Mobile Resilience Index 2026: How 34 Top Websites Hold Up on Phones

By MobileCheck · Data collected October 20, 2026 · Updated October 20, 2026 · 34 sites · CC BY 4.0

Preliminary results. These numbers come from a pilot sample of the 34 highest-ranked sites. The full study is in progress; this page updates automatically as the data grows.

38% of the 34 most popular websites break when a phone user enlarges text to 150%, and only 21% switch to a dark theme when the phone is in dark mode. We loaded each homepage on iPhone 14 Pro and Samsung Galaxy S22 profiles, then repeated every test in dark mode and with 150% text.

Key findings

• 38% of sites break at 150% text: 15% cut off text, 9% overlap text and 21% start scrolling sideways. 35% break on both phones.
• 68% declare dark-mode styles, but only 21% visibly switch to a dark theme; 24% have a dark design even in light mode.
• 21% have text below WCAG AA contrast, and 21% have low-contrast text in dark mode — 0% of sites with a dark theme get more low-contrast text in it than in light mode.
• 38% show text smaller than 12px on a phone (12% have ten or more such elements); median body text is 16.3px.
• 24% have tap targets under 24×24px crowded by other targets; 3% scroll sideways even at normal text size; 15% block pinch-zoom.
• The median homepage weighs 1.3 MB, with a median LCP of 1.2 s (9% weigh over 3 MB; 18% paint after 2.5 s).

Large text: what breaks at 150%

People with low vision — and many who just want to read without glasses — raise their phone’s text size. We enlarged every font on each homepage by 50% and compared the layout with the normal view. Problems that already existed at normal size don’t count; only what the larger text caused.

The usual culprits are fixed heights on buttons and cards, labels with white-space: nowrap and overflow: hidden, and navigation bars sized in pixels. Our guide Large text: keep your layout from breaking shows the CSS fixes; overlapping and clipped text covers the two most common symptoms.

Dark mode: declared vs. delivered

With the phone set to dark mode we checked two things: whether the site has dark-mode styles at all, and whether its first screen actually turns dark. Many sites ship a prefers-color-scheme rule for a widget or an icon while the page itself stays bright.

See how to support dark mode without losing contrast.

Everyday readability on a phone

Fixes: low-contrast text, text too small, tap targets, content wider than the screen.

Scores

Each test produces MobileCheck’s readability score from 0 to 100 (85+ is good, 60–84 needs work, below 60 is poor). A site’s Resilience score is the average of its normal, dark-mode and 150%-text scores across both phones. The median site scores 93.

Results for every site

Select a column heading to sort. Scores are 0–100 averaged over both phones. “Text <12px” and “Low contrast” count elements on the worse of the two phones. Rank is the site’s position in the Tranco list.

Method

Sample

The homepages (https://domain/) of the highest-ranked websites in the Tranco list (ID GQG5K, October 19, 2026), taken top-down. We skipped infrastructure domains without a website for people (CDNs, DNS, ad and analytics endpoints), link shorteners, adult sites, and the same brand on another country domain. 85 entries were skipped this way. Of 48 sites attempted, 14 were excluded after testing (bot protection, load errors, HTTP errors, pages without auditable text, or redirects to a site already in the sample), leaving 34.

Devices and engine

iPhone 14 Pro / 15 / 15 Pro / 16: 393×852 CSS px, device pixel ratio 3; Samsung Galaxy S22 / S23 / S24 / S25: 360×780 CSS px, device pixel ratio 3. Both profiles were rendered in Chromium with each device’s viewport, pixel ratio, touch input and user agent; the iPhone profile therefore shows the page iPhone visitors are served, laid out by Chromium rather than Safari’s WebKit engine. One page load per device, browser bars subtracted from the viewport, browser language en-US.

Scenarios

Normal: the page after loading, closing cookie banners and scrolling once to load lazy content. Dark mode: prefers-color-scheme: dark. 150% text: every element’s computed font size (and pixel line height) multiplied by 1.5.

Thresholds

Very small text: under 12px as rendered (after any page scaling). Contrast: WCAG 2.2 AA, 4.5:1 for normal text and 3:1 for large text (24px, or 18.66px bold). Tap targets: under 24×24px with another target within 12px of the centre (WCAG 2.5.8). Sticky bars: fixed elements covering over 25% of the screen. LCP over 2.5 s, CLS over 0.1, page weight over 3 MB.

Breaks at 150% text

Compared with the normal view on the same phone: the page becomes wider than the screen, or more text is clipped, or more text overlaps other text. A site breaks if this happens on at least one phone.

Dark theme

The site declares dark-mode styles (a prefers-color-scheme: dark rule, color-scheme or a color-scheme meta tag) and the mean luminance of its first screen drops by at least 0.15 compared with light mode. “Dark design even in light mode”: first screen mean luminance below 0.25 in light mode.

Percentages

“Share of sites” counts a site if the problem appears on at least one of the two phones. Each percentage is out of the sites where that scenario completed (shown as “x of n sites”).

Limitations

• Only homepages were tested; inner pages, logged-in areas and app-like flows may behave differently.
• The iPhone profile was rendered in Chromium, not Safari. Font metrics and a few CSS features differ, so some iPhone-specific bugs are missed. Real phones also use different system fonts.
• Enlarging every font size by 1.5 approximates the phone’s text-size setting and browser text zoom; sites that respond to -apple-system-body or Dynamic Type may react differently on a real device.
• Automated checks can’t judge everything: text in images, ads and content inside cross-origin iframes is not audited, and contrast over background images is skipped.
• Sites vary by region, A/B test and time of day; one load per device is a snapshot. Bot-protected sites are excluded, which may bias the sample toward sites without aggressive bot protection.
• LCP and CLS were measured once in a lab setting, not from real users.

Download the data

Per-site results for all 34 sites, plus the excluded sites and the reason, under CC BY 4.0.

Fix it on your own site