"Add More Visuals" Has No Support in the Research That Studied It

I set out to add figures to a corpus of long-form articles, and before doing it I went looking for the evidence that figures help. What I expected to find was a body of work quantifying the lift. What I found was two of the most-cited papers in the field warning that the lift had never been established.

Scaife and Rogers, writing in the International Journal of Human-Computer Studies, put it flatly: "little is known about the cognitive value of any graphical representations, be they good old-fashioned (e.g. diagrams) or more advanced" — and their review of the literature "reveals a fragmented and poorly understood account of how graphical representations work, exposing a number of assumptions and fallacies" (Scaife & Rogers, 1996).

Six years later Tversky, Morrison and Bétrancourt reached the same edge from a different direction: "research on static graphics has shown that only carefully designed and appropriate graphics prove to be beneficial for conveying complex systems" (Tversky et al., 2002).

What the record actually supports

Dismissing the whole field would be the opposite error. The literature is not silent — it is specific, and what it specifies is narrower than a count.

Format has to match the concept. Tversky's team named this the Congruence Principle: "the content and format of the graphic should correspond to the content and format of the concepts to be conveyed" (Tversky et al., 2002). A narrowing drawn as a bar chart fails it. So does a sequence drawn out of order, or a two-by-two where nothing is marked as the answer.

Some encodings are read more accurately than others. Cleveland and McGill established an ordering of elementary perceptual tasks "on the basis of how accurately people perform them" (Cleveland & McGill, 1984). Position and length sit near the top; area, angle and colour saturation sit lower. This is why a bar chart usually beats a pie, and why encoding a quantity in the size of a circle is a choice you should have to defend.

Words belong next to the picture. Moreno and Mayer measured a spatial-contiguity effect: "students learned better when visual and verbal materials were physically close" (Moreno & Mayer, 1999). A caption on the far side of a page break is not contiguity.

The words are what people remember. Borkin and colleagues ran what was then the largest visualization study of its kind, across 2,070 single-panel visualizations drawn from news media, government reports, scientific journals and infographic sources (Borkin et al., 2013). Titles and text carry more of the recall than the chart type does — which makes a generic label above a figure a wasted asset.

A diagram earns its place by making the implicit explicit. Larkin and Simon described diagrammatic representations as "indexed by location in a plane", displaying information "that is only implicit in sentential representations and that therefore has to be computed, sometimes at great cost, to make it explicit for use" (Larkin & Simon, 1987). A figure that restates the sentence beside it has added search cost and nothing else.

Two registers of the same advice, and they do not join

What the playbook says

What the record says

• Add more visuals — aim for one per sectionCongruence: only where the format matches the concept (Tversky et al., 2002)
• Any chart beats a paragraphAccuracy: encodings are ordered by how accurately they are read (Cleveland & McGill, 1984)
• Infographics are more memorableRecall: the words carry more of it than the chart type does (Borkin et al., 2013)
• Visuals aid comprehensionUnestablished: "little is known about the cognitive value of any graphical representations" (Scaife & Rogers, 1996)

Cox's qualifier, which is why this is a test and not a house style

Cox adds the constraint that stops all of the above from becoming a template. Effective reasoning with an external representation is "a three-way interaction" between the properties of the representation, "the match between the demands of the task and the type of information read-off afforded by the representation", and reader factors including prior knowledge and cognitive style (Cox, 1999).

That is why the useful output of this literature is a set of shape-fit tests, not a style guide. What works depends on the argument being made and the reader making it. A rule that says "one figure per section" cannot know either.

What I did with it

I turned the five findings above into five checks and ran them across the corpus: congruence and encoding-accuracy as hard failures, contiguity, title-carries-the-finding and locality as advisories. As of 2 September 2026 that is 173 files and 80 figures, this article's own two included Sagentix GTM Methodology, 2026. The count moves whenever the corpus does, which is the point: it is an inventory, not a target.

Two things came out of it that I did not expect. The first is that the checks worth automating are the ones about honesty, not taste — a bar drawn longer than another whose printed number is smaller is a false picture, and no citation fixes it. The second is that the count never mattered. Several articles improved by having a figure removed, because the figure restated a table that was already there, and Larkin and Simon explain exactly why that costs the reader.

The uncomfortable part: if you have been briefing agencies or writers on a visual target, the research does not back the brief. It backs a different one.

Where Sagentix fits

Sagentix builds go-to-market evidence for organizations that have to survive somebody else's review — a board, an investor, a regulator, an examining firm. Discipline about figures is the same discipline as discipline about citations: a chart is a claim, and it fails independently of whether the number beside it is true.

The research library behind that work runs to 1,414 catalogued artifacts, every deliverable passes an 18-check quality gate before it reaches a client, and full engagements run six to eight weeks at CA$4,500 to CA$45,000 depending on scope. Phase 1 carries a money-back guarantee, subject to terms.

Three ways to use this

Audit what you have, free, in an afternoon. Take your last ten published pieces and ask two questions of every figure: does the shape match the argument, and does the drawing contradict any number printed on it. You need no tooling. The second question alone usually finds something, and it is the one that costs credibility when a reader catches it first.

Rewrite the brief instead of the assets. If you commission content, replace any figure-count target with a shape-fit test. "Every figure must make explicit something the prose leaves implicit" is a brief a writer can meet and an editor can check. "Three visuals per article" is neither.

Bring in help when the audience is adversarial. Worth it when the reader is a diligence partner or an examiner rather than a subscriber — when a wrong chart is not an aesthetic problem but a finding. Ask anyone you evaluate to show you the check they run, not the portfolio they show.

References

• Borkin, M. A., Vo, A. A., Bylinskii, Z., Isola, P., Sunkavalli, S., Oliva, A., & Pfister, H. (2013). What makes a visualization memorable? IEEE Transactions on Visualization and Computer Graphics, 19(12), 2306–2315.
• 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.
• Cox, R. (1999). Representation construction, externalised cognition and individual differences. Learning and Instruction, 9(4), 343–363.
• Larkin, J. H., & Simon, H. A. (1987). Why a diagram is (sometimes) worth ten thousand words. Cognitive Science, 11(1), 65–100.
• Moreno, R., & Mayer, R. E. (1999). Cognitive principles of multimedia learning: The role of modality and contiguity. Journal of Educational Psychology, 91(2), 358–368.
• Scaife, M., & Rogers, Y. (1996). External cognition: How do graphical representations work? International Journal of Human-Computer Studies, 45(2), 185–213.
• Tversky, B., Morrison, J. B., & Bétrancourt, M. (2002). Animation: Can it facilitate? International Journal of Human-Computer Studies, 57(4), 247–262.