Design Token Drift: We Audited 375 Sites. Only 7.5% Got It Right.

Last updated: May 21, 2026

Average design token coverage across 375 product-company websites was 40.4% (median 38%), and only 7.5% of sites achieved 90%+ coverage. 17,731 hardcoded values were found across the dataset, averaging 47 per site. Tailwind sites averaged 44.1% coverage; CSS custom property sites averaged 39.5%.

Design tokens are supposed to be the single source of truth for color, spacing, typography, and elevation across a product. Define them once, reference them everywhere, and your UI stays consistent. That is the theory.

We wanted to know how the practice compares. So we scanned 375 production websites belonging to companies with dedicated design teams. Product designers, design directors, frontend engineers. Companies that invest in design infrastructure. If anyone should have strong token coverage, it is these teams.

The results suggest that token drift is not a niche problem for teams with weak systems. It is the default state, even for teams that have built token systems and actively use them.

Note: Key Takeaways

Methodology

This analysis comes from OverlayQA's automated token audit, which runs as part of our Design System Audit workflow. For each site, the scanner identifies the token strategy (CSS custom properties at :root, Tailwind utility classes, or no system detected), counts declared tokens and unique computed values, calculates coverage as the ratio of token-backed values to total values, and flags hardcoded values that bypass the token system.

The 375 sites in this dataset are production websites belonging to product companies with dedicated design teams. The population was identified through roles: product designers, design directors, design leads, and frontend developers at companies using Figma. This is not a random sample. It is a sample of teams that invest in design. The bias makes the findings more striking, not less.

All scans ran against the production homepage of each site. Token coverage on interior pages (dashboards, settings, documentation) may differ. Homepage coverage tends to be higher than interior pages because homepages receive more design attention.

The Coverage Distribution: A Long Tail of Drift

If design tokens were working as intended, you would expect most sites to cluster at 80-100% coverage. Instead, the distribution skews low.

Token CoverageSites% of Total
0% (no token usage)4311.5%
1-24%10227.2%
25-59%12232.5%
60-89%8021.3%
90-100%287.5%

38.7% of sites had less than 25% token coverage. Nearly 4 in 10 production sites are running with a token system where most values are hardcoded. The largest cluster (32.5%) sits in the 25-59% range: tokens exist and are partially adopted, but more than half of CSS values bypass them.

Only 28 sites (7.5%) achieved 90%+ coverage. These are the exception. And even among them, "90%" means roughly 1 in 10 values still bypasses the token system.

CSS Custom Properties vs. Tailwind: Which Drifts Less?

79% of sites in the dataset use CSS custom properties (variables at :root). 21% use Tailwind. The coverage gap between them is small but consistent.

ApproachSitesAvg CoverageAvg Hardcoded Values
CSS Custom Properties29839.5%47.8
Tailwind7744.1%45.2

Tailwind sites average 4.6 percentage points higher coverage and slightly fewer hardcoded values. The difference is modest, but it is directionally consistent with Tailwind's design philosophy. Utility-first CSS makes it harder to introduce arbitrary values because the framework provides a constrained set of classes. To use a hardcoded value, a developer must explicitly break out of the utility system with an arbitrary value bracket or custom CSS. With custom properties, there is no friction: a developer can type padding: 14px instead of padding: var(--spacing-md) and nothing stops them.

Within Tailwind sites, a further split emerged. Sites using Tailwind v4 (classified as "tailwind-full" in our scanner) averaged 55.0% coverage across 59 sites. Sites using older Tailwind with partial token adoption ("tailwind-sparse") averaged just 8.4% across 18 sites. The version and configuration of Tailwind matters more than whether you use Tailwind at all.

The Hardcoded Value Problem: 47 Per Site

17,731 hardcoded CSS values were detected across 375 sites. That is an average of 47 values per site that bypass the design system entirely.

Hardcoded values are the measurable evidence of token drift. Each one represents a CSS property that could reference a token but does not. Some are deliberate: a one-off marketing section might justify a custom color. Most are not deliberate. They accumulate through developer shortcuts, copy-pasted CSS, AI-generated code that does not know about the token system, and third-party components with their own styling.

Every hardcoded value creates a maintenance cost. When the design system updates --color-primary from #3468F8 to #2563EB, the 47 hardcoded values do not change. They become silent inconsistencies: elements that look almost right but do not match the system.

93% of Sites Have Tokens. Most Do Not Use Them Consistently.

Out of 410 sites with token system data, 383 (93%) had either CSS custom properties or Tailwind utility classes in place. Only 27 sites (7%) had no detectable token system at all.

This finding shifts the framing of the problem. Design token drift is not caused by teams failing to adopt tokens. Adoption is high. The problem is that adoption and usage are different things. A team can define 100+ CSS custom properties and still hardcode values throughout their codebase. In fact, that is exactly what the data shows.

