The researchers built their dataset from posts that named or tagged England players, replies to player and team accounts, and posts using Team England-related keywords, filtering out anything irrelevant. They then used an AI model trained on human-labelled examples to sort the remaining posts, checking samples by hand for accuracy (BBC Sport reported). The model flagged different forms of racism, far-right narratives, and hateful slogans, but counted each post on a simple yes-or-no basis rather than grading severity. That matters for how the 5,500 figure should be read: it's a prevalence count, not a scale of harm.

The abuse wasn't spread evenly. Jude Bellingham drew the most racist abuse in total, while Bukayo Saka and Kobbie Mainoo faced the highest proportion of racist posts relative to how often they were mentioned overall. Djed Spence, who converted to Islam in recent years, was targeted with a distinct stream of anti-Muslim abuse (BBC Sport reported). None of that is in dispute. It shows, in granular detail, who was targeted and how.

What it doesn't show is how that abuse reached the people who saw it. Hope Not Hate's senior researcher, Patrik Hermansson, points out that the abuse traces back to "a relatively small number of accounts and originating tweets" that then rack up huge reply counts. But a large reply count doesn't by itself reveal the route those replies took. Posts can spread through direct visibility to a poster's own followers, through reposts and quote-posts, through search, or through algorithmic recommendation, and a tally of replies doesn't distinguish between them. The BBC Sport account of the research doesn't include a breakdown of impressions or referrals by feed source, so the mechanism behind the reply volume is still an open question rather than a settled one.

Racism in football's online discourse isn't new, and academic research backs that up. A study covering eight international tournaments between 2008 and 2022, four men's and four women's, identified 1,048 racist tweets and built them into a balanced dataset for training detection models (Kicking Prejudice). Its best-performing model classified racist tweets with 96.18% accuracy on that curated test set, and its explainability analysis found that slurs targeting Black players and nationality-based insults were the strongest predictors of racist content, a pattern that lines up with real incidents from Euro 2020.

That's a useful confirmation that football-related racism is a recognized, recurring feature of the platform, not a one-off. It isn't a measurement of how often or how sharply abuse spikes during a live tournament. The dataset was deliberately balanced between racist and non-racist tweets for training purposes, which makes it well-suited for building a classifier and poorly suited for estimating real-world prevalence or timing. So the study supports treating tournaments as a known risk period worth watching. It doesn't, on its own, tell platforms or regulators when a spike is coming or how large one will be, a gap that matters once the conversation turns to designing interventions around "predictable" abuse events.

The idea that algorithms reward outrage isn't unique to X, and it isn't new. Whistleblower reporting from inside Meta and TikTok describes engineers being told to loosen restrictions on "borderline" content, including misogyny and racist material, to compete with rivals for attention. One Meta engineer said management framed the shift bluntly, tying it to a falling stock price. A former Meta employee described a "common trade-off between protecting people from harmful content and engagement" running through these decisions (BBC News reported). An internal Meta study cited in that reporting found the platform's algorithm offered creators "a path that maximizes profits at the expense of their audience's wellbeing," and separate internal data showed Instagram Reels carried 75% more bullying and harassment content, and 19% more hate speech, than the main Instagram feed. Meta disputed the characterization, telling the BBC that "any suggestion that we deliberately amplify harmful content for financial gain is wrong," and pointed to a decade of safety investment. TikTok called the whistleblower claims fabricated. Those denials matter: this reporting describes a general industry dynamic at different companies handling different content, and it shouldn't be read as direct evidence about X or about racism specifically. It's context for how these systems can behave, not corroboration of what happened during the World Cup.

Hermansson's framing of X's own role is more pointed. He argues the abusive posts "show up in the recommendation feed" and "are not obscure corners of this website, they are front and centre," describing this as a deliberate design choice rather than an unavoidable side effect. That's a claim about how the system is built, not a measurement of how many extra people actually saw the abuse because of it.

Testing a claim like that properly requires a specific kind of evidence: exposure figures broken out by referral source, a defined threshold for what counts as a genuine view rather than a passing scroll, and some comparison group that isolates the algorithm's contribution from everything else that spreads content online. Neither the BBC Sport report nor the methodology Hope Not Hate has described so far includes any of that for the England players' case. The "millions of feeds" figure is consistent with how these systems can work elsewhere. As it stands, it's asserted, not measured.

Hope Not Hate argues that current responses to online racism lean too heavily on removing abusive posts after they've already been seen, and it wants platforms to act earlier. Its central proposal is to downrank posts that attract heavy volumes of abuse, so they're recommended to fewer people, a specific and testable design change rather than a vague call for "more moderation" (BBC Sport reported). Hermansson has also said social platforms already have the tools to adjust recommendations during periods when abuse is surging, calling it "entirely feasible to stop this, or a vast amount of it."

Before the new Premier League season, the group called on Ofcom to use its existing Online Safety Act powers more assertively and to designate major sporting events as "high risk," which it says would compel platforms to adopt safety-by-design rules, remove abuse proactively, and downrank escalating viral hate in real time. That's Hope Not Hate's proposed reading of what the Act should do, not a power Ofcom has confirmed or a designation it has committed to using this way. The FA's response stayed at the level of principle, saying players "have the right to play without being exposed to hateful, discriminatory and distressing content online" and calling on providers to "take all of the necessary steps to create a safer life online for everyone." Ofcom's statement to BBC Sport was supportive in tone but general, saying it is "pushing tech firms hard to make their services safer" and will "hold them to account if they don't," without addressing how or whether an event-based high-risk designation would function under its current duties.

One voice is missing from all of this. The BBC Sport report did not include a response from X, so the company's own account of its recommendation policies and its in-tournament enforcement during the 2026 World Cup remains untested by anyone outside the organization.

Put the pieces together and a clear picture forms, along with an equally clear gap. Hope Not Hate's dataset is large, methodically built, and shows real, disproportionately distributed racist and anti-Muslim abuse aimed at specific England players. Academic research confirms this fits a recognized, recurring pattern in football's online discourse rather than a one-off event. What's missing is any published evidence, from Hope Not Hate's report or elsewhere, that ties this specific abuse to X's recommendation system rather than to follower networks, reposts, quote-posts, or search. No independently verifiable methodology, one built to isolate the algorithm's effect from those other channels, has been applied to this case.

Readers weighing this claim, or the next one like it, should look for four things before accepting "amplified by the algorithm" as demonstrated rather than plausible: a breakdown of impressions by referral source, separating recommended feeds from following feeds and search; reach figures attached to specific flagged posts; a comparison against some kind of counterfactual or control group; and a response from the platform itself addressing the methodology. Absent those four things, "amplified" describes a mechanism that's consistent with how these systems can behave, not one that's been shown to have behaved that way here.

The FA has asked for stronger platform action in general terms, and Ofcom has promised to hold companies to account without specifying how. Hope Not Hate wants a firmer legal hook before the next major tournament rolls around. Whether any of that produces auditable referral and reach data, the kind that could actually resolve the amplification question rather than just restate it, is the part of this story still waiting to be written.