Processing fluency

Processing fluency is the subjective ease with which information is taken in, and the claim is that people read that feeling as information about the content. Alter and Oppenheimer's review gathered manipulations as different as semantic priming, visual clarity and phonological simplicity under one continuum from effortless to demanding, which is why fluency gets cited for almost any design decision. The evidence underneath is far more uneven than that framing suggests, and it is strongest for the least interesting claims.

The famous claim — easy to read means judged true — has largely collapsed. Reber and Schwarz showed trivia statements in high or low colour contrast, and across the published set only two experiments produced a clear positive result, at d = 0.252 and d = 0.396. Aktepe and Heck ran two online replications, the second preregistered, and found no effect of contrast on truth judgments even though high-contrast statements were read reliably faster, so the manipulation itself worked. Their robust Bayesian meta-analysis of the eight experiments put the effect at Cohen's d = 0.023, with BF01 = 3.66 favouring the null and moderate evidence of publication bias. Making a sentence more legible does not make readers believe it.

What does reliably raise perceived truth is repetition, the other common source of fluency. Fazio, Rand and Pennycook's preregistered experiment found that repetition increased belief at every level of plausibility, including statements that were obviously implausible. That is the dangerous half of this principle, and the reason repetition sits in the ethical line below rather than in the build advice: a familiar, repeatedly surfaced claim gains credibility it has not earned.

Fluency does shift judgments other than truth. Oppenheimer found that heavier lexical complexity lowered the judged intelligence of a text's author regardless of the quality of the original essay, and that the same text set in a hard-to-read font was judged to come from a less intelligent author. His fifth experiment is the useful caveat: when an obvious irrelevant cause of the difficulty is available, such as a bad photocopy, people discount the fluency cue and overcorrect against it, so a visibly broken render does not cost you the way bad writing does.

Low fluency also reads as unfamiliar, and unfamiliar cuts both ways. Song and Schwarz showed that hard-to-read instructions raise predicted effort and lower willingness to adopt a routine, which is the finding closest to onboarding work. Their companion study found hard-to-pronounce food additives rated more harmful, and hard-to-pronounce amusement rides rated both more likely to make you sick and more exciting and adventurous. Expect an awkward name to read as risky, which helps a ride and hurts an ingredient list.

The inverse idea, that deliberate disfluency makes people think harder, has not survived. Diemand-Yauman, Oppenheimer and Vaughan reported that hard-to-read fonts improved memory in the lab and in high-school classrooms, and it became the standard citation for ugly-on-purpose typography. Meyer and colleagues then pooled the original disfluent-font reasoning experiment with 16 replication attempts and found no effect on solution rates and no benefit under any moderator they tested, although disfluent fonts did slightly slow responses. Xie, Zhou and Liu's meta-analysis of text-based learning is titled for its own result, null effects; Weissgerber, Brunmair and Rummer found coding and sign errors in it and corrected the pooled transfer effect from +0.026 to −0.008, so the number moved but stayed at zero. Nothing here supports degrading legibility on purpose.

The best field evidence for simplicity is about cutting content, not polishing it. Bertrand, Karlan, Mullainathan, Shafir and Zinman randomised the content, price and deadline of direct-mail loan offers from a South African consumer lender, and showing fewer example loans raised loan demand by about as much as a 25% reduction in the interest rate. The authors stress that they could not predict in advance which advertising features would matter, which is the right posture: simplification moved real money, but through what the offer said rather than through how it was set.

So treat fluency as a floor rather than a lever. Legible contrast, plain words and short sentences are a baseline obligation — WCAG 2.2 sets 4.5:1 for normal text, UK GDPR Article 12(1) requires clear and plain language, and the Unfair Commercial Practices Directive Article 7(2) makes material information presented in an unclear, unintelligible or ambiguous manner a misleading omission. Expect comprehension and completion gains from clearing obstacles; do not expect polish alone to make a claim persuasive, and do not use it to smooth over a claim that is not true.