Declared TokensSites% of Total
0 tokens368.8%
1-105713.9%
11-304711.4%
31-10012530.4%
100+14635.5%

66% of sites have 31 or more declared tokens. These are teams that have invested in building a token system. The gap between declared tokens and actual usage is the definition of drift.

Color Token Coverage: The Most Visible Drift

Color is where token drift is most visible to users. A mismatched button color, an inconsistent text shade, or a slightly-off background all erode brand consistency.

Across 374 sites with color data, average color token coverage was 51.5%. Teams used 4,691 total unique colors, of which 1,618 (34.5%) were covered by the token system. That means nearly two-thirds of colors in production are not referenced by the declared color palette.

Color coverage is higher than overall token coverage (51.5% vs. 40.4%), which makes sense. Colors are the most obvious category to tokenize, and most design systems start with a color palette before expanding to spacing, typography, and elevation tokens. But even with this head start, nearly half of production colors bypass the system.

Component Drift: Links Are the Worst Offender

Beyond token-level drift, we measured component drift: how many distinct visual variants exist for each component type on a given page. A link element with 3 style variants is normal (navigation, inline text, footer). A link element with 7 or 8 variants suggests uncontrolled styling.

Links were the most common drifting component across the entire dataset. On sites with measurable component drift, link elements averaged 4-8 distinct style variants per site. This pattern is consistent with how links are styled in practice: different sections of a page often override link color, weight, and decoration independently, creating variant sprawl.

Other commonly drifting components included anchor elements (effectively the same pattern as links, measured at the DOM level), section containers with inconsistent padding, and heading elements with multiple font-size/weight combinations.

What This Means for Design Teams

The zeroheight Design Systems Report 2026 found that only 8% of teams describe their design system as "very stable," while 44% report instability. Our data explains why: even teams that build token systems are not fully using them. The 40% average coverage we measured means the typical design system is a partial source of truth, not a complete one.

A design system's value depends on consistency. When 60% of CSS values bypass the token system, the consistency promise is only partially delivered. The design system creates value, but token drift limits how much.

The Tailwind finding is instructive. Constraint-based systems that make drift harder (not impossible, but harder) outperform permissive systems where any value is equally easy to use. The same principle applies beyond CSS frameworks: linting rules that flag hardcoded values, CI checks that measure token coverage, and runtime audits that catch drift in production all add friction in the right places.

Three Steps to Reduce Token Drift

How This Compares to Our Product Hunt Study

In April 2026, we ran a smaller study auditing 43 Product Hunt launches. That study found 37.7% average token coverage, compared to 40.4% in this larger dataset. The Product Hunt sample skewed toward AI-built apps and early-stage products. This dataset skews toward established companies with design teams.

The convergence of these two numbers is notable. Whether you are a startup launching on Product Hunt or a growth-stage company with a design director and dedicated design system, token coverage lands in the same 37-41% range. Token drift appears to be a universal problem at this stage of design system maturity, not a sign of neglect.

Run a free design system audit on your site to see your token coverage, hardcoded value count, and component drift score.

Frequently Asked Questions

What is design token coverage?

Design token coverage is the percentage of CSS values in a production page that reference design tokens (CSS custom properties or utility classes from a design system) rather than hardcoded values. A site with 80% token coverage has 80% of its CSS values flowing through the token system and 20% hardcoded.

What is a good design token coverage percentage?

Based on this study of 375 sites, 90%+ coverage puts you in the top 7.5% of sites. 60-89% coverage means your token system is well-adopted. Below 40% (the average) suggests significant token drift. The goal is not 100% since some values legitimately fall outside the token system, but most teams have substantial room to improve.

What causes design token drift?

The most common causes are developer shortcuts (typing a literal value instead of a token reference), copy-pasted CSS from external sources, AI-generated code that does not reference the token system, third-party components with embedded styling, and CSS specificity conflicts that override token values at runtime. For a detailed breakdown, see our guide to design system drift.

Does Tailwind prevent design token drift?

Partially. In our data, Tailwind sites averaged 44.1% token coverage vs. 39.5% for CSS custom properties. Tailwind's utility-first approach adds friction to hardcoded values (you must use bracket notation to bypass the system), but it does not eliminate drift. Tailwind v4 with full token integration performed significantly better (55.0% average) than older Tailwind with partial adoption (8.4%).

How do I measure design token drift on my site?

You need to compare declared tokens against computed CSS values in the browser. Static analysis tools check source code, but runtime auditing catches drift caused by specificity conflicts, third-party overrides, and build pipeline transformations. OverlayQA's Design System Audit workflow does this automatically: it scans your rendered page, identifies your token strategy, and reports coverage percentage, hardcoded values, and component drift.