A few years ago, the New York Times ran an experiment about fonts where participants would identify if a given statement was true or false.
The quiz was delivered in one of six typefaces: three serifs (Baskerville, Computer Modern or Georgia) or three sans-serifs (Trebuchet, Helvetica or Comic Sans).
With the only variable being the shape of the letters, the New York Times observed whether typefaces influenced the believability of statements. They found that Baskerville, a traditional serif font, made statements the most believable, while Comic Sans, a sans-serif resembling chalkboard writing, was likely to harm believability.
A test of our own
To designers, the New York Times’ results are unsurprising. It’s taught early in design education that serifs are more traditional, trustworthy, and formal with roots going back to the Roman Empire. Their tiny wings also make them far more readable for long paragraphs (like the one you’re reading now) which is why you rarely see books set in a sans-serif typeface.
By contrast, sans-serifs are informal, cleaner, and generally perceived as more accessible (see Google’s recent switch to a friendlier look-how-cute-we-are-with-your-personal-data sans-serif).
So at Yieldmo, our designers and data scientists used our massive mobile ad network which reaches 3 out of every 4 smartphones users in the US (an audience reach comparable to Facebook, Google, Amazon, or Yahoo!) to replicate the experiment. We tested three serif and three sans-serif typefaces on more than 2 million people across the mobile web — a sample size 44 times larger than the New York Times’ and likely larger than any other test run on the subtle influences of these typefaces.
Now, we couldn’t replicate the New York Times test to the letter (pun!). Instead, we tested only mobile-web-safe fonts (Helvetica, Arial, Verdana, Times New Roman, Georgia, and Garamond) so no Baskerville or Comic Sans. We were also measuring clickthrough rates (CTR) in a mobile ad context — very different from the true or false experience of a quiz. The example below shows the fonts on a GEICO ad, but the actual test applied to our total inventory of advertisers ranging the spectrum of industries and placements.
By measuring the change in CTR on our network’s ads, we were able to infer the impact a typeface had on a person’s likelihood to tap a mobile ad.
We found that serifs were among both the best and worst performing typefaces.
Arial (+4.3%), Helvetica (0%), Georgia (+3%), and Verdana (0%) all performed relatively the same (despite Georgia being a serif typeface redesigned specifically for the screen in 1996). It’s possible that with retina displays Georgia’s squared off accommodations for the 72dpi era are now obsolete. We don’t really have a good hypothesis on why Arial and Helvetica performed so differently. To the untrained eye, they usually appear identical. We plan to take a closer look as we expand the test.
The biggest surprise came from Garamond (-15%), a serif typeface from the 16th century. If any font in our group says trust and tradition, it’s Garamond, and yet it hurt CTR! The problem may have been that our test did not account for the typeface’s naturally shorter x-height and so Garamond, although the same pixel size as the other tested fonts, appears smaller and less legible. We plan to run our test again to see if bumping up Garamond’s size helps it out.
The clear winner was Times New Roman (+15%), a serif from 1932 that is ubiquitous on both the page and screen. Our guess for the moment, as we are planning to revise and expand this test, is that Times New Roman has the best of Baskerville’s formality but with superior legibility over obsolete Georgia and high contrast Garamond. As a result of this test, we are considering shifting our mobile ads’ default font from Helvetica to Times New Roman. This could mean a double digit lift in taps all from the subtle magic of serif typefaces.
We’re still trying to resolve a number of unexpected results that came from our experiment and we welcome any insights you might have. Maybe you have a better hypothesis around Arial and Helvetica’s oddly different performance? Or maybe you have other ideas as to why Garamond failed so miserably? We’d love to read your comments.