The Musk Bump: Quantifying the rise in hate speech under Elon Musk
Since his formal takeover of Twitter on October 27, Elon Musk has repeatedly promised to clamp down on hate speech while taking unilateral policy decisions, such as reinstating abusive accounts, that encourage the spread of hate speech on his platforms.
- On the day Mr Musk completed his deal to buy Twitter, he promised that Twitter would not become a “free-for-all hellscape, where anything can be said with no consequences”.
- A week later, he tweeted: “Twitter’s strong commitment to content moderation remains absolutely unchanged. In fact, we have actually seen hateful speech at times this week decline *below* our prior norms, contrary to what you may read in the press.”
- On November 18, three weeks after taking over the platform, he tweeted to claim that Twitter’s new policy would be “freedom of speech, but not freedom of reach” and that hate tweets would be “max deboosted & demonetized”.
- On November 23, he tweeted: “Hate speech impressions down by 1/3 from pre-spike levels. Congrats to Twitter team!”
In our work quantifying hate speech on Twitter, we face a fundamental asymmetry of information. Those inside Twitter have enormous access to its data, including everything from engagement to impressions. But despite Elon Musk’s characterization of Twitter as a “public town square”, access to this information isn’t shared freely.
Earlier in November, researchers at the Center performed a rapid review of hate speech on Twitter, and since then we have been hard at work investigating this issue in more depth.
This blog collates some of our recent work in this area: the picture that emerges is of a “Musk Bump” in the volume of hate speech in Twitter, and a failure to address dangerous anti-LGBTQ+ slurs that stands contrary to Musk’s promise that hateful tweets would be “max deboosted”.
1. The number of tweets containing slurs is far higher under Musk
On Friday, the New York Times published findings from the Center for Countering Digital Hate on how the volume of hate speech on Twitter has risen since Musk took over.
The research shows that despite Musk’s claim to have cut the number of times that users view tweets containing hate speech, there has been a notable bump in the volume of tweets containing slurs being posted since he took over.
On Friday, Musk tweeted to claim: “Hate speech impressions continue to decline, despite significant user growth!”
However, we found:
- There have been 3,876 daily tweets mentioning the racist n-word on Twitter, up 202% on the average daily rate in 2022 before his takeover of 1,282
- There have been 3,964 daily tweets mentioning the homophobic term f****t on Twitter, up 58% on the average daily rate in 2022 before his takeover of 2,506
- There have been 17,937 daily tweets mentioning the misogynist term c**t on Twitter, up 33% on the average daily rate in 2022 before his takeover of 13,514
- There have been 5,117 daily tweets mentioning the transphobic term t****y on Twitter, up 62% on the average daily rate in 2022 before his takeover of 3,159
Our research is based on over-time data from Brandwatch, a social media analytics tool that provides data on the volume of tweets that match a given query over time.
Brandwatch’s figures cover all mentions of a given slur or its plural equivalent in English-language tweets worldwide and include retweets and quote retweets. The data provider says it has “unlimited access” to Twitter data via the Firehose API.
The approach of analyzing tweets containing hateful slurs is the same technique used by Twitter itself. Twitter’s former Head of Trust and Safety, Yoel Roth, previously tweeted about efforts to tackle tweets containing hate speech under Musk’s ownership of the platform, using metrics based on impressions for tweets containing at least one hateful slur.
Musk’s changed emphasis on impressions attempts to obscure the jump in actual usage of slurs on the platform. While Musk has promised greater transparency, Twitter does not currently provide data on tweets impressions to researchers and has not published any data about the volume of hate speech on the platform since Musk took over.
On November 23, Musk tweeted to claim that “hate speech impressions” were down by a third from “pre-spike levels”. His tweet was accompanied by a graph showing the number of “hate speech impressions” declining after a spike when he first took over the platform.
However, CCDH analysis shows that the volume of tweets containing hateful terms remained far above pre-Musk levels during the week in which he made the claim.
|2022 daily average before takeover||1,282||13,514||2,506||3,159|
|Average after takeover||3,876||17,937||3,964||5,117|
|The Musk Bump||202%||33%||58%||62%|
|Average during week Musk made the “pre-spike levels” claim (21st–27th November)||4,650||18,998||4,246||5,290|
2. Engagement on tweets containing slurs is up under Musk
In one of his tweets, Musk claimed that hate speech would be “max deboosted” under a new policy of “freedom of speech, but not freedom of reach”.
But average engagement with tweets containing slurs has jumped since Musk took over, according to analysis using SNScrape, a Python package that gathers data using Twitter search queries.
In the two weeks running up to Musk’s takeover, the average number of replies, retweets and likes on tweets containing at least one of the slurs “n****r”, “tr**ny” or “f***ot” was 13.26. This jumped to 49.5 since his takeover, amounting to an increase of 273%.
