Week 29 · Human Interaction Psychology

You Have Never
Used This Site

And yet you already know where the cart icon lives. In July 2000, Jakob Nielsen wrote four words that explain why: users spend most of their time on other sites. Your interface is never judged alone — it is judged against everything else they have already used.

Part I

Two Stores, One Task

Below are two mock stores. Both sell the same $18 mug. Both ask you to do the same three things: add it, open the cart, reach checkout. One store places everything exactly where a hundred other checkouts have already trained you to look. The other rearranges the same task — on paper, more thoughtfully.

Run the attempt. Each store is timed from its own first tap to its own last, with a live step and miss count. Run it again if you like — every attempt adds a point to the plot below.

Predict

Before you run it: will the reimagined layout get you to checkout faster, slower, or about the same?

The checkout race

Same task, both times: add one item, reach checkout

Two mock stores. Same task: add the item, reach checkout.

One follows the checkout shape you have used a hundred times. The other rearranges it — on paper, more thoughtfully.

Run the attempt above and this paragraph will fill in with your own two numbers, side by side.

Part II

Four Words, One Mechanism

Nielsen’s original sentence, from the 22 July 2000 Alertbox column “End of Web Design,” carries no citation and no percentage: “Users spend most of their time on other sites.” The consequence he draws is the whole law: users prefer your site to work the same way as all the other sites they already know.

Jon Yablonski’s later reformulation is the sentence worth carrying into a review, because it names the mechanism the original leaves implicit: “Users will transfer expectations they have built around one familiar product to another that appears similar.” The qualifier — that appears similar — is doing all the work. Resemblance is the trigger, and the trigger fires before comprehension.

If the organism carries a “small-scale model” of external reality inside its head, it can react to future situations before they arise.

Kenneth Craik, The Nature of Explanation, 1943

That sentence, from 1943, is the mechanism underneath Nielsen’s 2000 observation. A mental model is a simulator built from frequency-weighted experience. Donald Norman’s contribution names the gap it leaves you with: your design model never reaches the user directly — only the system image, everything the product shows and does, does. The user’s own model is built from that image plus everything they already knew, and the second term is the one you do not control.

The two gulfs are Norman’s: the Gulf of Execution (what you want to do vs. what the system lets you do) and the Gulf of Evaluation (what the system did vs. what you can tell it did). A familiar affordance narrows the first; a familiar feedback pattern narrows the second.

Transfer has a sign. Positive transfer makes a new task easier because prior learning carries over. Negative transfer makes it harder, and it needs two conditions at once: high surface similarity and different behaviour. That is a design-created failure mode — it cannot happen to something that looks as different as it behaves. Style your own custom dropdown to look like a native select, then give it half the keyboard behaviour, and you have manufactured negative transfer on purpose.

Which is also where convention and the surface-level sameness people call a cliché come apart. The dossier’s own split is exact: conform in structure, behaviour, terminology and placement — that is convention, and it is what Jakob’s Law actually asks for. Diverge freely in typography, colour, imagery, motion character and voice — that is the expressive layer the law never touches. Copying the surface anyway, out of habit rather than need, is what produces what this dossier calls the beige-SaaS aesthetic: uniform not because it had to be, but because nobody separated the two layers.

Part III

The Anatomy of a Convention

A cart control is not one decision — it is at least four, stacked. Position, icon, label, order. Scrub through them one at a time and watch the modelled cost of breaking each.

Anatomy of a convention

scrub through position → icon → label → order

"Cart"

product → add → cart → checkout

Every dimension standard — cart top right, cart icon, “Cart”, product → add → cart → checkout. Scrub forward to break one at a time.

Novel dimensions, added one at a time

0 — all standard4 — Order

0 — all standard

Novel so far: none
Modelled cost: 0; no relearning — the model still matches

Modelled, not measured — there is no dossier figure for “cost in units.” The weights are this page’s own construction, built from four separate dossier claims: position carries the heaviest weight because Rule 3 (recognition) and the reported six-fold centred-logo navigation-failure finding both hang on it; icon carries the lightest because a mislabelled glyph survives if the text label stays put (Rule 1’s checklist).

Part IV

Four Cases, Graded

Jakob’s Law is the shortest, least-measured principle in this curriculum — the 2000 column cites no study at all. What it does have is two decades of case evidence, of uneven quality. Step through four, each graded for how far you should trust it.

Four cases, graded

1999

1-Click

verified

Amazon patented one-tap purchase (US 5,960,411) and sued Barnes & Noble for its "Express Lane" feature in 1999; the case settled in 2002, and Apple licensed the patent in 2000. Until the patent expired in 2017, the fastest checkout on the web was legally unavailable to almost everyone — the multi-step flow that formed instead became the model users learned, and the one Baymard has measured ever since.

Case

1-ClickThe homogenisation study

1-Click

Graded per this dossier’s own verification pass: the Nielsen 2000 quote, the Baymard 70.22% figure and its 50-study methodology, and the Amazon patent facts (US 5,960,411, the Barnes & Noble injunction, Apple’s licence) were independently confirmed. Windows 8’s narrative and the CHI 2021 homogenisation numbers are secondary — the dossier itself flags a discrepancy between two secondary citations of the same CHI study, not a confirmed error in either.

Part V

Innovate in One Place

The common mistake this dossier names is spreading novelty evenly — a little bit unusual everywhere, rather than concentrated in one place. Pick which of the four elements to make novel yourself and watch the modelled cost move.

One element stays bounded. Three or more stop behaving like three separate small costs.

Spend your novelty budget

pick which elements to make unfamiliar

Elements made novel: 0
Modelled total cost: 0 units; no relearning — the model still matches

Modelled cost = (sum of the selected elements’ weights) × (1 + 0.18 × (k − 1)), for k selected elements. The multiplier is this page’s own construction, built to dramatize the dossier’s own claim (2.5, common mistake): “twelve slightly unusual interactions feel harder than one genuinely new one, because the user never establishes a stable model at all.”

Challenge · Five deviations

Open

A checkout wireframe has five things that no longer match convention. Three carry real transfer cost. Two are free — the dossier’s own expressive layer. Flag exactly the three that matter.

Flagged: 0 of 5
Target: exactly 3, the real ones; keep testing

Structure, behaviour, terminology and placement cost something. Typography, colour and motion character do not.

Insight

Users do not learn your interface. They recognise it.

Every one of this dossier’s rules — consistency, real-world match, recognition over recall, user control — is a version of the same fact: a mental model is not built from your onboarding screen, it is built from the thousand other interfaces the user met first. You do not own that training data, and you cannot see it. The only lever you actually hold is which reference set your design resembles — free, at design time, and the one thing every case in Part IV has in common, win or lose.

Mental model

Whenever a review says ‘users expect it this way’ or ‘let’s do something different’,

look for which reference set you are actually being judged against, and whether the deviation touches structure, behaviour, terminology or placement — not typography, colour or motion,

because transfer is triggered by resemblance, not by category, and the cost of breaking it is paid immediately, in full, by every first-time user.

Tomorrowname the three to six products your users have spent the most hours in for this exact task. That list, not your competitors, is the model your interface is actually being tested against.

End of experience

← Back to the constellation

You Have NeverUsed This Site

Two Stores, One Task

Four Words, One Mechanism

The Anatomy of a Convention

Four Cases, Graded

Innovate in One Place

Where this idea goes next

You Have NeverUsed This Site

Two Stores, One Task

Four Words, One Mechanism

The Anatomy of a Convention

Four Cases, Graded

Innovate in One Place

You Have Never
Used This Site

You Have Never
Used This Site