- Dark mode needs two designed palettes; an inverted single palette breaks the contrast relationships.
- Larger body text and line height raised time on page every time we have shipped it.
- Container queries, subgrid and :has() remove most reasons to compute layout in JavaScript, which is where interaction latency comes from.
- Scroll-jacking, auto-playing hero video and custom cursors each fail at least two of our three tests.
- Redesigns break on content, not visual direction: design against the longest real page, not the shortest.
Design trends articles are usually written by people who do not have to maintain the result. We build what we design, so every trend below is judged on three things: what it costs to ship, what it costs to maintain, and what it does to the page on a mid-range Android phone.
Keep: dark mode as a designed theme
Dark mode has stopped being a feature and become an expectation. What changed is the standard. An inverted colour scheme with the same tokens flipped looks broken, because contrast relationships do not survive inversion.
Doing it properly means designing two palettes, checking contrast in both, and testing every image and illustration in each. Illustrations built from theme tokens rather than fixed colours adapt on their own, which is why we draw ours that way.
The cost is real and one-off. The benefit is a site that respects a setting a large share of visitors have already made.
Keep: bigger type, more space
The drift towards larger body text and more generous spacing is the most reliably good trend of the last five years. Sixteen pixels is a floor, not a target. Eighteen reads better on most screens.
Every time we have increased body size and line height on a client site, time on page has gone up. That is not fashion, it is people being able to read.
Related and equally welcome: fluid type scales that interpolate between breakpoints rather than jumping, which modern CSS makes trivial.
Keep: layout done in CSS, not JavaScript
Container queries, subgrid and :has() have removed most of the remaining reasons to compute layout in JavaScript. A card that rearranges itself based on the space it occupies rather than the viewport width is a genuinely better component, and it ships nothing to the browser beyond CSS.
This is the trend with the largest performance dividend, because JavaScript running on resize is a common cause of poor interaction latency. We put numbers on that in what actually moves Core Web Vitals.
Keep: designed illustration over stock photography
Stock photography of people pointing at laptops has become a signal that a business has nothing specific to show. Illustration drawn in the brand's own palette, or genuine photographs of the actual team and the actual work, both beat it.
The practical advantage of illustration is that it can be built from theme tokens, adapts to dark mode by itself, and weighs a fraction of a photograph.
Keep: motion that respects the setting
Subtle entrance animation is fine. Two rules make it acceptable: honour prefers-reduced-motion, and never let animation delay content becoming readable. An element that fades in after 400 ms is an element that was not there when the reader looked.
Skip: scroll-jacking
Hijacking the scroll to drive a narrative breaks the one interaction every visitor already understands. It fails badly with keyboards, screen readers and trackpads, and on the sites where we have measured it, it increases exits rather than engagement.
If the story needs several steps, use several sections and let people move at their own speed.
Skip: auto-playing hero video
A video background is usually the largest asset on the page, and it competes with the LCP element for bandwidth at exactly the wrong moment. On mobile it either does not autoplay at all or drains battery to no purpose.
If the video is genuinely the point, use a poster image and let the visitor press play. The people who want it get it, and everyone else gets a fast page.
Skip: custom cursors and hover-heavy interactions
They do not exist on touch devices, which is where most traffic is, and they interfere with assistive technology. The effort is better spent on the states that matter, focus, active, disabled, error, which are usually missing on the same sites that have a custom cursor.
The thing that is not a trend
Accessibility settles more design arguments than taste does, and it settles them faster.
Body text at 4.5:1 contrast. A visible focus ring on every interactive element, including the ones inside custom components. Tap targets of at least 44 pixels. Real text rather than text inside images. Form labels that stay visible instead of vanishing into placeholders. Headings in order, one h1 per page.
None of that is decoration and none of it is negotiable in the work we ship. It also has the useful property of ruling out several trends before anyone has to have an opinion about them.
Redesigns that keep the rankings you already have: content audit, design, staged migration and redirects checked before anything goes live.
What actually breaks a redesign
Not the visual direction. The content.
The pattern we see: a design approved on three pages with beautifully short copy, then a real site with a 90-word product paragraph, a nine-item navigation and a legal footer nobody mentioned. Cards break, headings wrap to four lines, and someone starts adding exceptions to the CSS.
The cure costs nothing. Design against the longest real content you have, not the shortest. Ask for the actual page titles before the layout is signed off. Build one page with everything the client might ever put on it, then judge the design there. Every redesign we have run late was late for a content reason, never because a gradient was hard.
How we decide
When a client asks for something from a trends article, we ask three questions. What does it cost to build and maintain when the content changes? What does it do to the page on a mid-range phone? Does it work with a keyboard?
Most trends survive one of those. The ones worth keeping survive all three. If you are weighing up a rebuild against a tidy-up, the ecommerce architecture piece is the better read for a store, and our web development page explains how we scope both.
Questions
How often should a website be redesigned?
Every four to six years for most businesses, or sooner if the content structure no longer matches what you sell. Redesigning because the site looks dated, when it converts well, is usually an expensive way to feel better.
Does dark mode matter for a business website?
It matters to the substantial share of visitors who have set it at the system level. Doing it badly is worse than not doing it. An inverted palette with broken contrast reads as carelessness.
Are animations bad for SEO?
Not directly, but they often hurt interaction latency and delay content becoming readable, which are measured. Animation that runs after content is visible and honours reduced-motion settings is fine.
What is the most common design mistake you see?
Low-contrast grey text on white, closely followed by removing focus outlines because they looked untidy. Both make the site harder to use for a lot of people and neither buys anything.
Should I follow design trends at all?
Follow the ones that improve readability, performance or accessibility. Those tend to be permanent improvements. Treat the visual fashions as optional, because the redesign that follows them will arrive in two years.