Can a Font Lie: experiments on the persuasive power of typographyCan a Font Lie: experiments on the persuasive power of typography

In August 2012, the documentary filmmaker Errol Morris published a short quiz in The New York Times titled "Are You an Optimist or a Pessimist?" Readers read a paragraph by the physicist David Deutsch, then marked whether they agreed with the proposition that humanity could be protected from an asteroid strike. What nobody was told was that the paragraph was shown in one of six typefaces chosen at random: Baskerville, Computer Modern, Georgia, Helvetica, Trebuchet, and Comic Sans. About 45,000 people took part. The result: significantly more people agreed with the same sentence when it was set in Baskerville; Comic Sans was the least believed typeface. The difference was small, but across 45,000 people it was too consistent to be explained by chance. Morris called this "the Baskerville effect." The typeface did not change what the text said; it changed how much of it was believed.

Morris's observation rests on a mechanism well known in psychology. In 2008, in an experiment published in Psychological Science, Hyunjin Song and Norbert Schwarz gave participants the same exercise instructions in two typefaces: plain Arial and Brush Script, which imitates handwriting. Those who read them in Arial estimated the exercise would take about eight minutes; those who read them in Brush Script, fifteen. Same text, same movements. Type that was hard to read had created the feeling that the task itself was hard; the participants were also less willing to add that exercise to their daily routine. In a second experiment in the same paper, a sushi recipe presented in the difficult typeface was judged to "require more skill."

The processing fluency principle, named by Reber, Schwarz, and Winkielman in 2004, operates here too: the brain treats whatever it processes easily as true, easy, and trustworthy. Choosing a corporate typeface is therefore not a question of "which one is beautiful"; the right question is "which one gives the customer's brain the least work." A heading that is hard to read makes the service beneath it look hard as well.

In 2011, Diemand-Yauman, Oppenheimer, and Vaughan published an intriguing finding in Cognition: when course material was presented in "disfluent" typefaces such as Comic Sans Italic or Bodoni MT, students did better on the exam. Difficulty raised attention, and attention strengthened memory. The idea was so attractive that in 2018 RMIT University announced to the world Sans Forgetica, a typeface with parts of its letters deliberately left out, as "the font that boosts memory"; the news was shared hundreds of thousands of times.

Two years later, Taylor, Sanson, Burnell, Wade, and Garry tested Sans Forgetica in four experiments in the journal Memory: it was harder to read, yes; but it did not improve recall, and under some conditions it even reduced it. Diemand-Yauman's original finding could not be replicated in larger samples either. Today's scientific consensus is clear: tiring the reader teaches the reader nothing. A brand that picks a hard-to-read typeface "to get noticed" is investing in a finding from 2011 that no longer holds.

"Serif type reads better on paper, sans serif on screen" is a sentence still drilled into students at design schools. In 1999, in his doctoral thesis at the University of Reading, Ole Lund examined one by one 72 legibility studies conducted since 1896 and found no solid evidence that serifs increased or decreased legibility; most of the studies were methodologically weak. The difference lay not in the serif but in x-height, letter spacing, line length, and contrast.

In 2004, Microsoft's legibility researcher Kevin Larson broke another piece of received wisdom: the theory that "we recognize words by their shape." This view, which goes back to Cattell in 1886, attributed the faster reading of lowercase text over uppercase to the outline of the word. Eye-movement research showed something else: the reader decodes a word by recognizing its letters in parallel; uppercase text reads slowly not because the shape is lost, but because of habit and because the letters resemble one another. The thirty years of measurements Miles Tinker collected in Legibility of Print in 1963 still give the valid limits today: text set entirely in capitals slows reading noticeably; 45 to 75 characters per line read comfortably; as Bringhurst wrote in 1992, 66 characters is close to ideal.

The most expensive legibility experiment was conducted on highways. In the mid-1990s, Garvey, Pietrucha, and Meeker of Penn State University tested the newly designed Clearview against Highway Gothic, the typeface American highway signs had used for fifty years, in both nighttime and daytime conditions. Clearview's open counters and larger x-height increased nighttime recognition distance by about sixteen percent, especially among older drivers: at the same letter height, a driver could read the sign several car lengths earlier. The Federal Highway Administration gave Clearview interim approval in 2004; thirty states changed their signs. In 2016 the approval was withdrawn on the grounds that later studies had produced mixed results; in 2018 it was granted again. The lesson of the story is not which typeface won, but that legibility is a quantity that can be measured in meters.

In 2004, in the Journal of Business Research, John Doyle and Paul Bottomley set the same product's name in different typefaces and had participants choose. The brand presented in the typeface judged "appropriate" to the product category was chosen significantly more often, even though it was the same product. Appropriateness is the overlap between the formal character of the typeface and the promise of the product: soft letters for a chocolate, hard-cornered ones for an engineering firm. The typeface says something about the product before the text is read.

