatial memory when they had to navigate without it, and in 13 participants retested three years later, more GPS use was associated with a steeper decline in hippocampal-dependent spatial memory (Dahmani & Bohbot, 2020). The longitudinal sample was small, so this is a caution rather than an established effect.

Based on the distinction between overview and route, Mars treats planning and travelling as different tasks. Planning a trip, Mars renders an interactive map in the conversation: switch days to see each day’s stops and route, tap a place for photos, ratings and address, and open the route in your maps app when you set off (Münzer et al., 2012).

6. From evidence to design

These findings point to one conclusion: a visual is useful when its form fits the task, keeps related information together and points attention to what matters, and not merely because it is a picture. Mars meets each requirement as follows:

1. Diagrams make relations explicit → diagrams that draw as the answer streamsFlowcharts, mind maps, sequence diagrams, timelines and data charts draw while the answer streams, with key insights and follow-up questions, and re-flow to a vertical layout on narrow screens. Steps, structure and order appear where they can be seen at once instead of being reconstructed from paragraphs (Larkin & Simon, 1987; Nesbit & Adesope, 2006; Schroeder et al., 2018).
2. Graphs for patterns, tables for exact values → charts and tables side by sideMars draws monthly figures as a bar, line or pie chart in the Mars chart style and puts values that must stay exact into a table. Every number comes from what you gave it or records it read, and charts and tables stay editable shapes and text in PowerPoint (Vessey, 1991; Cleveland & McGill, 1984; Spence & Lewandowsky, 1991).
3. Cues and contiguity lower load → big-number headlines with their evidence beside themGive Mars your figures and the key metrics become bold big-number headlines, two to four on one slide, with charts and tables beside them; in a conversation, a chart arrives with its key insights inside the answer (Xie et al., 2017; Ginns, 2006).
4. Overview and route serve different tasks → an interactive trip map, then your maps appMulti-day trips land on an interactive map: switch days to see each day’s stops and route, tap a place for photos, ratings and address, and open the route in your maps app when you set off (Münzer et al., 2012).

7. Limitations

First, the studies cited here test how people learn from and read visual displays, not Mars itself. Most outcome evidence comes from learning settings, such as students from Grade 4 to postsecondary (Nesbit & Adesope, 2006) or high-school students on a campus tour (Münzer et al., 2012); perception results come partly from crowdsourced participants (Heer & Bostock, 2010); and some cited works are theoretical analyses rather than new experiments (Larkin & Simon, 1987; Vessey, 1991). The GPS finding rests on a cross-sectional sample of 50 drivers and a longitudinal sample of 13 (Dahmani & Bohbot, 2020). Reading a chart at work is not the same as a post-test, and the product’s effect needs its own evaluation.

Second, the benefit of a visual depends on who reads it. Techniques that are highly effective with inexperienced learners can lose their effectiveness and even have negative consequences with more experienced learners, the expertise reversal effect (Kalyuga et al., 2003). A diagram that helps a newcomer may be redundant for an expert, and much remains unknown about complex displays used for ill-defined tasks (Hegarty, 2011).

Third, decoration is contested. A meta-analysis of 39 experimental effects found that interesting but irrelevant additions, including illustrations, reduce retention and transfer, with small-to-medium and medium effects (Rey, 2012); yet in one experiment, embellished charts were described no less accurately than plain ones and were recalled better after two to three weeks (Bateman et al., 2010). Mars decks give every slide a picture in one consistent style, so this design choice should be judged against that evidence rather than assumed to help.

Fourth, popular claims about visuals are often stronger than the evidence. Larkin and Simon’s own title says a diagram is sometimes worth ten thousand words (Larkin & Simon, 1987); the case against pie charts is disputed (Cleveland & McGill, 1984; Spence & Lewandowsky, 1991); and the well-known modality effect concerns graphics paired with spoken narration and is strongest when the presentation is system-paced (Ginns, 2005), conditions that do not describe a chart read at one’s own pace, so we do not rely on it.

8. Conclusion

A diagram is not better than text by default. The research shows that visuals help when they make relations explicit, match the form to the task, keep related information together and cue what matters, and that their value depends on the reader and on what is left out. Mars diagrams and charts drawn with the answer, deck slides that pair big-number headlines with charts and tables, and interactive trip maps are designed on this evidence.

