I agree with Tony — writing a program to flag a new story that is n% similar to an existing Medium story isn't hard. The hard part is having people in the loop to actually read the stories to decide if the similarity is intentional plagiarism as opposed to, say, both authors quoting the same source verbatim.

But that's where having a continuous feedback loop comes in, where the human judge makes tweaks to the percentage of similarities, decides that n% similarity is not wroth a look but n+x% similarity is, while n+y% similarity is an automatic shutdown. So cost is in the human side, not the programming. (But if Tony disagrees, listen to him!)