17 Studies Plus Advertising Data Map Out Twitter’s Decline

I rarely look at Twitter these days, but it’s obvious people are having extremely different experiences there. I see some people talking of it spiralling into its end stage, a right-wing haven with rampant abuse. The word “cesspool” comes up often. Meanwhile others say it’s still the place to be, that “everyone” is there. Impressions are surely distorted by people’s social media bubbles and biases – including wishful thinking.

To try to get a handle on this, in this post I ask 5 questions that get at what’s happening from different angles. And I’ve used 2 broad types of data to try to answer them. The first type comes from the group of 17 studies in this post’s title. That includes studies analyzing tweets and Twitter accounts, mostly using Twitter’s sample data feed for researchers (via API), as well as surveys of internet/social media users. I list them all here, but skim over them. If you want to dig into a more detailed summary of those studies, that’s in my post tracking Twitter and Twitter-Mastodon migration studies here.

The second type of data is on advertisers, including the somewhat murky data that Twitter provides to advertisers.

The 5 questions start with whether Twitter use has declined, and end with whether the drop in advertisers could be an existential threat. My main takeaways? There has been a very substantial drop in Twitter use – except for hate speech, misinformation, and far-right tweets, which boomed immediately after the takeover. And there’s enough data to get a rough idea of the magnitude of the overall decline in quality and quantity of tweets.

Has Twitter use declined?

Yes. There was a substantial decline in the months after Musk took over. The decline wouldn’t be consistent across the board: Some Twitter networks may be dead in the water, while others may be fairly intact. And as you’ll see in later sections, several studies show a boom in far-right activity, hate speech, and misinformation.

Three studies put the decline in use around 20% or 30%. The decline is in part people stopping tweeting altogether, and partly a reduction in tweeting. The studies were a survey of general users in the US, a study of tweets in German, and another on users participating in the climate debate in English and French.

Twitter’s data to advertisers estimated the drop in global advertising reach to be over 32%, suggesting that this high rate of decline might be globally representative.

There was a survey in the US with a lower rate, but it didn’t include data on use or ask about reduced use. In that survey, about 10% of Twitter users said they were no longer active at all. A study of traditional media Twitter accounts found a drop in engagement (6%).

Sources (most recent first):

• Analysis of about 90% of all tweets in German between December 2020 and May 2023, including network analysis (1.2 billion tweets). (Study in German – my summary is my own translation) (Hammer and Schories 2023)
• Survey of 10,701 people from a representative panel of US internet users, oversampled with Asian, Black, and Hispanic users, in March 2023 (American Trends Panel, Wave 123) (Pew Research Center 2023b, 2023c)
• Analysis of the climate debate on Twitter in English and French from 2016 to March 2023, using 400 million tweets (Chavalarias 2023)
• Survey of 11,004 people from a representative panel of US internet users in December 2022, with an analysis of respondents’ tweets from January 2022 to April 2023 (American Trends Panel, Wave 123) (Pew Research Center 2023a)
• Analysis of 514 misinformation super-spreader Twitter accounts from September 2022 to the end of the year, with a comparison group of 130 Twitter accounts from traditional media outlets (Carniel 2023)
• Survey of more than 24,000 people in the US surveyed in December 2022 and January 2023, with comparisons to similar surveys at 3 prior points in 2022 (The Covid States Project) (Schulman 2023)
• Twitter data on advertising global reach, quoted in a discussion of multiple sources of indirect data on Twitter use (Kemp 2023) (Not counted as one of the formal studies)
• Analysis of Twitter Blue subscribers who signed up from November 9 to November 11, including a sample of 961 associated with far-right networks and 933 that were not (Barrie 2022)

Has Twitter tilted or swung to the right?

Yes, it has, though users with center-left political leanings were still the majority of highly active Twitter users in a December 2022 survey in the US. However, in the US at least, the Twitter users who have been leaving or reducing their activity have been predominantly from the center-left. People from the center-left who are still using Twitter have been more likely to say there are major problems with the platform in the Musk era.

People from demographic groups likely to be the target of abuse from the right – including women and black Americans – also reported higher levels of concern and reduced use of Twitter.

Meanwhile, activity on the far right has boomed. For example, this was the only cluster with increased activity in a large analysis of German-language Twitter. And an analysis of Twitter Blue subscribers associated with far-right networks found a 70% increase in retweets after the takeover compared to a 28% increase for the subscribers who didn’t associate with far-right networks. (Algorithmic favoritism is one of the selling points of Twitter Blue subscriptions, so it’s a group likely to get increased engagement.)

As well as the studies addressing this directly, the later questions mapping the rise in hate speech and misinformation, including on climate change, also speak to the increase of far-right activity on Twitter.

Sources (most recent first):

• Study of over 1.2 billion tweets in German from 2020 to 2023 (Hammer and Schories 2023)
• Survey of US internet users in March 2023 (Pew Research Center 2023b; 2023c)
• Survey of US internet users in December 2022, with analysis of their tweets from January to April 2023 (Pew Research Center 2023a)
• Survey of Americans in December 2022 and January 2023 (Schulman 2023)
• Analysis of a sample of 1,894 Twitter Blue subscribers (Barrie 2022)

Has hate speech increased?

Yes. There’s definitely been a change in moderation standards, and people who want to tweet hate know it and are taking advantage of it. Even where Twitter is taking tweets down, a group of researchers pointed out they just can’t keep up.

