Why all websites look the same in 2026

Between 2010 and 2019, the average distance between website layouts fell by 44%. The figure comes from a study presented at the CHI 2021 conference (opens in a new tab) by researchers at Indiana University, who compared more than 227,000 screenshots from more than 10,000 websites over seventeen years. That all websites look the same is not the impression of a tired designer. It can be measured, it keeps growing and it started long before any AI generated a single home page.

Generative AI did not create the convergence, it accelerates it. A model trained on the web produces the most probable layout in its corpus, the one millions of sites already share. The result is a site that is correct, readable, fast to ship and impossible to tell apart from its neighbours.

This article first establishes what the measurement says and what it does not say. It then details the layers of tooling that produce the same site and what a visitor decides in front of a page they feel they have already seen. It reports what brand data has measured about indifference, before showing where singularity lives when everyone uses the same tools.

The study by Sam Goree and his colleagues combines two methods. The first is computational. Computer vision algorithms compare archived screenshots from 2003 to 2019 along two dimensions, colour and layout, across more than two million pairs of sites. The second is qualitative. Eleven web designers were interviewed about their practice and the sites they designed.

The curve is not a steady slope. From 2003 to 2007, websites diversify. The web leaves HTML tables behind and everyone experiments. Then the trend reverses and never turns back. The authors identify two technological breaks. The first follows the iPhone in 2007, when mobile browsing imposes responsive design and reduces the number of viable layouts on a narrow screen. The second arrives around 2013 with the spread of libraries and frameworks, which hand everyone the same ready-made components. A further acceleration appears around 2017.

The designers interviewed do not describe laziness. They describe constraints. They cite mobile, budget, the expectations of clients trained by the sites they visit and optimisation for search engines. Convergence is an industrial effect, not a lapse of taste. That is precisely what makes it hard to fight, because every decision that feeds it is rational when taken on its own.

A site built in 2026 passes through several layers of tooling before it reaches the screen. Each one narrows the space of possible forms.

Platforms and their themes

WordPress powers 40.3% of all websites (opens in a new tab) according to W3Techs in September 2026, which is 58.8% of the sites running an identified content management system. Shopify follows at 5.4%, Wix at 4.2%, Squarespace at 2.4%. More than one site in two therefore starts from a theme, which is a layout drawn to suit thousands of different buyers.

Share of websites by content management system according to W3Techs, September 2026

CSS frameworks and component libraries

Of the 3,977 developers who answered the question in the State of CSS 2025 survey (opens in a new tab), 2,041 use Tailwind CSS, 1,194 Bootstrap and 766 shadcn/ui. These are good tools. They also ship, by default, the same spacing scale, the same corner radii, the same shadows, the same hierarchy of greys and the same family of components. Anything not explicitly decided takes the default value. That value is the same for everyone.

Typography

The Fonts chapter of the 2025 Web Almanac (opens in a new tab) by HTTP Archive lists the fonts most often served through Google Fonts. On desktop, Poppins leads with roughly 3.8% of requests, followed by Open Sans, Lato and Montserrat. On mobile, Roboto joins the group. The same families show up among sites that self-host their fonts. A free, readable, neutral typeface is a sound choice. It is also everyone's choice, which strips it of any ability to sign.

Generative AI adds a layer on top of the others

Site generators sit on top of the other three layers. Lovable states in its documentation (opens in a new tab) that it uses Tailwind for styling. Since 13 May 2026, its new applications run on TanStack Start. Vercel presents v0 (opens in a new tab) as working with Next.js, Tailwind and shadcn/ui. A language model produces what is statistically most probable given its corpus. Asked without a precise art direction, it returns the median of the web. The technical cost of these sites is covered in the article on the debt of generated sites (opens in a new tab). The aesthetic cost does not show in the code.

Resemblance is not a flaw in itself. The Nielsen Norman Group has stated Jakob's Law (opens in a new tab) since 2000. Users spend most of their time on sites other than yours. They therefore expect yours to work like the ones they know. The menu sits at the top, the logo returns to the home page, the cart waits in the top right corner. Breaking these conventions makes the visitor pay a learning cost, the one described in the article on mental load (opens in a new tab).

The useful distinction runs between structure and expression. Structure is the grammar of the site. It says where things are, how to navigate and how to pay. It benefits from staying conventional. Expression is the voice. It gathers the rhythm of the spacing, the typographic choice, the palette, the motion, the way a button answers a hover and the tone of the copy. Nothing in Jakob's Law requires expression to be the same everywhere. Yet the convergence measured by the Indiana study affects both. Websites no longer share only a grammar, they also share a voice, which therefore no longer sets anyone apart.