References

Listed in APA 7th edition style, each with its original DOI link.

1. Bateman, S., Mandryk, R. L., Gutwin, C., Genest, A., McDine, D., & Brooks, C. (2010). Useful junk? The effects of visual embellishment on comprehension and memorability of charts. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 2573–2582. https://doi.org/10.1145/1753326.1753716
2. Cleveland, W. S., & McGill, R. (1984). Graphical perception: Theory, experimentation, and application to the development of graphical methods. Journal of the American Statistical Association, 79(387), 531–554. https://doi.org/10.1080/01621459.1984.10478080
3. Dahmani, L., & Bohbot, V. D. (2020). Habitual use of GPS negatively impacts spatial memory during self-guided navigation. Scientific Reports, 10, 6310. https://doi.org/10.1038/s41598-020-62877-0
4. Ginns, P. (2005). Meta-analysis of the modality effect. Learning and Instruction, 15(4), 313–331. https://doi.org/10.1016/j.learninstruc.2005.07.001
5. Ginns, P. (2006). Integrating information: A meta-analysis of the spatial contiguity and temporal contiguity effects. Learning and Instruction, 16(6), 511–525. https://doi.org/10.1016/j.learninstruc.2006.10.001
6. Heer, J., & Bostock, M. (2010). Crowdsourcing graphical perception: Using Mechanical Turk to assess visualization design. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 203–212. https://doi.org/10.1145/1753326.1753357
7. Hegarty, M. (2011). The cognitive science of visual-spatial displays: Implications for design. Topics in Cognitive Science, 3(3), 446–474. https://doi.org/10.1111/j.1756-8765.2011.01150.x
8. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38(1), 23–31. https://doi.org/10.1207/S15326985EP3801_4
9. Larkin, J. H., & Simon, H. A. (1987). Why a diagram is (sometimes) worth ten thousand words. Cognitive Science, 11(1), 65–100. https://doi.org/10.1111/j.1551-6708.1987.tb00863.x
10. Mayer, R. E., & Moreno, R. (2003). Nine ways to reduce cognitive load in multimedia learning. Educational Psychologist, 38(1), 43–52. https://doi.org/10.1207/S15326985EP3801_6
11. Münzer, S., Zimmer, H. D., & Baus, J. (2012). Navigation assistance: A trade-off between wayfinding support and configural learning support. Journal of Experimental Psychology: Applied, 18(1), 18–37. https://doi.org/10.1037/a0026553
12. Nesbit, J. C., & Adesope, O. O. (2006). Learning with concept and knowledge maps: A meta-analysis. Review of Educational Research, 76(3), 413–448. https://doi.org/10.3102/00346543076003413
13. Rey, G. D. (2012). A review of research and a meta-analysis of the seductive detail effect. Educational Research Review, 7(3), 216–237. https://doi.org/10.1016/j.edurev.2012.05.003
14. Schroeder, N. L., Nesbit, J. C., Anguiano, C. J., & Adesope, O. O. (2018). Studying and constructing concept maps: A meta-analysis. Educational Psychology Review, 30(2), 431–455. https://doi.org/10.1007/s10648-017-9403-9
15. Spence, I., & Lewandowsky, S. (1991). Displaying proportions and percentages. Applied Cognitive Psychology, 5(1), 61–77. https://doi.org/10.1002/acp.2350050106
16. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. https://doi.org/10.1207/s15516709cog1202_4
17. Vessey, I. (1991). Cognitive fit: A theory-based analysis of the graphs versus tables literature. Decision Sciences, 22(2), 219–240. https://doi.org/10.1111/j.1540-5915.1991.tb00344.x
18. Xie, H., Wang, F., Hao, Y., Chen, J., An, J., Wang, Y., & Liu, H. (2017). The more total cognitive load is reduced by cues, the better retention and transfer of multimedia learning: A meta-analysis and two meta-regression analyses. PLoS ONE, 12(8), e0183884. https://doi.org/10.1371/journal.pone.0183884

Cite this article

Mars Research. (2026). Why a diagram is sometimes worth more than a paragraph. MarsDataAI. https://marsdata.ai/research/seeing-with-diagrams