The trend lists published last January have now had eight months to be tested. Most of what they promised didn't happen, several things they praised turned out to be expensive mistakes, and the two developments that actually defined the year weren't on anybody's list.
The scorecard
Graded against what teams actually shipped, rather than against what looked good on a mood board.
| Trend | What happened | Verdict |
|---|---|---|
| Bento grids | Adopted so widely they stopped meaning anything | Held — but now a default, not a statement |
| Dark mode | Continued to win; users expect the option | Held — table stakes |
| Heavy 3D and WebGL | Drained performance budgets far more than teams expected | Out — cost exceeded return on most sites |
| Kinetic typography | Looks impressive in a showreel; thin on substance in production | Overrated — fine as accent, poor as strategy |
| Deep AI personalisation | Ran into consent and privacy constraints quickly | Out — the legal reality arrived first |
| Motion everywhere | Replaced by small functional feedback interactions | Evolved — micro over macro |
| Machine readability | Nobody predicted it. Became the year's real story | In — and undersupplied |
| Raw, imperfect aesthetics | Emerged as a reaction to AI-generated polish | In — and accelerating |
Look at the last two rows. Neither appeared in the January forecasts, both became significant, and — this is the interesting part — they have the same cause.
One force, two opposite directions
Generative AI did something strange to web design this year. It pushed the visible surface and the underlying structure in opposite directions at the same time.
Downward, toward the machine. A growing share of people now research through AI systems that read your page before any human does. That makes the page's parseability a design concern — clean semantic markup, content present in the HTML rather than assembled by script, accurate structured data, clear hierarchy. Design that reads well to a machine.
Upward, toward the hand. Simultaneously, AI made competent visual output nearly free. Templates, imagery and layouts that would have looked professional in 2023 are now generatable in seconds, which collapsed the value of looking professional. The response from brands that want to seem real has been to visibly demonstrate human authorship — rough edges, hand-drawn marks, unretouched photography, deliberately broken grids.
The year in one line The surface of the web got more human because AI made polish worthless. The substructure got more machine-readable because AI became the reader. Same cause, opposite directions.
Once you see that, the rest of the year's trends sort themselves. Almost everything that gained ground serves one of those two movements. Almost everything that lost ground served neither.
What's in
Designing for machines as well as people
The most consequential and least visually interesting development of the year. If AI systems can't parse your page, they can't quote you — and being quoted is increasingly how people find you at all.
What this means in practice is unglamorous: semantic HTML rather than a stack of generic containers, content rendered in the markup rather than assembled entirely client-side, accurate structured data and schema, and copy that makes specific, quotable claims instead of atmospheric brand language. Much of it overlaps with accessibility work, which is a rare case of two obligations pointing the same way.
The reason to treat it as a design trend rather than a development task is that it's determined at design time. A layout specified as a series of decorative panels with text baked into images is unparseable no matter how good the engineering is afterwards. Our fuller treatment is in designing websites for AI search.
Deliberate imperfection
Hand-drawn arrows and annotations. Photography that hasn't been retouched into oblivion. Slightly-off alignment used on purpose. Visible seams. Type that looks set by a person rather than a system.
This reads as an aesthetic choice and functions as a credibility signal. When flawless is free, flawed becomes evidence — which is the same dynamic making brand trust harder to earn in an AI-generated world, arriving in visual form.
The caution: this is easy to do badly. Imperfection applied as a style layer on top of a generic template reads as affectation, and users are getting quick at spotting it. It works when the roughness is genuinely yours — your team's actual photos, your actual handwriting, your actual product mid-use.
Chemicloud
Affordable, reliable web hosting with free migration and 24/7 real human support — no AI chatbots
Best for: Web Hosting for Growing Websites
Anti-grid and neubrutalism, as a counter-move
Bento's ubiquity created its own opposition. Deliberately broken layouts, raw structural aesthetics, monospace type, harsh contrast, visible borders and no rounded corners anywhere — a rejection of the soft modular look that had become the default.
