If you do the counting properly, this can be taken into account. Just like in poker, where you can bluff sometimes but if you do it always, people will call you upon. It's called Reification Done Right (in quadstore terminology), where you store metadata about the statement. Then it's just about building a reputation score for each statement. You can use simple bayesian updates like it's done in spam filters. Pretty quickly there is very little incentive to lying in an open book world.
Even more interestingly you can compute a score between author of the statement and viewer of the statement, basically telling you whether or not you can trust the information given your current belief system. For example if you are a flat-earther, you can get spaceX news filtered quite easily. From the computational point of view it's faster not to compute the whole matrix of score but use some lower dimensional embeddings so you do a matrix factorization, and that's how you get all the user recommendation systems ala Netflix.