The Surface That Arrives First

Before anyone reads a word you have written, they have already decided something. This chapter is about what gets decided in that gap, how little of the popular evidence for it survives being checked, and what the surviving part actually licenses you to do.

A number sold to you by a company that sells the thing it recommends

Start where the last two dismantlings started: with the number everyone opens on.

“Colour increases brand recognition by up to eighty per cent.” It is in pitch decks, brand guidelines and design courses, and it is nearly always credited to a study at the University of Loyola, Maryland. That attribution is specific enough to feel checked. So I checked it.

The Loyola research is two papers by Dr Ellen Hoadley, from 1990 and 2000, on how colour affects the processing of information in graphs and text. Neither paper contains the figure. Neither paper is about brand recognition. They were conducted when office screens and printers were monochrome by default and colour was an expensive novelty, and what they compared was colour against no colour at all.

The eighty per cent traces instead to a marketing leaflet published around 2005, titled 20 Ways to Share the Color Knowledge. It was produced by Xerox.

I want to be careful about the lesson here, because “that statistic is fake” is the boring half of it. Nobody fabricated anything in bad faith. A printer company made a reasonable-sounding claim in a sales document, somebody quoted it and attached a university to it, and after that the attribution did all the work. It is the same failure chapter five documented in a story that travelled without its source, arriving in a different costume.

The interesting half is the direction of the interest. The number that says colour is enormously valuable came from an organisation whose revenue depends on people believing colour is enormously valuable. That is the third time this pattern has appeared in my own published work in a single month, in three unrelated fields. It is worth learning as a reflex: before asking whether a number is true, ask who benefits if it is believed.

What is genuinely measured: fifty milliseconds, and what happens after them

Now the finding that does survive, because it is a good one and it deserves to be stated precisely rather than sold.

In 2006, Gitte Lindgaard and colleagues published three studies in Behaviour & Information Technology. Participants rated the visual appeal of web homepages shown for half a second. The third study repeated the exercise at fifty milliseconds — deliberately, to test whether what they were seeing was merely an exposure effect. The appeal ratings held. They correlated strongly from one phase to the next, and they correlated between the fifty-millisecond condition and the five-hundred-millisecond one.

So a visual judgement forms in about fifty milliseconds, and giving people ten times longer does not much change it. That is a real, replicated-within-the-paper result and it is genuinely striking.

Here is the part that gets removed in transit. The study measured the speed and the stability of an aesthetic rating. It did not measure trust. It did not measure recall, or preference a week later, or willingness to buy, or whether the fast impression predicted anything at all about what someone subsequently did. Every deck that says “you have fifty milliseconds to win the customer” has attached a consequence that the research does not contain.

The replication that embarrassed the person who ran it

In 1995 Masaaki Kurosu and Kaori Kashimura showed people layouts of cash-machine interfaces and asked them to rate both how attractive each one looked and how easy it looked to use. The two ratings tracked each other closely. Nobody had used any of the machines.

Two years later Noam Tractinsky set out to test whether that was a Japanese phenomenon. His expectation, stated in the paper, was that it would not hold elsewhere — that aesthetic sensitivity of this kind was culturally specific. He ran three experiments in Israel with a translated instrument. It replicated. The correlation between how attractive an interface looked and how usable it was assumed to be was stronger among his participants than in the original.

I put weight on this study for a reason that has nothing to do with its size. It is a replication that surprised its own author in the direction of the original finding. That is a very different kind of evidence from a confirmation produced by somebody hoping for one, and there is much less of it about than the citation counts suggest.

And then a 2023 result that changes the advice

If the story stopped there, the practical instruction would be “make it beautiful”, and that is roughly what the design industry has taken from it for twenty-five years.

A short paper presented at CHI in 2023 complicates it usefully. When the researchers statistically controlled for processing fluency — how easy a thing is to perceive and make sense of — the aesthetic-usability effect got smaller. Which suggests that a good deal of what we have been reading as a response to beauty is a response to ease. Attractive things are frequently easier to parse: cleaner contrast, more predictable structure, fewer competing signals. The pleasure may be the feeling of that ease rather than a separate reaction to the beauty itself.

This is a more useful conclusion than the one it replaces, because it is actionable by people without taste and it is falsifiable. “Make it beautiful” cannot be checked. “Reduce the number of decisions a stranger has to make in the first half second” can be, and mostly by counting things.

T4 — the Flash Test

T3, in the last chapter, asked whether your positioning survives being retold by a stranger. T4 asks something earlier and more brutal: whether your surface says anything at all before it is read.

Take one of your surfaces — a homepage, a deck cover, a profile, a package. Remove your name and logo from it. Show it to someone who does not know your work, for about one second, and take it away. Then ask three questions, in this order, and write down the first answer rather than the considered one.

What kind of organisation was that? Not which one — what kind. If they cannot answer, your surface is generic, and generic is the expensive failure, because it means the fifty milliseconds went to somebody else's category.

How much does it cost? People price a surface almost instantly, and the answer tells you what tier you are competing in whether you chose that tier or not.

What did you look at first? This one checks the fluency question. If different people look at different things, you have not made a hierarchy, you have made a collection.

Two disclosures, because I would rather say them than have them found. T4 is a working instrument, not a validated one — it has no reliability data behind it and I am not going to pretend otherwise. And the research it leans on studied web pages and cash machines, not brands. Carrying it across to your surface is my argument, not the researchers'.

A position gives you something worth retrieving. A memory map tells you what it must be retrieved from. A story is how it moves without you. A surface is what decides, in fifty milliseconds, whether any of that ever gets a chance. But everything so far has assumed the surface stays still — that there is one homepage, one cover, one mark, holding its shape while people look at it. That has not been true for some time. The next chapter is about what happens when the artefact moves, adapts, and is assembled differently for every person who encounters it — and what “consistent” can possibly mean then.