Reading with Diversity in Mind: Pupillometry and Typography Towards Inclusive Design for ADHD Readers

Llana and colleagues at the University of Rhode Island examine how typographic choices interact with ADHD when reading digital text. The motivation is practical: digital reading is now the default for most education and work, and small typographic decisions (font family, size, spacing) materially shape comprehension, speed, and cognitive effort - yet ADHD readers are an under-studied population in readability research. The authors ran a controlled in-lab study with an EyeLink Portable Duo eye tracker recording fixations, saccades, and pupil dilation while 21 participants (6 with ADHD, screened with the Adult ADHD Self-Report Scale ASRS-v1.1) read four eighth-grade-level passages, each randomly assigned one of four Roboto variants (Default sans-serif, Mono, Serif, Slab) and one of three sizes (14/16/18 px), followed by five-question multiple-choice comprehension quizzes. As a methodological contribution they also released ReadGen, an open-source tool that generates consistent text images for SR Research's Experiment Builder so pupillometry data are not contaminated by viewing-angle artifacts. Quantitative analysis used Pearson correlations linking pupil size to reading speed, mixed-effects linear models for ADHD effects on speed, and a Poisson regression for saccade counts. They followed the lab study with five semi-structured interviews of ADHD participants on what helps or hurts focus during digital reading.

Reading speed correlated negatively with mean pupil dilation (r=-0.376, p<.001) across all participants - faster reading meant smaller pupils, consistent with lower cognitive effort - and 61.9%% of participants showed an ~8.5%% gradual decrease in pupil size over the session, suggesting effort drops with sustained reading. The clearest typographic effect was a familiar speed-comprehension trade-off: Roboto Default produced the fastest reading (~10%% above mean) but the lowest accuracy (~27%%), while Roboto Serif was the slowest but most accurate (~34%%). Font size effects on comprehension were small. The ADHD-specific findings are the headline. Mixed-effects models found ADHD participants read significantly faster than non-ADHD participants (p=.029), and Poisson regression showed they made roughly 22%% fewer saccades (IRR=0.78, beta=-0.245, p=.013), suggesting broader, less detailed visual scanning. The two groups also responded to typography differently: ADHD readers were fastest with the default sans-serif Roboto at smaller sizes (14-16 px), while non-ADHD readers were fastest with Roboto Serif at 18 px. Qualitative interviews reinforced the design implications: ADHD participants reported that ads, notifications, AI summaries, visual clutter, and continuous scrolling derail reading; 80%% wanted user-controlled font, size, brightness, and background; all preferred dark or warm color palettes; 67%% favored interfaces that emulate physical books with chunked text, consistent formatting, and annotation.

For accessibility practitioners and reading-tool designers, this paper is useful on three fronts. First, it provides direct evidence that ADHD readers are not just non-ADHD readers with worse focus - their visual scanning, preferred font, and preferred size are systematically different, which means fixed defaults optimized for the general population may actively underserve them. Second, the speed-comprehension trade-off across font variants reframes 'good for ADHD' as a choice between throughput and accuracy that should be user-controlled, not pre-selected by designers. Third, the qualitative themes (distraction-free chrome, customization controls for font/size/contrast, chunked text, dark/warm color modes, no auto-injected AI summaries) form a concrete design checklist for any product targeting reading by people with ADHD. The pupillometry-as-cognitive-effort signal is also methodologically useful for evaluating future readability tools beyond self-report. Caveats are significant: only 6 ADHD participants, all university-affiliated and 90%% with a bachelor's or higher, short laboratory passages rather than long-form reading, and the study cannot disentangle ADHD-specific scanning strategies from individual variation. The authors frame the paper as a prequel - the right next steps are larger samples, longer passages, and ablations of which interface customizations actually help.