This is why corporate typography is not the choice of a single font but a system. When Adrian Frutiger designed Univers in 1957, he coded its weights and widths with two-digit numbers: 55 regular, 65 bold, 75 black, 57 condensed. For the first time a typeface was conceived not as a "family" but as a coordinate system. The same year, in Zurich, Miedinger and Hoffmann drew Neue Haas Grotesk; in 1960 it took the name Helvetica. Both are still in use today, because both delivered not a single beautiful letter but a system.

Among the languages written in the Latin alphabet, Turkish is one of those that punishes typeface choice most harshly, and the reason is a single dot. The capital of the dotted i is İ, the capital of the dotless ı is I, and most typefaces apply this distinction correctly only if OpenType's locl (localized forms) feature has been defined for Turkish. If it has not, "İSTANBUL" becomes "ISTANBUL" when converted to capitals; in the word "fiyat" the fi ligature swallows the dot and the word reads as something else. In 2008 in Turkey, a mobile phone's lack of Turkish character support meant that the word "sıkışınca" was written without its dots and read as a different word, and a family argument ended in death; for years the technology press retold the incident as "the cost of a missing dot." In Turkish, typography is not decoration; it is the safety of meaning.

That is why, before we accept a corporate typeface, we test it with three words: İSTANBUL (capital İ), fiyat (the fi ligature), ÖĞRENCİ (the ratio of the accents to the x-height). A typeface that fails any of the three cannot enter the identity, however beautiful it may be.

In September 2016, at the ATypI conference in Warsaw, something rarely seen in the history of typography happened: Adobe, Apple, Google, and Microsoft announced OpenType 1.8 on the same stage. The novelty was the "variable font": a typeface that carries axes such as weight, width, slant, and optical size as continuous values in a single file. A website that until then had loaded twelve separate files for one family could now produce every intermediate weight from a single file.

From a design standpoint, the real gain is the optical size axis. In the age of metal type, the 8-point and 72-point cuts of a typeface were drawn differently; at small sizes the letters were more open and the spacing wider, at large sizes the fine details sharper. The digital age forgot this knowledge for a hundred years; the opsz axis of the variable font brought it back. If a corporate typeface has this axis today, the same family stands on a business card and on a facade sign with the correct stroke weight.

We turned these findings into a selection protocol. A candidate typeface first goes through Song and Schwarz's measure: body text is read at 16 pixels on screen and 9 points in print, with the line set to 66 characters; a candidate that lowers reading speed is eliminated. Then the Turkish test: İSTANBUL, fiyat, ÖĞRENCİ. Then Doyle and Bottomley's appropriateness question: do the three adjectives that describe the brand's promise also suit the typeface; if they do not, its beauty is irrelevant. Finally the system question: does the family have at least four weights or one variable axis, can the print and screen licenses be bought together, is the x-height compatible with the secondary font.

Morris's 45,000 readers told us this: the typeface is the invisible coefficient of credibility. Every font chosen without measuring that coefficient adds an unknown multiplier to every sentence the brand writes.

Morris, E. (2012). Hear, All Ye People; Hearken, O Earth. The New York Times, Opinionator, August 8–9, 2012 (analysis: David Dunning).

Song, H. & Schwarz, N. (2008). If it's hard to read, it's hard to do: Processing fluency affects effort prediction and motivation. Psychological Science, 19(10), 986–988.

Reber, R., Schwarz, N. & Winkielman, P. (2004). Processing fluency and aesthetic pleasure. Personality and Social Psychology Review, 8(4), 364–382.

Diemand-Yauman, C., Oppenheimer, D. M. & Vaughan, E. B. (2011). Fortune favors the bold (and the italicized). Cognition, 118(1), 111–115.

Taylor, A., Sanson, M., Burnell, R., Wade, K. A. & Garry, M. (2020). Disfluent difficulties are not desirable difficulties: The (lack of) effect of Sans Forgetica on memory. Memory, 28(7), 850–857.

Lund, O. (1999). Knowledge construction in typography: The case of legibility research and the legitimisation of sans serif typefaces. Doctoral thesis, University of Reading.

Larson, K. (2004). The science of word recognition. Microsoft Typography.

Tinker, M. A. (1963). Legibility of Print. Iowa State University Press.

Bringhurst, R. (1992). The Elements of Typographic Style. Hartley & Marks.

Garvey, P. M., Pietrucha, M. T. & Meeker, D. (1997). Effects of font and capitalization on legibility of guide signs. Transportation Research Record, 1605, 73–79; FHWA Clearview interim approval (2004), withdrawal (2016), and reinstatement (2018).

Doyle, J. R. & Bottomley, P. A. (2004). Font appropriateness and brand choice. Journal of Business Research, 57(8), 873–880.

OpenType 1.8 announcement, ATypI Warsaw, September 14, 2016 (Adobe, Apple, Google, Microsoft).