X Is Considering Downvotes for Post Replies

As Elon Musk’s X continues to seek the most “truth seeking” systems to ensure accuracy within online debate, it’s re-examining the use of up and downvotes on post replies, as a means to better rank engagement, and display the most relevant comments within each thread.

Well, downvotes anyway.

As you can see in this example, X is experimenting with a downvote option on post replies, which would give users another way to share their feedback about user comments.

The option was spotted in the back-end code of the app by researcher Aaron Perris, with X engineer Jay Baxter then confirming the internal test.

Though it’s probably not exactly what you’d think.

As per Baxter:

“Community Notes wouldn’t work well without negative rating signal. But you have to be smart about how you use them. If you naively add them all up, you’ll get a hivemind like Reddit. One way: only downrank if you see negative ratings from people who typically disagree.”

So the downvote option for replies would be weighted based on political perspective, so that you couldn’t have users of a certain ideology or viewpoint brigading the replies, and downvoting whatever they don’t agree with into oblivion.

The ultimate outcome, then, would ideally be a system that downranks replies without political bias, so that when there are junk and spam comments, or clearly misleading and untrue statements, they are the ones that get slid down the list, or potentially even hidden, as opposed to silencing dissenting opinion.

It’s an interesting concept, and is similar to how the former Twitter team had also experimented with downvotes in replies, not as a means to express an opinion on the comment itself, but to push down spam and junk.

Back in 2021, Twitter tried out virtually the exact same process, though without the weighting for political leaning.

The idea, then, was for the downvote arrows to indicate relevance and value in post replies, not personal opinion on the expressed comments, in order to improve the quality of its rankings.

As you can see in this image, Twitter actually tried out several variations of the up and downvote UI, though part of the problem at the time was that users weren’t entirely clear on what these buttons meant.

That saw many did use them as an indicator of disagreement, as they’re used on Reddit, which wasn’t the intended purpose.

X is now accounting for this by making the process simpler to understand, in that you will, in fact, downrank comments that you don’t like. X is effectively acknowledging that those on one side of the political divide will undoubtedly downvote opinions that they disagree with, as will the other, but the added weighting will account for this, and ensure that any impacts are only applied where there’s agreement.

Which, in the end, should ensure that only the worst examples of spam and junk are penalized, not opinions as such. Which is a better approach, though a lot does also depend on how X determines political leaning, and whether that process is accurate enough to maximize relevance.

Either way, it’s an interesting experiment, which could help to maximize in-app engagement by helping to highlight the most valuable post replies.

I mean, weighting for political preference is a slippery slope in itself, which that can also lead to certain types of misinformation being upheld in the app, as there are some topics that will never see universal agreement.

But in this context at least, it makes sense as a means to weed out junk comments.

X hasn’t said that this will make it to a live test, but it is in the early experimental phase.