What this page is demonstrating is a "hyperlink everywhere" reading mode: instead of authorshand-picking a few anchor links, the content is treated as raw text and the UI layer turns pieces of it into interactive targets (it's presented as a short story-style block of prose). The effect is that linking becomes something that can be applied to any text after the fact, rather than something baked into the copy.
At a high level, the feature works by splitting the text into tokens (usually words), wrapping those tokens in lightweight elements, and then using JavaScript to attach interaction states—hover, focus, click, maybe selection—to "activate" a token and make it behave like a link. The same approach can also support richer behaviors than normal anchors, like showing a preview, offeringmultiple destinations, or progressively "revealing" links so the page doesn't look like asea of blue underlines.
Customization is basically a set of knobs around (1) what becomes linkable (every word vs. nouns only vs. matched phrases), (2) when it activates (hover, click, long-press, keyboard focus), (3) how itlooks (highlight style, cursor, underline rules, animation), and (4) where it goes(a deterministic mapping, a search route, a contextual menu, or a resolver function that can return different targets). Even if you keep it simple, just changing tokenization (word-level vs. phrase-level) and activation (hover vs. click) radically changes the feel.
Why it's an interesting UX experiment: it challenges the default "links are editorial choices" model and treats linking as a dynamic lens over content. That opens questions you can actually test—doesdense linkability help exploration or just add noise? do people discover it without instructions? does it interrupt reading flow?—and it forces you to confront practical design constraints like accessibility (keyboard/screen readers), mis-taps on mobile, and the balance between discoverability and visual clutter.