Misinformation About Misinformation

Yet another Government that does not understand that empowering itself means empowering its successors, who are currently its opponents. They never learn. David Rose writes:

The title of this article is not mine. The term “misinformation about misinformation” was coined by five academics, led by Ceren Budak of the University of Michigan, in a paper published in Nature in June 2024. It describes the gap between what public figures say about falsehood online and what the research shows. I borrowed it as the title of a briefing for the Free Speech Union, published this week, because it also describes the evidential foundations of what Andy Burnham’s Labour Government now proposes: the most far-reaching machinery for controlling public speech that this country has seen in peacetime since the seventeenth century.

Culture Secretary Lisa Nandy’s Green Paper, Watch this Space, suggests that the algorithms governing social media and video-sharing platforms be compelled by law to give prominence to “regulated public service media”, so that outlets such as the BBC are “seen and heard in the fierce battle against mis and disinformation”. This is a zero-sum game: privileging the Beeb will demote everything else. Lucy Powell, Labour’s Deputy Leader, wants the Representation of the People Bill amended so that platforms come under “some of the same requirements” of accuracy and balance as licensed broadcasters during election campaigns – a plan that would require a vast, and as yet non-existent apparatus of surveillance and censorship. Meanwhile, following amendments earlier this year to the Online Safety Act, the media regulator Ofcom has instructed platforms to draw up protocols for restricting content during a “crisis”, defined as an “extraordinary situation in which there is a serious threat to public safety”.

Beneath all of this sits a single document: a 136-page “evidence review” of mis- and disinformation and their impact, commissioned by the Government Office for Science (GO-Science) from a seven-person team at Sheffield University and published in May. It was commissioned to supply the “full evidence base” ministers admitted they previously lacked. Its tone was apocalyptic, concluding that online fake news “is now widely recognised as a systemic national risk”, that “algorithms and business models contribute to [its] amplification and impact”, and that “system-level interventions are needed”. It went on to recommend statutory “quality signals” in the ranking of content, a state-linked “database of false and misleading narratives”, and official “prebunking” across education and public communications. Debates about what misinformation actually is, it added, “should not become a distraction from action”.

Two often repeated claims underpin this programme. The first is that mis- and disinformation saturate the internet and inflict grave harm: riots, vaccine refusal, subverted elections. The second is that platform algorithms are the engine of the crisis, pushing inflammatory content at innocent, vulnerable users because outrage maximises engagement and engagement maximises revenue, and that this radicalises them permanently. Yet tested against the primary sources and the peer-reviewed literature, neither proposition holds.

Somewhat surprisingly, the review’s authors made no attempt to establish the extent of what was true or false on the internet. Instead, by their own account, they relied on summaries generated by large language AI models of 228 academic papers they had not read, and on the monthly bulletins of the European Digital Media Observatory (EDMO), a body that coordinates fact-checkers across the EU. Those bulletins record replies to questionnaires asking fact-checkers what subjects they have dealt with. Hence, the review’s claim that health misinformation was “detected” every month means only that this was something that fact-checkers chose to check. It does not record the number of misleading posts, the size of their audience, by what standard they were judged false, or how many people believed them.

Nor were the authors and their eleven “expert” reviewers equipped to make such judgements. They included no climate scientist, no economist, no clinician, immunologist or epidemiologist, and no scholar of migration, though inter alia the report pronounces on misinformation about climate, “EU financial policy”, transgender sportspeople and “migrant entitlement”. However, they did include longstanding members of what has been termed the censorship industrial complex, such as Clare Melford, chief executive of the Global Disinformation Index; Chris Morris of the self-appointed fact-checking organisation, Full Fact; Paula Gori, the head of EDMO itself; and Bob Ward, the London School of Economics communications officer whose output largely consists of complaints against journalists who question climate and energy policy. All are partisans in the contest the review was supposed to adjudicate, and their own work is cited repeatedly in its text. They were literally marking their own homework.

