How I Made Google's "Web" View My Default Search

Nah, you need some way to consider the quality of your text. No way around that.

Let's say I publish a page -- nothing fancy to it, no dark patterns at all -- just the word "architecture". For anybody searching for the term "architecture", this is now the perfect page -- a 100% match to their search, with no dark patterns at all. Completely useless, but perfect.

What would my motivation be for doing something like this? Well, imagine that I now deploy 100,000 identical or very similar pages. Your search engine is now fully _nuked_ for those terms. Does your competitor have a product that you don't want being found via the search engine? I'll happily nuke its search terms for you, for a fee of course. That's my motivation. Remember that advertising is not just a competition for you to be seen, but also for your competitors to not be seen. Your search engine, as-is, would be great at enabling the latter.

(Maybe you could fix this by allowing down-votes from logged-in, IP-logged users... but with 100k pages to downvote, that's not gonna help. So maybe you take pages which have been downvoted and do similarity comparisons to pages which haven't. A simple Levenshtein distance is too crude and easily-gamed, but maybe you could feed the downvotes into a neural-network classification system... and down the rabbit hole you go.)

There’s quite a lot of dark patterns to watch out for. One is including text on a page meant to be read by your crawler but which is hidden from users by various CSS tricks. To get around this you’d either need to specifically blacklist certain CSS / DOM structures or do something very radical: render the page and then attempt to OCR it back into text to match against the original text.

Remember when Google said they would penalize websites that use dark patterns like Quora, Pinterest, Twitter, etc? Google can easily penalize those websites for their dark patterns so websites and blogs by actual people rise to the top but they don’t.

I keep thinking of a hierarchical invite model for these sorts of problems. I'd be very curious if someone could give a second opinion on this idea.

The mechanism:

Everyone has to be invited by someone, so it all traces back to the creator. The creator knows they themselves are legit, but let's say someone online asked for an invite and bad inputs keep coming from somewhere down that branch of the invite tree. Either the person the creator invited is a spammer, or someone they've invited in turn is the spammer. All accounts leading up to them can be progressively killed (and their inputs nullified), starting with the ones actually causing trouble → if it keeps happening in that branch then kill one layer up, and so on.

Incentives:

People risk losing their own account when inviting someone who they don't trust to be good netizens. Maybe there needs to be an incentive why you care about your account in the first place, or maybe (looking at Wikipedia or OpenStreetMap, or HN with its voting system and the homepage meaning a lot of attention for your page) a majority of people are simply honest and happy to contribute and that suffices

Problem it solves:

Wouldn't such a hierarchical invite system work around the online identification problem?

If you would DNA check everyone and ban people who abused the system in the last decade, you'd also not have any spam online, but that's way too invasive (besides not being legal and prohibitively expensive, it's also not ethical). However, a pseudonymous (all that is known about you is a random user ID) invite tree seems to me like it would have similar properties. It requires banning the same person perhaps a hundred times until they run out of people that will give them invites, but wouldn't it eventually distill the honest people from the population? (Which is probably almost everyone if there is no gain from systematic cheating and there's social pressure to not ask for multiple invites because account holders know that means you were either messing with the system they enjoy using or invited someone onto it that did that.)

(Implementation details: One bad input isn't an instant ban: people misclick or misunderstand interfaces, but eventually it gets to a point where, if they can't click the right buttons, there's also no point having them be moderators of the search engine (or whatever this is used for) and so their account is removed. If multiple removals happen in a tree that's deep and recent, remove more than one layer at a time to get rid of malicious sockpuppet layers. The tree's maximum depth can be limited to something on the order of 50: it doesn't take many steps to find a chain of relationships that links two random persons on the planet, so a fairly low depth is enough for the whole world. People should be told on the invitation page how many bad apples were removed in each layer below them, so if they're 1 bad apple removed from having their own account pruned then they know to only invite people they're very sure about. One problem I see with the system is that it reveals social graphs which not everyone is happy about. If that means being able to kill virtually all spam, content farms, etc., maybe it's worth it, but or course more research is needed beyond an initial proposal.)

"I invented a ranking system that cannot be gamed!"

By my experience, ALL ranking systems can - and will be gamed, no matter what. I'm still willing to give you the benefit of the doubt - maybe you are the one who can finally truly come up with an un-game-able system. But I'll consider that an extraordinary claim - and will require to see extraordinary proof for that.