I weighed every character on Notifio's site by its contrast ratio a few days ago, and the single worst thing in that measurement was the footer: the same 14-link column repeated on 38 pages, drawn in #525252 on #0a0a0a, which is 2.53:1 and fails AA at any size.
That post measured the colour. It did not look at what happens when you point at it. This one does, because the answer turned out to be the most interesting thing about the site's CSS.
#525252 on #0a0a0a 2.53:1 footer link, resting #a3a3a3 on #0a0a0a 7.85:1 footer link, hovered A footer link on this site goes from failing to comfortably passing when a pointer touches it. That is a 3.1x change in luminance contrast, triggered by an input device that nobody reading it on a phone has.
Every hover rule on the site
The whole stylesheet is 39,976 bytes and contains five media queries:
@media (min-width:40rem) 1 sm @media (min-width:48rem) 1 md @media (min-width:64rem) 1 lg @media (hover:hover) 2 Those last two blocks are 649 bytes and hold every hover declaration on the site. All eleven of them, unminified:
@media (hover:hover){ .hover\:border-\[\#2a2a2a\]:hover { border-color: #2a2a2a } .hover\:border-\[\#404040\]:hover { border-color: #404040 } .hover\:border-\[\#525252\]:hover { border-color: #525252 } .hover\:border-\[\#737373\]:hover { border-color: #737373 } .hover\:bg-\[\#0f0f0f\]:hover { background-color: #0f0f0f } .hover\:bg-\[\#1e1e1e\]:hover { background-color: #1e1e1e } .hover\:bg-\[\#141414\]:hover { background-color: #141414 } .hover\:bg-\[\#e5e5e5\]:hover { background-color: #e5e5e5 } .hover\:text-\[\#a3a3a3\]:hover { color: #a3a3a3 } .hover\:text-\[\#fafafa\]:hover { color: #fafafa } } @media (hover:hover){ .group-hover\:text-\[\#fafafa\]:is(:where(.group):hover *) { color: #fafafa } } Border colour, background colour, text colour. Not one of them changes a text-decoration, a font-weight, a size or a position. The site's entire interactive vocabulary is "this thing changes colour".
Tailwind v4 is the reason they are inside a media query, and it is the right default: hover styles that stick to an element after a tap are a classic touch bug, so the generated CSS only applies them where the pointer can actually hover. I did not write that wrapper and I would not remove it. It does mean that on a phone, those 649 bytes are downloaded and never match anything.
Against the 1,091 links that are actually on the pages
I then walked every anchor on all 38 live pages and read its classes and computed styles.
anchors 1,091 footer 532 content 348 nav 211 carrying at least one hover token 1,048 change colour only 1,048 add an underline 0 change weight, size or position 0 carrying no hover response at all 43 the nav wordmark, once per page 38 the five cards on /guides 5 The 656 most common ones are hover:text-[#a3a3a3] on a base of #525252, which is the footer measurement above plus the citation links on the fifteen site pages.
And then there are the 43 with nothing. Thirty-eight of them are the wordmark in the nav, which is fine: it is a logo, it reads as a home link by convention. The other five are the five guide cards on the guides hub, and the reason that is worth a paragraph is what sits directly underneath them on the same page:
/guides, content links hover classes "How fast do rental listings..." none "How to be first to a new..." none "What to write in your first..." none "The documents to have ready..." none "How to spot a rental scam..." none "For students" hover:bg-[#0f0f0f] "For people relocating" hover:bg-[#0f0f0f] "For Amsterdam" hover:bg-[#0f0f0f] "See pricing" hover:bg-[#e5e5e5] "Compare the alternatives" hover:border + hover:text Two card grids, one page, and only the lower one responds to a pointer. The sites hub and the comparison hub both give their cards hover:bg-[#0f0f0f] as well, so the five guide cards are the only clickable cards on the site that do nothing at all when you point at them.
What is left when hover is gone
So I asked the narrow version of the question: with no pointer, no hover and no colour perception, how do you know something is a link? The answer for a standalone link in a card or a list is position, which is legitimate. The answer for a link inside a sentence has to be an underline, because WCAG 1.4.1 requires a non-colour cue when a link sits in a block of text.
underlined anchors 90 of 1,091 anchors inline in prose, next to other text 25 of those, underlined 25 of those, the same colour as the text around 20 Twenty-five is a small number for a 38 page site, and it is the right number to care about: everything else on this site is a card, a nav item or a list row. All 25 of them are underlined, so the site passes the one rule most marketing sites fail. Twenty of the 25 are drawn in exactly the same colour as the sentence they sit in, which means the underline is not a decoration there, it is the entire signal. Delete underline from the shared class string and twenty links become invisible prose on the terms, the privacy policy and the about page.
Focus is the third cue, and the stylesheet has exactly two authored focus rules, both of them on the one email field in the app. Every link keeps Chrome's default focus ring, which is the only state cue on this site that works for a keyboard user and the reason the keyboard story is in better shape than the touch story.
The actual problem, stated plainly
Hover is an enhancement. It is not allowed to be the thing that makes text readable, and in our footer it is.
532 footer links resting: #525252 2.53:1 fails AA for all text sizes hovered: #a3a3a3 7.85:1 passes AAA on touch: no hover state exists A reader on a phone gets the 2.53:1 version and has no way to reach the other one. The fix is one token: make #737373 (4.18:1) or #a3a3a3 the resting colour and let the hover go to #fafafa. That is the same number of CSS declarations, it costs nothing in bytes, and it moves the readable state from "if you have a mouse" to "always".
The two smaller ones while I am in there: give the five guide cards the hover:bg-[#0f0f0f] their sibling hubs already have, and keep the 25 prose underlines exactly where they are.
Everything above is measurable from outside. Open the footer of any page, compare it on a phone and on a laptop, and the difference you are looking at is 649 bytes of CSS that one of those two devices has decided not to use.

