Visual Compression

It is Monday morning and a security engineer faces forty browser tabs: weekend vulnerability disclosures, three vendor advisories, a thread dissecting one of them, two papers someone said she must read. She has an hour before stand-up. She will open perhaps six tabs and skim two. The other thirty-four will be judged, kept or closed, read or abandoned, on the thinnest of evidence: a title, a favicon, a remembered recommendation, the first three lines glimpsed on hover. Every judgement is a bet. Most bets are cheap when wrong. A few are expensive: the advisory she dismisses because its title reads like routine patch noise is the one describing the library her team ships.

This chapter is about that bet, and about the first thing a flexible representation can do for it. Not help someone understand a page, which is a later question, but the narrower decision that comes first: is this worth my attention at all? Previews have always existed to answer it: titles, snippets, thumbnails, abstracts. The new possibility is to generate a preview from what the page says and what the reader is trying to do, in a form (a few small visuals, say) chosen for speed of reading.

The chapter follows one path. The page is fixed. A smaller representation is derived from it. Making it small is compression. Something becomes easier to see. Something may disappear. And the question the whole book keeps returning to follows: what had to survive?

A smaller representation, and what it is for

The purpose comes first, because everything else hangs on it:

A compressed representation succeeds when a smaller, faster representation preserves enough of what matters to the task for the reader to make an accurate selection decision.

Not fewer words. Not fewer pixels. Not prettier cards. A preview that halves reading time while hiding the decisive caveat has failed. A preview that takes the same ten seconds as a title but prevents one catastrophic skip has succeeded. The unit of analysis is the decision, and the discussion should never drift far from it.

Readers deciding what to open behave like foragers, and foraging theory gives the mechanism. Pirolli and Card’s information-foraging theory describes people who judge content they have not yet reached (the page at the end of a link) from the cues in front of them: titles, snippets, surrounding text, images, layout. That judgement is called information scent, the imperfect, subjective sense of how valuable a source will be and what it will cost to reach. Two behaviours follow. People follow the trail whose cues promise the most, and they leave an area when the scent thins. Anyone who has watched themselves skim search results has watched both happen.

A visual preview, in this vocabulary, is an engineered scent cue. It stands where the snippet stands and makes the same promise: this is what lies down this trail, and this is what it will cost you. That locates its job exactly. It does not need to convey the content. It needs to predict the value of engaging with it. And it predicts the failures, because scent can be wrong in two directions, and the two are not alike:

FALSE OPEN
the preview looks relevant
→ the reader opens it, wastes some minutes, and is unharmed

FALSE SKIP
the preview looks irrelevant
→ the reader never sees the source, and the loss has no ceiling

A false open costs time. A false skip can cost the one item that mattered: the advisory, the contraindication, the contradictory result. Any assessment that averages the two into “accuracy” buries the number that matters most. A preview that trades fewer false opens for more false skips has gone backwards, however attractive its cards.

Compression is not neutral shortening

It is easy to talk about compression as though it were shortening: the same thing, with less of it. It is not. Shortening a representation is a series of decisions about what to keep, and each decision has consequences. A compressed representation makes some things much easier to see (the topic, the stance, the headline result) and puts others out of sight (the qualifications, the denominators, the “except when”). Three separate gains are often run together, and the whole chapter depends on keeping them apart:

attention gained
   ≠ understanding gained
   ≠ correct inference
   ≠ safe action

A preview can draw the eye without helping anyone grasp the page. A reader can grasp the page and still draw the wrong conclusion from it. A correct conclusion can still be a poor basis for acting on, if the decisive caveat was left out. Evidence for one link in this chain says nothing about the next.

The nearest body of research is not about AI-generated previews at all. It concerns visual and graphical abstracts in science communication, and it shows the chain at work without needing any dramatic claims. Used as an illustration, and not as a literature review, it makes four points.

Visuals attract attention, and attention is not the same as correct routing. In a randomised trial of ways to disseminate surgical research (Chapman and colleagues, 2019), visual abstracts drew more engagement overall, but the engagement came mainly from healthcare professionals. The outcomes reported were measures of engagement, not of understanding.

Liking a form is not the same as learning more from it. In three randomised trials of Cochrane review summaries (Buljan and colleagues, 2018), people who read an infographic gained no more knowledge than those who read a plain-language summary, though the infographic was preferred.

Compression can carry a distortion with it. A 2025 study of randomised trials with non-significant primary results found “spin” (emphasising secondary findings, or presenting a non-result as equivalence) in about a third of infographics against about a quarter of abstracts and full texts. The difference was not statistically significant overall, so this is not a finding that infographics spin more. It does show that spin survives into the compressed form. Dropping a qualifier while illustrating the encouraging subgroup is not a harmless side effect of shortening. It changes what the page appears to say.

And readers do not always use a representation the way its designer meant. Reporting-guideline work on infographics (the RIVA-C project, 2024) raises the concern that readers may treat an infographic as a substitute for the full text, while many omit elements needed to interpret a result, such as the comparator, harms and uncertainty. A preview built for selection will be used for understanding whatever the designer intended.