A study published in 2006 (opens in a new tab) in Behaviour & Information Technology by Gitte Lindgaard and her colleagues showed that the judgement of a page's visual appeal forms within 50 milliseconds and that it stays stable when the page is shown for longer. The visitor has an opinion before reading a single word.

That aesthetic judgement reaches further than pleasantness. In 2003, B.J. Fogg and the Stanford Web Credibility Project team had 2,684 people assess the credibility of websites (opens in a new tab). The look of the design is the criterion mentioned in 46.1% of the comments, far ahead of information structure and the content itself. The Nielsen Norman Group documents the same phenomenon under the name aesthetic-usability effect (opens in a new tab), the tendency to judge as more usable whatever looks more polished.

These two results combine in a way that is uncomfortable for a generic site. The visitor judges in the blink of an eye and judges mostly on form. If that form is the form of ten thousand other sites, the judgement carries no information about you. It carries information about the category. A law firm whose site looks like every other law firm is perceived as an average law firm, whatever the real quality of the firm. The judgement then applies to the whole category and the site adds nothing to it.

The web does not yet have a direct measure of what website sameness costs. Brand marketing, on the other hand, has measured the cost of indifference. The mechanisms involved are the same, attention, memory and price.

In 2024, Peter Field, Adam Morgan and System1 published The Extraordinary Cost of Dull (opens in a new tab), drawing on the IPA databases and System1's advertising tests. In the UK market, a campaign that triggers no emotion has to buy almost £10 million of additional media to reach the effect of an interesting one. Nearly half of the people exposed to an advert feel nothing. The work of Karen Nelson-Field cited in the same report adds a detail that concerns small organisations directly. Established brands absorb dullness, small brands suffer a drop six times steeper.

Kantar reaches the same point through price. In its Meaningful Different Salient framework, 94% of a brand's pricing power (opens in a new tab) is explained by the meaningful difference the public attributes to it. Brands with strong difference in 2006 were twice as likely (opens in a new tab) to still appear in the BrandZ ranking twenty years later.

The Be Distinctive Everywhere (opens in a new tab) report by JKR and Ipsos, published in 2023, finally measures how rare the thing is. Of the brand assets tested, only 15% reach the genuinely distinctive level. For colour, the proportion drops to 4%. Most brands believe they own a colour. Almost none owns one the public attributes to them without the logo.

Transposing these results to websites calls for one precaution. A website is not an advert and these studies did not measure home pages. The reasoning holds through the mechanism, however. A site that nothing sets apart has to buy the attention it does not earn, in advertising, in paid search, in discounts. It defends its price poorly, because nothing it shows justifies a gap with the neighbour. It does not settle into memory and therefore does not benefit from the brand search described in the article on brand demand (opens in a new tab). These three costs spread across three different budget lines, which explains why no dashboard ever adds them up.

Changing tools settles nothing. A hand-coded site using a top-10 typeface, the default palette and standard components looks as much like its neighbours as a generated one. Singularity does not live in the stack, that is, the set of technologies a site is built with, from the CMS to the typeface. It lives in the decisions that stack leaves open.

The defaults the tools provide and the decisions they leave open

Motion deserves a separate mention because it is the least exploited dimension. A study of screenshots does not measure it, for the simple reason that a screenshot is still. It is also the dimension where default tooling decides the least. Two sites can share the same grid and the same typeface and produce an entirely different feeling depending on how elements arrive, respond and leave.

Accessibility is not the price of singularity. Sufficient contrast, complete keyboard navigation and respect for reduced-motion preferences apply to an original typeface as much as to Roboto. The constraint bears on readability and perception, not on neutrality.

That websites look the same is a measured fact, older than AI and amplified by it. It results from individually rational decisions, a proven theme, a solid framework, a readable typeface, a fast generator. None of these decisions is a mistake. Their sum produces a site the visitor has already seen, files into a category within fifty milliseconds and forgets before the end of the day.

No direct measurement yet puts a figure on what that sum costs a website. Brand data gives the order of magnitude by analogy. Perceived difference explains almost all of pricing power, a dull campaign needs millions in additional media to catch up with an interesting one and only 4% of brand colours are recognised without a logo. A site's singularity is decided where the tools leave a choice, in rhythm, typography, palette and motion. Those decisions build an asset that depreciates more slowly than a theme.