Researchers ran what I'd call a design battle. Three versions of personalized drinking feedback - same data, sent by email - with only the format varying. After 90 days, the format mattered, at least for how often people drank.

What they did

222 young adults who drank regularly (mean age 21) were assigned to one of three feedback conditions, split using block allocation stratified by alcohol intake level and impulsivity:

  • Text. Researcher-written, paragraph-form feedback on alcohol intake and brain health.
  • Researcher image. The same content, designed by the research team as a visual.
  • Co-designed image. A visual format built around what users said they wanted.

The co-designed format came from a separate phase: 40 participants in focus groups described what feedback would actually reach them. Six themes came up: visual representation, clarity, accessibility, brevity, peer comparison, and harm. The researchers built the third condition around those criteria.

Feedback went out by email. The team measured total alcohol consumption, drinking frequency, and alcohol-related harm over about 90 days. A few weeks after the first feedback arrived, they also tracked capacity, opportunity, and motivation to change - three factors the study treated as predictors of behavior change.

What they found

Every group reduced their drinking. Even a plain-text email moved the needle. Total alcohol consumption dropped across all three conditions (η²=0.11), and alcohol-related harm dropped too (η²=0.05). The study had no no-treatment control group, so I can't rule out regression to the mean - but the reductions were consistent.

The format gap showed up in drinking frequency:

  • Image-based (both researcher and co-designed): effect size d=0.56 for reducing drinking days
  • Text-based: d=0.39

Effect sizes are modest by conventional standards. The difference between 0.56 and 0.39 is real but not dramatic. Still, it was consistent: both image-based conditions outperformed text on frequency.

On preference: participants liked the co-designed visual best - d=0.55 over the researcher-built image, d=0.83 over the text. But that preference did not translate into larger reductions. The co-designed and researcher-designed image formats performed identically on actual consumption.

Format also had no effect on capacity, opportunity, and motivation to change (η=0.01). Format shifted behavior without shifting stated intentions first.

What it means

The sample skews young - mean age 21 - so I wouldn't push this too far toward older drinkers. Self-reported consumption adds noise. Without a no-feedback control, it's hard to isolate the format effect from just getting any feedback at all.

But the visual-versus-text gap rings true to me regardless.

A chart of your drinking data and a paragraph about it are different kinds of information, even when the underlying facts are identical. A chart lands as a fact about you right now. A paragraph is a report. I think those two things trigger different responses - but that's speculation; the study doesn't explain the mechanism, only that the gap exists.

The co-design finding is the part I find most interesting. Participants worked to build the feedback they wanted, and they genuinely preferred it. But that preference did not produce better outcomes. A better user experience did not mean better behavior change.

On a night when you're keeping pace with the table, a text summary of your drinks doesn't land the same way a curve does. Seeing your BAC arc toward its peak and then fade puts you in a different relationship to the next drink than reading "3 standard drinks." AlcoBalance shows you that curve in real time - not a report of what happened, but where you are now and where you're heading.

Source: Journal of Medical Internet Research, 10.2196/87393