Melford’s record shows what that means in practice. The Global Disinformation Index, which she co-founded in 2018, does not correct falsehoods but tries to defund publishers by pressurising advertisers. It rates news websites for “disinformation risk”, compiles a “dynamic exclusion list”, and supplies it to ad agencies and their clients, so that those of which it disapproves are starved of revenue. In 2020, in the first wave of the Covid-19 pandemic, the GDI rebuked Senator Tom Cotton for suggesting on Fox News that it might have escaped from a laboratory in Wuhan, China. According to the GDI, through the broadcast of his words, a “debunked conspiracy theory” was being given “authority, validation and amplification”. It went on to insist that the lab leak hypothesis had been “fact checked and proven untrue” and that advertisers must withdraw from sites that echoed it. Its position has not aged well, for many experts, governments and intelligence agencies have come to believe that far from being a conspiracy theory, the likelihood is the virus did escape from the poorly-shielded Wuhan lab.

The GDI did not confine itself to origins. For example, in 2021 it denounced a Spanish-language site that had reported that a third of British deaths from the Delta variant were among the vaccinated, describing this as “clearly untrue”, and persuaded the restaurant chain Chipotle, whose advertisement sat beside the article, to withdraw support. Public Health England’s Technical Briefing 16, published on 18 June 2021, recorded 73 deaths in England among people confirmed to have the Delta variant. Of these, 34 were unvaccinated, ten had received one dose, and 26 – 36 per cent – had been fully vaccinated for more than a fortnight; more than half of the dead had received at least one dose.

Meanwhile the GDI and its allies have been redefining what constitutes dis- and misinformation, applying the terms not only to falsehoods but “adversarial narratives”: content that may be accurate but which, in the GDI’s judgement, sets people against an individual, an institution or “the science”. In 2024 it told UnHerd that the site would stay on its advertiser exclusion list because it published gender-critical writers such as Kathleen Stock, whose views are protected under British law. For UnHerd, the consequences were severe: it estimated that it was receiving between 2 and 6 per cent of the advertising revenue its audience would normally command.

Another leading component of the censorship lobby, the Centre for Countering Digital Hate, which was co-founded by Morgan McSweeney, now campaigns against the “new climate denialism” – a category that includes dissent not from the laws of physics but from Britain’s Net Zero energy policy, and voicing concern over the drawbacks of heat pumps and electric vehicles. It is a matter of record that between 2019 and 2023 the Foreign Office paid the GDI grants worth £2.6 million.

It is not therefore difficult to see why the GO-Science review reached the conclusions it did. Its proposed “database of false and misleading narratives”, refreshed monthly and linked to the state, is the GDI’s bank of disagreeable “narratives” with a government seal of approval. Having commissioned an evidence base, ministers were given a predetermined manifesto.

Maybe its authors should have used a different AI, but the fact remains that their sweeping assertion that “the governance of online platform algorithms is exclusively profit- and engagement-driven” and so fosters hatred and division rests on a single citation: a 2025 paper from staff at the Universitas Muhammadiyah Buton, a small private institution in the province of Southeast Sulawesi, which the UniRank directory says is the 433rd best of the 562 universities in Indonesia, and the 10,370th best in the world. I filed a freedom of information request to Sheffield University to elicit the cost of the review to the taxpayer. The answer: £496,355.

More significant is what the review missed: a body of research in the world’s leading journals, none of it cited, which contradicts its every finding. In the words of the Nature paper by Budak and her colleagues, “public intellectuals and journalists frequently make sweeping claims about the effects of exposure to false content online that are inconsistent with much of the current empirical evidence”. Their own review documented “a pattern of low exposure to false and inflammatory content that is concentrated among a narrow fringe with strong motivations to seek out such information”. They noted that visits to 490 websites deemed “untrustworthy” on the basis of fact-checkers’ lists made up just 5.9 per cent of Americans’ clicks on news sites, and when television is added to the denominator, “fake news” accounts for roughly 0.15 per cent of the average media diet. “This reality,” the Nature authors observed, “is not reflected in public discourse about social media.” The GO-Science review appeared two years later, but left this paper unmentioned.