There was an explosion of hate speech immediately after the Musk takeover on October 27, 2022. For example, the day afterwards, use of a particular set of hate speech rose from 80 tweets an hour to 4,500 an hour afterwards. A longer term analysis found hate speech quadrupled early on, then settled down to a rate much higher than previous to the takeover. Another of anti-semitic tweets up to February 2023 found the rate of the group they could identify went from one over 6,200 a week to over 12,700 a week.

Musk’s interactions on his own Twitter account, and its popularity with Twitter users from hate groups, is another bellwether. And Twitter is reinstating accounts previously banned for hate speech.

I think this conclusion of one set of researchers reflects the contents of the whole group of studies on this question: “Cumulatively, results would seem to indicate what results of a less stringently moderated Twitter may look like.”

Sources (most recent first):

• Network analysis before and after the takeover of the relationship between Musk’s own account and those of people from hate groups (Auten 2023)
• Lexicon and filter to identify hate speech, as well as bot activity, around 4 weeks before and after the takeover (Hickey 2023)
• Search for tweets using both a keyword for LGBTQ+ and specific slurs (without sentiment analysis) (Center for Countering Digital Hate 2023b)
• Tally of tweet impressions for 10 reinstated accounts, previously banned for hateful speech or misinformation (Center for Countering Digital Hate 2023a)
• Analysis of anti-semitic tweets between June 2022 and early February 2023 (Miller 2023)
• Analysis of the use of a homophobic slur on Twitter in November after a mass shooting at a LGBTQIA nightclub (Benton 2022)
• Analysis of anti-semitism on Twitter between January and November 2022, particularly from October (Jikeli and Soemer 2022)
• Fact check of Musk’s claim about hate speech declining at Twitter in November 2022, using a social media analytic service (Center for Countering Digital Hate 2022)
• Analysis of hate terms (particularly racist, anti-semitic, and homophobic slurs) with sentiment analysis, from October 22 to 28, 2022 (Benton 2022)

Has misinformation increased?

Yes. Again, by a lot. So much so, that, for example, researchers conducting a particularly intensive longterm study of the climate science debate in English and French say that post-Musk, Twitter can now be considered an “active place of conversion to climate skepticism.” Where previously, climate change denialist tweets were half those of pro-climate-science/concern ones, the drop in Twitter activity from the non-denialists now makes them roughly equal.

Engagement with misinformation superspreaders has been shown to be booming, and Twitter is reinstating accounts previously banned for misinformation – including notorious superspreaders. A project has been monitoring sharing of links on Twitter and Facebook from news websites that are rated as frequently “iffy” since 2016. The rate of sharing these links (weighted for engagement with them), was higher post-Musk than at any other time than after the 2020 US presidential election.

Incidentally, if you think Facebook is bad for sharing misinformation, the proportion of “iffy” sharing has mostly been higher at Twitter since early 2019. Since Musk, it’s been consistently 2 or 3 times as high as Facebook.

Sources (most recent first):

• Analysis of most-shared news articles on Twitter and Facebook between 2016 and the end of May 2023 (University of Michigan’s Iffy Quotient).
• Analysis of the climate debate on Twitter in English and French from 2016 to March 2023, using 400 million tweets (Chavalarias 2023)
• Analysis of 514 misinformation super-spreader Twitter accounts from September 2022 to the end of the year, with a comparison group of 130 Twitter accounts from traditional media outlets (Carniel 2023)
• Tally of tweet impressions for 10 reinstated accounts, previously banned for hateful speech or misinformation (Center for Countering Digital Hate 2023a)

Is there a decline in advertising that’s big enough to be an existential threat to Twitter?

Yes. As I mentioned above, Twitter is telling advertisers that global advertising reach on their platform has dropped by close to a third. If it gets worse, or doesn’t recover, the drop in advertising could threaten Twitter’s survival. In the 5 weeks from April 1 this year, Twitter advertising revenue was reportedly around half what it was in the previous year, and at least some on the inside aren’t optimistic about advertising prospects – with a growth in advertising for online gambling etc potentially making Twitter’s image worse. Insiders also reported that there’s a growth in pornography.

Advertisers are leaving because of the combination of concern about harm to their image of their promoted tweets appearing alongside hate speech etc, and the decline in Twitter use. The problematic tweets seem likely to get worse. In recent weeks, US Fox News has come under fire from the right, and Twitter may have positioned itself as the new center of the right-wing media, just as a US presidential gets underway – and the company’s head of trust and safety resigned. And more accounts of previously-banned far-right extremists are being re-instated.

Source: Twitter data on advertising global reach, quoted in a discussion of multiple sources of indirect data on Twitter use (Kemp 2023) (Not counted as one of the formal studies).

Interested in Mastodon? Check out my Shortcuts to Giving Mastodon a Try.

You can keep up with my writing from here and elsewhere at my newsletter, Living With Evidence. And I’m @hildabast@mastodon.online on Mastodon.

~~~~

Other Absolutely Maybe tagged Mastodon and/or Twitter

Disclosures: I joined Twitter in October 2010. My now-inactive Twitter account had just under 10,000 followers before the Covid-19 pandemic, and peaked a few hundred above 32,000. They were under 31,000 as of writing. I joined Mastodon on October 31, 2022, and currently have over 4,800 followers there. (PLOS is on Mastodon, as are PLOS journals.)

The cartoon at the top of this post is my own (CC BY-NC-ND license). (More cartoons at Statistically Funny.)The image in the thumbnail at the foot of the post is public domain. It includes a data chart of mine, and a public domain Mastodon mascot via Wikimedia Commons.