Users spend most of their time on other products. They expect yours to work the same way as everything they already know — and when it doesn't, even small deviations feel like the product is broken.
Before a user ever opens your product, they have already spent thousands of hours inside other products. They know where the shopping cart goes. They know what a hamburger icon does. They know the back button lives in the top-left corner and the search bar lives at the top of the screen. This accumulated knowledge is not a blank slate you can override — it is a set of deeply reinforced expectations that your design either aligns with or fights against.
Jakob Nielsen, co-founder of the Nielsen Norman Group, formalised this observation in what became known as Jakob's Law: users spend most of their time on other sites, so they come to your site with existing mental models built elsewhere. When your interface matches those models, users feel competent immediately. When it doesn't, even a small deviation — a cart icon in the wrong corner, a nav bar in an unexpected position — generates friction that users experience as the product feeling “off” without being able to explain why.
Jakob Nielsen published this law in 1998, at a time when the web was still young enough that many designers treated every site as a blank canvas. His argument was deceptively simple: the majority of a user's experience happens on other people's products, not yours. Every hour a user spends on Amazon, Google, or Instagram trains a set of expectations about where elements should be and how interactions should work. Your product inherits those expectations whether you designed for them or not.
The mechanism is transfer learning — the same process that lets you drive a rental car without reading the manual. Steering wheel, pedals, mirrors, turn signals: they're all in the same place because every car manufacturer follows the same conventions. Imagine if one manufacturer moved the brake pedal to the right side. It might be a “better” position biomechanically, but the cost of retraining millions of drivers would be catastrophic. Interface conventions work the same way.
“Users spend most of their time on other sites. This means that users prefer your site to work the same way as all the other sites they already know.”
— Jakob Nielsen
The shopping cart in the top-right corner of an e-commerce site is one of the most consistent interface conventions on the web. Amazon, ASOS, Nike, Zara, Apple — every major e-commerce site on the planet puts the cart in the top-right. Users know this. When they want to check their cart, they move their eyes or cursor to the top-right without thinking.
Moving the cart — hiding it inside a hamburger menu, putting it on the left, or removing the icon entirely in favour of a text link — forces users to search for something they expect to find instantly. The cart doesn't become harder to use. But the moment of confusion, the need to visually scan the header, is enough to introduce doubt. And in a checkout flow, doubt is expensive.
Cart is inside the hamburger menu. Users scan the top-right first and find nothing.
Cart is top-right with a badge showing item count. Zero searching required.
The cart badge is also part of the convention: a red dot or number indicating items in the bag is a pattern users recognise instantly from every other shopping site they use. Even if your visual design is completely unique, the structural position of the cart and the badge convention should remain standard. The creativity goes into everything around it — not into moving the cart.
In the time since ChatGPT launched in late 2022, an entire generation of AI chat interfaces has converged on a single layout: model selector at the top, conversation in the middle scrolling upward, input field pinned at the bottom with suggestions above it. This pattern has been repeated across ChatGPT, Claude, Gemini, Copilot, Perplexity, and dozens of smaller AI products.
It is, in Jakob's Law terms, already a convention. Users who use AI products regularly have built a mental model of where the input is. Placing the input at the top — or removing the bottom input in favour of a different layout — breaks that expectation immediately. The interface doesn't feel innovative. It feels wrong.
Users look at the bottom for the input — it's not there. New messages push the conversation further down.
Input is exactly where muscle memory from ChatGPT, Claude, and Gemini has trained users to look.
The AI input-at-bottom pattern is only three years old — far younger than most interface conventions. But it has been reinforced by an unusual density of usage: AI chat is now a daily tool for hundreds of millions of people. Three years of daily use creates stronger mental models than ten years of occasional use. If you're building an AI product today, the bottom input is not a stylistic choice — it's an expectation.
Jakob's Law doesn't mean every product should look identical. Visual design, tone, branding, and personality can all be radically different. What should remain consistent are the structural positions and interaction patterns that users have internalised from thousands of hours of use elsewhere. The logo links home. Search is at the top. The cart is top-right. The back button is top-left. Primary mobile navigation is at the bottom. These aren't limitations on creativity — they're the invisible scaffolding that lets users get comfortable quickly enough to appreciate everything you've done differently.