None of this shows that visual summaries cause wrong decisions, and the chapter does not claim it. What it shows is how a compressed representation can win on attention and preference while the things that would make it safe to rely on have quietly dropped out. That is exactly the pattern a preview for selection has to be tested against.

Three ways a preview could win

If a visual preview does beat a title or an excerpt, it matters why, because three different victories hide inside one number.

1. It carries more useful information. The cards surface the page’s real structure, its claims, comparisons and exceptions, in less space than the excerpt. This is the victory worth wanting, and it should survive even ugly, unstyled cards.
2. It conveys structure faster. The information is the same as in a text summary, but decoded sooner, because spatial layout does work that sentences do slowly. This shows up as faster decisions at equal accuracy, which is valuable when attention is scarce.
3. It is merely more attractive. Colour, polish and fluency pull the eye and raise confidence without improving the judgement underneath. This shows up as preference, with accuracy flat or worse, and it is the most dangerous outcome because it feels like success while producing false skips and false opens with greater confidence.

A fair test therefore measures decision quality, decision speed and preference separately. Its most useful finding might be a preview that wins on preference while losing on selection accuracy. That would show that attractiveness is something to control, not something to maximise.

When the right preview is no preview

Compression is an intervention, and interventions have to earn their place. In a chapter about visual previews, a large class of pages should get none: just the title and the source, left as they are. Four cases show why.

An argument that is its hedges. A methods critique, a legal analysis or a careful negative result compresses into apparent certainty. Any card set either reproduces the full structure of qualification, which means nothing has been compressed, or drops it, which destroys the point. The honest preview is the title and the abstract.

A distinction that is mostly numbers. Two treatments differing by a hazard ratio of 0.87 against 0.91, or two plans differing by 3% of the budget: visual encoding invites magnitude illusions in bar lengths and areas that the numbers stated plainly would not cause. Text carries this kind of content. Pictures distort it.

A short piece that its title already summarises. “Unauthenticated remote code execution in Acme Router firmware before 4.2” needs no cards. Generating a preview spends compute, attention and distortion risk to restate what a dozen words already decided.

A concept that invites a false picture. Some ideas come with stock imagery that smuggles in the wrong structure: the brain as a circuit board, evolution as a ladder, a network protocol as plumbing, risk as a traffic light. A generated image that reaches for the obvious metaphor can teach the wrong model faster than prose would.

These are not edge cases to patch later. They are the chapter’s lasting contribution: a transformation has to earn its use, and a system that cannot leave a page alone cannot be trusted with anything stronger.

A preview is not a summary

One last distinction keeps this chapter from spilling into the next. A preview exists for one purpose: to help the reader decide whether to spend attention on the source. A summary implies something broader: to let the reader understand the source without reading it. These carry different obligations. A preview may legitimately leave out everything except what distinguishes this source from its neighbours (topic, stance, key result, decisive caveat), because its job ends when the reader decides. A summary that leaves out the same material has failed. And a preview that tries to teach the content has overreached: slower to scan, more to get wrong, and doing a different job.

Likewise the claim here stops at selection. Nothing in this chapter says a preview helps anyone understand, remember or apply what the page says.

Where this leaves us

A compressed representation is legitimate as decision support: it earns its place by improving the accuracy of the reader’s selection, with false skips counted separately and weighted heavier, and without hiding what the decision depends on. Compression is not neutral shortening, and attention gained is not understanding gained, correct inference or safe action. Information scent explains how a preview works, attractiveness is the confound to guard against, and leaving the page alone is always a candidate.

The preview keeps the reader outside the page. The next question is what can be done once they are inside it: when does adding a diagram, a table or a timeline to the page itself help someone understand it?

References

• Pirolli, P. & Card, S. (1999). Information foraging. Psychological Review, 106(4), 643–675.
• Pirolli, P. (2005). Rational analyses of information foraging on the Web. Cognitive Science, 29(3), 343–373.
• Chi, E.H., Pirolli, P., Chen, K. & Pitkow, J. (2001). Using information scent to model user information needs and actions on the Web. Proceedings of CHI 2001.
• Chapman, S.J. et al. (2019). Randomized controlled trial of plain English and visual abstracts for disseminating surgical research via social media. British Journal of Surgery, 106(12), 1611–1616.
• Buljan, I. et al. (2018). No difference in knowledge obtained from infographic or plain language summary of a Cochrane systematic review: three randomized controlled trials. Journal of Clinical Epidemiology, 97, 86–94.
• Do infographics “spin” the findings of health and medical research? (2025). BMJ Evidence-Based Medicine, 30(2), 84 onwards. (Authors include J.R. Zadro; full author list and end page to be confirmed.)
• Zadro, J.R. et al. (2024). Development of the Reporting Infographics and Visual Abstracts of Comparative studies (RIVA-C) checklist and guide. BMJ Evidence-Based Medicine, 29(5).