X adds comment downvotes to improve algorithmic ranking

Listen to the article 3 min

This audio is auto-generated. Please let us know if you have feedback.

The platform formerly known as Twitter is trying out comment downvotes once again, as it works to address the influx of artificial intelligence generated spam that’s impacting overall user engagement.

X’s head of product Nikita Bier shared a preview of the feature in an X post on Wednesday, with a new thumbs-down icon added to the lower function bar on post replies.

When tapped, users will be able to choose a reason for downvoting a reply, with five options to choose from:

• Not interested in this post
• Incorrect or misleading
• AI generated
• Spam
• Report post

Downvotes won’t subtract from a post’s like count, and the intention of the option, Bier says, is not to influence what gets traction more generally in the app (as it function does on Reddit), but is more designed as a user preference tool, with each users’ responses helping to dictate their preferences, while also providing a simplified reporting path.

As explained by Bier: “The Reply Algorithm is currently is the worst product in the company. There is no logic, no signal, just garbage. Replies could be ordered randomly with crypto spam at the top and it would probably be better than the current system.”

As such, Bier is hoping that the new downvote option will provide more signals for X to improve this element, though users have expressed concerns that it could be used for brigade attacks, and mass downvoting of commentary that they disagree with.

In order to combat this, the downvote option will only be available to X Premium subscribers. That should ensure that bots can’t game the process, though it also means that downvotes are only available to around 1% of all X users.

But for those who are paying for the app, they now have another option to consider, which could help them guide the algorithm, and help X to better identify spam and junk replies.

It’s the latest in X’s broader push to fend off the rising onslaught of automated reply bots, with AI tools enabling the creation of more sophisticated X bots that are harder to detect. And with many people looking to use X to influence public opinion, many well-resourced, even state-backed groups are building AI bot armies that are gradually taking up more space within the X stream.

Indeed, Wired recently reported that X has been inundated with AI fakes since the beginning of the U.S.-led incursion in Iran, which X has already taken action to address in other forms.

Reply downvotes are another element within its defenses, which Bier has designated X’s “Iron Slopdome” of advanced protection.

X/Twitter has tried out comment downvotes several times in the past, though never as a direct ranking mechanism, as such, more as a simplified link to report issues with comments in the app.

This new test is along the same lines, and it’ll be interesting to see what data X can glean from the option, and whether it does become a more valuable feedback element, or another vector for spam attacks.