It's genuinely distinctive and it's genuinely inappropriate for most businesses. It suits studios, indie software, publications and portfolios — contexts where attitude is part of the offer. On a site selling professional services to cautious buyers it reads as amateurish rather than confident.
Micro-interactions over motion
Motion for its own sake is out; small functional responses are in. A button that acknowledges a press, a field that reacts to input, a toggle with weight to it. These give feedback, which is a usability function, and they cost almost nothing in performance.
The distinction that matters: does the animation tell the user something, or is it decoration? Decoration is what got cut this year, largely for the reasons in the next section.
Typography carrying the page
Large, confident type replacing the hero image entirely is one of the clearer aesthetic shifts of the year, and it has a practical logic behind it that pure fashion doesn't.
Type is nearly free to load, scales cleanly to any screen, is fully readable by the systems now parsing your page, and — unlike a photograph — cannot be generated indistinguishably by a model. A hero built from a strong sentence in a distinctive typeface does the same job as a stock image at a fraction of the weight, and says something specific rather than atmospheric.
The failure mode is treating scale as a substitute for having something to say. Enormous type wrapped around a generic value proposition just makes the emptiness bigger. This trend rewards teams with sharp copy and punishes teams without it, which is why it looks so good on some sites and so hollow on others.
Warmer, higher-contrast palettes
The cool grey neutrals that dominated the early 2020s have been giving way to warmer tones — browns, terracottas, greens, softer yellows. Partly fashion, partly a reaction to the same sterility problem driving the imperfection trend.
Worth noting the practical constraint underneath: colour contrast is measurable and increasingly audited, so warm palettes still have to clear accessibility thresholds for body text. Run the palette through a contrast checker before it gets approved, not after it ships.
What's out
Heavy 3D and WebGL
The clearest failure of the year's predictions. These consistently cost more in load time and interactivity than they returned in engagement, and the cost lands hardest on the mid-range mobile devices that determine your Core Web Vitals grade. A hero animation that delights on a designer's laptop can push a page out of the passing band for everyone else.
Not dead — appropriate for product visualisation where seeing the object in three dimensions is the point. Dead as decoration.
Deep AI personalisation
Widely predicted, widely underdelivered. Reshaping page content per visitor requires knowing enough about them to do it, and the consent and privacy constraints on that knowledge arrived faster than the design capability. Most implementations quietly narrowed to something closer to segment-based content swapping.
Generic stock photography
Already ignored by users, and now carrying a second problem: it reads as AI-generated whether or not it is. The visual signature of generic professional imagery has become the signature of synthesised imagery, so it now actively signals the opposite of what it's meant to.
Soft corporate UI
Heavy drop shadows, very large corner radii, floating card stacks, gentle gradients. The default look of the early 2020s SaaS site is aging visibly — not because anything is wrong with it, but because it's now what a template produces when nobody makes a decision.
Autoplaying video heroes
Expensive, frequently muted, often skipped, and a reliable way to fail a performance budget on the exact page where first impressions form.
Scroll-driven storytelling, at scale
Long narrative pages where content reveals and transforms as you move down them. Impressive when done properly and expensive in a way that's usually underestimated — they need real editorial content to justify the format, they're difficult to make work on small screens, and they're often the first thing to break on a slower device.
The version that survived is lighter: sensible reveal-on-scroll for sections, not a full cinematic sequence. Reserve the ambitious form for pages where the story genuinely is the product — an annual report, a launch, a case study with a real arc.
What isn't a trend and never was
Worth separating, because trend articles routinely list these alongside aesthetics and they belong in a different category entirely.
Accessibility is a requirement, increasingly a legal one in several jurisdictions, and it doesn't cycle. Contrast ratios, keyboard navigation, focus states, alt text and sensible heading structure are baseline competence.
Performance is a constraint that every aesthetic decision spends against. It's the budget, not an item on the list.
Mobile-first stopped being a trend around a decade ago and is now simply where most of your visitors are.
Clarity about what you do and what happens next is the thing that actually determines whether the site works, and it's unaffected by fashion — the substance of conversion-focused design and, at page level, landing pages that convert.