On algorithms, the evidence is if anything more awkward for ministers. The largest study of the question was conducted by researchers at Twitter itself and published in PNAS in 2022, covering seven countries and millions of accounts. It found the algorithm amplified right-of-centre legislators somewhat more than those on the left, but as for extremism, “contrary to prevailing public belief, we did not find evidence to support this hypothesis”: far-right and far-left parties in France, Spain and Germany were amplified less than moderate ones. A landmark experiment published in Science in 2023 switched tens of thousands of Facebook and Instagram users from an algorithmic feed to a simple chronological one for three months. Their political attitudes did not measurably change, and the algorithmic feed turned out to have served them less content from untrustworthy sources, not more. Studies of YouTube found that consumers of extremist videos were overwhelmingly subscribers who went looking for them, not ingenues led astray by recommendations.

Budak and her colleagues drew the obvious conclusion: “Although algorithms indisputably shape the content people see, we interpret recent empirical evidence as suggesting that, on average, these algorithms tend to push users to more moderate content and to offer extreme content predominantly to those who have sought it out.”

The GO-Science review infers that misinformation may have caused “societal harms” such as Covid vaccine hesitancy. The most substantial study of that question was published in Science in 2024. Combining survey experiments with Facebook exposure data covering 233 million American users, it found that content flagged as false by fact-checkers had almost no aggregate effect on vaccination intentions, for the simple reason that hardly anyone saw it: such material attracted 0.3 per cent of the 2.7 billion views of the vaccine stories in the dataset. What did depress intentions was unflagged “vaccine-sceptical” content that contained no falsehoods at all, which had 46 times the aggregate impact of the material flagged as false. The single most influential item was an accurate report in the Chicago Tribune headlined “A healthy doctor died two weeks after getting a COVID vaccine; CDC is investigating why”. It was seen by more than 50 million people – over six times the views of all flagged vaccine misinformation put together.

Advocates of regulation cite this study as proof that moderation must go further. It proves the reverse. The material with the measurable effect was truthful reporting by an established metropolitan newspaper: precisely the “trustworthy” journalism Nandy’s Green Paper proposes to amplify. A regime aimed at misinformation would have missed it. A regime capable of suppressing it would have crossed a baleful line: from correcting falsehood into suppressing accurate news that officials found inconvenient. That is a choice the censorship lobby usually leaves unsaid, although Melford of the GDI came close in 2021, explaining that her organisation had moved beyond merely checking facts because “something can be factually accurate but still extremely harmful”.

If misinformation really were the threat ministers describe, one would expect its footprint to be visible in what people believe. A study published in PLOS ONE in 2022 went looking for it. The team examined national US surveys of 46 separate conspiracist beliefs, repeated with identical wording since 1966, alongside polling on six theories in six European countries, including Britain, and repeated measures of generalised “conspiracy thinking”. “In no instance,” they reported, “do we observe systematic evidence for an increase in conspiracism.” Of the 46 American beliefs, significant declines outnumbered increases by 15 to six. In Europe, just one theory had gained ground in a single country. Yet another belief that had grown was the assertion that conspiracy theories themselves were out of control: 73 per cent of Americans thought so, a conviction the authors attributed to alarmist statements by politicians and the press. Online conspiracy content, they concluded, “might not persuade as much as reinforce existing views.”

Britain has been here before. In July 1637 the Star Chamber of Charles I decreed that nothing might be published without a licence from approved authorities, with enforcement delegated to a private body, the Stationers’ Company. The categories of the forbidden – “seditious”, “scandalous”, “offensive” – meant whatever the licensers wished. The proposals of 2026 are not an exact echo. But exalted legal status for approved publishers, emergency powers over content, and official programmes to instruct children and adults in which voices to trust are that regime’s closest peacetime relatives in nearly four centuries. Of course they failed, and Charles was beheaded twelve years later.

Seven years after the King’s measures took effect, John Milton inaugurated England’s free speech tradition with Areopagitica, and his answer to the licensers still stands. If we believe in truth, “we do injuriously, by licensing and prohibiting, to misdoubt her strength. Let her and Falsehood grapple; who ever knew Truth put to the worse, in a free and open encounter?”

The Government’s case rests on a review that read nothing, measured nothing and ignored the research that contradicted it. On the evidence, the case for the most far-reaching apparatus to restrict peacetime speech since the ousted Stuart king indeed amounts to misinformation about misinformation.