Musk claimed on November 23 that “hate speech impressions” were down by a third from “pre-spike levels”, but engagement on tweets containing slurs during that week remained above pre-Musk levels. The average engagement on tweets containing the slurs during the week in question was 33.5, over twice what it was before he took over.
|Average engagement on tweets with slurs|
|Average engagements in the two weeks before Musk’s takeover (14 October–27 October)||13.26|
|Average engagements since Musk’s takeover (28 October–29 November)||49.5|
|The Musk Bump||273%|
|Average engagements in the week Musk made the “pre-spike levels” claim (2022-11-21 to 2022-11-27)||33.5|
3. LGBTQ+ hate went viral on Twitter in the wake of the Colorado Springs shooting
In stark contrast to Musk’s claim about hate speech being “max deboosted”, high-reach tweets from prominent Twitter users promoting hate towards LGBTQ+ people were viewed tens of millions of times in the wake of the Colorado Springs shooting.
CCDH analysis using Brandwatch shows that just 20 prominent hateful tweets can be estimated to have picked up a total of 35 million views.
The viral tweets identified are all instances of anti-LGBTQ+ “grooming” rhetoric, a form of hate speech smearing the LGBTQ+ community with labels like “groomers” and “pedophiles” centered around the baseless lie that LGBTQ+ people are “grooming” children.
Previous research by CCDH and the Human Rights Campaign showed that “grooming” tweets also spiked earlier this year in the wake of the passing of the ‘Don’t Say Gay or Trans’ bill in Florida.
The 20 tweets identified in the wake of the Colorado Springs shooting were the most high-reach examples of the “grooming” narrative out of a wider sample of 656,709 tweets from between 20 November and 23 November. The sample was identified using Brandwatch based on keyword filters for slurs including “grooming”, “groomer”, “pedo”, “paedo”, “paedophile” and “pedophile”. High-reach tweets included:
- A tweet by Tim Pool, viewed 10 million times, which said: “We shouldn’t tolerate pedophiles grooming kids… Club Q had a grooming event…How do prevent the violence and stop the grooming?”.
- A separate tweet by Pool, viewed 5 million times, said: “the grooming of children is not stopping…people are calling for more violence…I do not think legislators will stop the grooming…People will not stop calling for violence…so you tell me what happens next”.
- A tweet by Kurt Schlichter, viewed 2 million times, which said: “I don’t think we have to tolerate pedophiles because some asshole shot up a gay bar. Frankly, a lot of people trying to convince us we need to tolerate pedophiles seem to be happy to use any excuse to silence our opposition.”
- A tweet by far-right online personality James Lindsay, viewed 165,000 times, that included a photo of a transgender person with the comment “Groomer eyes”
- A tweet by Missouri State Representative Ben Baker, viewed 165,000 times, that described drag events involving children like those taking place at Club Q as “grooming” by “some gender appropriating sicko”
Notably, James Lindsay, the author of the “Groomer eyes” tweet, had previously been banned permanently from Twitter for violating its rules on hateful conduct, but was recently reinstated by Musk.
Our analysis also shows that the overall volume of tweets containing terms linked to the grooming narrative – including “grooming”, “groomer”, “groomers” and “pedo” – spiked in the wake of the shooting, with an average of 188,715 tweets in the three days following the attack, over triple the daily average in the week before the shooting.
This analysis includes retweets and quote retweets and covers any mention of the terms, including those expressing outrage or defending the LGBTQ+ community as well as tweets unrelated to the anti-LGBTQ+ “grooming” narrative.
4. Anti-LGBTQ+ extremists are picking up followers at quadruple the pace under Musk
The followings of anti-LGBTQ+ extremists responsible for spreading the “grooming” narrative are growing at quadruple their previous speed, according to CCDH analysis of historical follower figures from the social analytics tool Social Blade.
Earlier this year, CCDH identified ten individuals with outsized influence in proliferating the anti-LGBTQ+ “grooming” narrative on Twitter. The list includes LibsofTiktok, Christina Pushaw, Jack Posobiec and James Lindsay, among others.
In the three months running up to Musk’s takeover, these ten individuals gained a combined total of 222,709 followers a month, taken as an average across the three months. In the month following Musk’s takeover, they together gained 944,205 followers. That amounts to a pace of growth over four times as high as before Musk’s takeover.
While this growth will have causes including the US midterm elections, where anti-LGBTQ+ rhetoric played a large role, it indicates that accounts spreading anti-LGBTQ+ hate are not having their content “max deboosted” as Elon Musk promised.
|Who?||August New Followers (pre-Musk)||September New Followers (pre-Musk)||October New Followers (pre-Musk*)||November New Followers (post-Musk)||Average pre-Musk||Multiple|
|DrewHLive||4,622||-1,798||-2,934||8,018||-37||NA (negative growth before Musk)|
* Musk completed the deal on the 27th of October, so October is counted as “pre-Musk”