A site can be entirely on-trend and fail all four. Plenty are.
The sameness problem underneath all of this
One more force worth naming, because it explains why trend-following has become riskier than it used to be.
AI-assisted design tools produce good work fast, and they produce it from the same underlying training distribution. When a whole industry uses them, the visual baseline flattens — the same layouts, the same spacing rhythms, the same imagery treatments, arrived at independently. It's the content sameness problem in visual form, and it's the reason bento grids stopped signalling anything.
The practical consequence: adopting a trend now confers less distinctiveness than it did, because everyone can adopt it in an afternoon. What still differentiates is the material a tool can't generate — your own photography, your own voice, your own product shown honestly. Which is the same conclusion arriving from AI-generated video, and from most other places this year.
What to actually do with a redesign this year
- Fix the machine layer first. Semantic markup, structured data, content in the HTML. Invisible, unfashionable, and the highest-return work available right now.
- Pick two aesthetic directions, not eight. Ones that say something true about the business. Execute them properly rather than sampling widely.
- Budget performance before you design. Decide the weight the page can carry, then design inside it. The reverse order is how 3D heroes get approved.
- Replace stock imagery with anything real. Your team, your product, your work. Even imperfect real photography now outperforms polished generic imagery.
- Check contrast and keyboard navigation before sign-off. Cheaper now than after launch, and increasingly not optional.
If the aesthetic direction is what you're missing rather than the execution, that's a positioning question before it's a design one — the sort of work a branding and design partner handles better than a template library.
The short version
Bento and dark mode held up and became defaults. 3D, kinetic type and deep personalisation cost more than they returned. The two things that mattered were on nobody's list: designing for machines to read, and designing to prove a human made it. Both came from the same source. Pick two directions, budget your performance first, and remember that most trends are wrong for most businesses.
Site looking like everyone else's template?
We build distinctive sites that stay fast, accessible and readable to the systems your buyers now use.
Explore Web Design →Frequently asked questions
What are the biggest web design trends in 2026?
Two that were widely predicted and two that were not. Bento grid layouts and dark mode held up and are now closer to standard practice than to trends. The two nobody forecast are the ones that actually matter: designing for machine readability, so AI systems can parse and quote your pages, and a raw, deliberately imperfect aesthetic that emerged as a reaction to AI-generated polish. Both trace to the same cause, which is generative AI changing what a website is competing against.
What web design trends are going out in 2026?
Heavy 3D and WebGL, which consistently cost more in performance budget than they returned in engagement. Motion for its own sake, replaced by small functional micro-interactions that give feedback rather than decorate. Deep AI-driven personalisation, which ran into privacy and consent constraints faster than most teams expected. Generic stock photography, which was already ignored and now reads as AI-generated by default. And the soft corporate look of heavy shadows, extreme rounded corners and floating cards is aging noticeably.
Are bento grids still a trend in 2026?
They held up, which is the problem. Bento layouts became so widely adopted that they stopped signalling anything about the brand using them — a modular grid of rounded cards now reads as a default rather than a decision. They remain genuinely good at organising dense information across screen sizes, so the sensible position is to use them where the content is actually modular and to stop expecting them to carry any distinctiveness on their own.
What is designing for machine readability?
It means building pages so that automated systems can parse, understand and quote them, not just render them for humans. In practice that means clean semantic HTML, content present in the markup rather than assembled entirely by client-side JavaScript, accurate structured data, clear heading hierarchy, and specific answerable statements rather than vague marketing copy. It matters because a growing share of discovery now happens through AI systems that read your page before any person does, and a site those systems cannot parse is effectively invisible to them.
How many design trends should a website actually use?
Two or three, executed properly, rather than a dozen applied thinly. Trends are a vocabulary for saying something specific about a brand, and using all of them at once says nothing. The practical filter is to ask what each trend costs — in performance, in accessibility, in build time — and whether it communicates something true about the business. Most trends are wrong for most sites, and choosing not to follow one is a legitimate design decision rather than a failure to keep up.
