Clickbait is effective where it can be employed without consequence. YouTube is one such domain.
It's inordinately effective where algorithms promote material based on short-term appeal rather than long-term benefit.
Where either or both of these factors are addressed, clickbait doesn't disappear entirely (it does hook into human psychology), but its effectiveness and power fall markedly.
YouTube does not support blocking of individual channels by users, or at least didn't as of the last time I used it whilst logged in (some years ago). If I'm presented with content that is absolute bilgewater, I cannot in any way indicate to the system "do not show me content from this source at all ever". This is the equivalent of a privately-functioning super-downvote, and is an exceedingly strong signal YouTUbe are denying themselves over content quality. Its availability would be among the very few possible reasons I'd consider accessing the service other than anonymously (and preferably through a front-end such as Invidious or FixYT, or commandline / terminal tools such as youtube-dl, mpv, or mps-youtube). Eventually, YouTube might themselves consider re-jiggering their recommendations away from channels which are consistently blocked.
In the absence, attention and misdirection games can be played, repeatedly, without risk.
(Oh: and quality contributors such as Derek Muller are also harmed by this as they're competing with scum-sucking bottom feeders for their own far superior productions.)
The other aspect is that the algorithm itself promotes short-term engagement at the cost of not only long-term value and quality, but genuine social harm, as has been well documented over recent years. (Including by Destin Sandler in an excellent 3-part series on online disinformation, the YouTube segment is here: https://yewtu.be/watch?v=1PGm8LslEb4) This means that the existing psychological highjacking is further juiced algorithmically.
On desktop systems where the option is available to me, I use CSS hacks to remove both recommendations and comments from YouTube. The experience improves tremendously. (Stylish: https://addons.mozilla.org/en-US/firefox/addon/stylish/ https://chrome.google.com/webstore/detail/stylish-custom-the... though the uBlock Origin elements remover can also be used for this.)
I've just had an exchange with dang about a recent submission and site which makes no bones about its dedication to pseudoscience and disinformation. Hacker News does have a policy of penalising (though not completely blocking) such sites. In our back-and-forth, dang offered (tongue-in-cheek, I'm pretty sure), that any strict "no bullshit" rule would render much of the present-day Internet inadmissable on HN. The longer I think that over, the more my response is "Yes, and that would be a Good Thing."
(I'm well aware if issues with truth-determination, epistemology, simple errors, politically-motivated censorship, and more. I'm referring here to publishers and authors who specifically and categorically defend posting of bullshit, though that's not the full extent of what I'd object to. I refer to Dante's Inferno in which it is the 8th of the 9 circles to which false counselors, falsifiers, hypocrits, schismatics, corrupt politicians, flatterers, pimps, and seducers are consigned. Violations of trust strike at the very heart of polity and civilisation.)
http://thumbnails-visually.netdna-ssl.com/dante-alighieris-i...
In my experience on the non-commerial, non-algorithmic, and distributed social platforms of Mastodon / the Fediverse and Diaspora, I find that whilst bullshit isn't entirely absent, its role is much reduced.
There is no algorithm to juice its spread.
On Mastodon, instances can elect to federate fully, partially, or not at all with others. Smaller instances can define and tune their interactions with others quite specifically. (The instance I'm on, toot.cat, run by a long-time online friend, has a substantial list of limited and silenced servers: https://toot.cat/about/more)
Diaspora dates from somewhat earlier, and has a more open philosophy toward content, though repeat disinformation, attacks and threats, and spam are dealt with in at least part. Among my contacts, we share information on block-worthy profiles based on various criteria. (Such informal lists are a frequent occurrence, especially in the absence of more powerful tools. They're not without their own risks and failure modes, including inadvertent or unwarranted inclusion.)
The larger point though is that clickbait is a phenomenon actively and intentionally supported by algorithmic social media to serve its short-term benefits. It does so at its own long-term cost.