Freeland post puts attention on how Twitter decides to label tweets ‘manipulated’

When asked this question, Twitter referred The Canadian Press to its synthetic and manipulated media policy, which it confirmed it used to label Freeland’s tweets.

Twitter often refers to the policy when people flag tweets using the platform’s reporting process. The company currently allows people to report tweets as misleading in the U.S., Australia and South Korea, but the option is not yet available in Canada.

That means Twitter often finds out about Canadian tweets that may be misleading when users report posts for other violations, when third parties bring them to the company’s attention or when technology it uses to unearth potential violations discovers an issue.

To determine if a tweet should be marked as manipulated, Twitter uses three questions embedded in its synthetic and manipulated media policy.

The first — Is the content synthetic or manipulated? — involves a look at whether the video’s composition, sequence, timing, or framing has been changed. It also examines if visual or auditory information like overdubbing, modified subtitles or frames have been used and whether the content depicts a real person.

Twitter’s policies show it is more likely to deem a tweet synthetic or manipulated based on the first question if it has wholly-synthetic audio or video or has been spliced, reordered or slowed down to change its meaning.

However, Twitter also looks for more subtle forms of manipulation, such as isolative editing, omission of context or presentation with false context.

It does not consider retouching photos or colour-correcting video in ways that don’t alter the meaning of content to be manipulated.

Twitter then considers its next question: Is the content shared in a deceptive manner?

To determine that answer, the company analyzes the metadata embedded in the tweet, information on the profile of the account sharing it and websites linked in that profile and in the post.

Twitter then arrives at its final question: Is the content likely to impact public safety or cause serious harm?

The company considers whether the tweet would cause threats to physical safety, risk of mass violence or widespread civil unrest, stalking or unwanted or obsessive attention and voter suppression or intimidation.

Media that are found to be synthetic or manipulated after all three questions are subject to removal and accounts found violating the policy several times may be suspended.

When only some of the criteria have been met, Twitter often labels the tweet as misleading.

In most cases, it also shows a warning to people before they share or like the content, reduces its visibility, provides a link to additional explanations and turns off likes, replies and retweets for the post in question.

When the visibility of a tweet is reduced, it doesn’t appear in search results, replies and timelines and isn’t recommended for use by Twitter’s algorithms that can sometimes draw attention to popular tweets.

The company also has a civic integrity policy that tells people they can’t use the platform to manipulate or interfere in civic processes by “sharing content that may suppress participation or mislead people about when, where, or how to participate in a civic process.”

Violations of that policy, which Twitter didn’t cite in Freeland’s case, result in the company often labelling, removing or minimizing the tweet’s visibility.

This report by The Canadian Press was first published Aug. 23, 2021.

Tara Deschamps, The Canadian Press

by Canadian Press

Freeland post puts attention on how Twitter decides to label tweets ‘manipulated’

Our Community