If you're worried about title originality, why don't you write a scraper instead of asking for people to submit urls manually? Would be more efficient. But, If you care about real people posting stuff, then they should be allowed submit whatever they see appropriate. To mitigate the frustration, I think you could allow urls to be posted multiple times with a specific scientifically backed grace period of "n-hours" until no more duplicates are allowed. People can still up-vote the titles they like best and the winner takes all, making the other titles appear like collapsed sub-captions below the main-title. Hope you understand what I mean. This would solve two problems at once. A) Custom title's, no censorship. B) Valuable data about winning-titles, that can be used to train a stochastic model to predict the best titles. How you use the data from B) is up to you.

I mean people aren't stupid enough to change the Title of the "Higgs Boson" to "Bananas". Sorry, if this comes over wrong. I respect you and this is just critics on your software's policies.