What social media can, and cannot shape or predict around elections

Elon Musk’s $ 366 million, and the Twitter/X feed he owns, did not keep the House Republican, and nobody can show that Lula’s volunteer army of millions on WhatsApp won him Brazil’s election. However, both consequential results bear out what I studied for my PhD around Sri Lanka’s socio-political developments, and how platforms reflect a political mood far more often than they make one.

Elon Musk spent roughly $ 366 million on this year’s American midterms, $276 million of it since July, and owns the platform whose feed a Harvard study found rewarding divisive posts and which the statistician Nate Silver argues favours right-leaning accounts, yet Democrats have won the House with what NBC News projects as 258 seats and flipped at least seven Senate seats, among them Texas, Ohio, Iowa and Maine, where his political action committee (a vehicle for funding campaigns) had targeted its spending. Nine days earlier in Brazil, Luiz Inácio Lula da Silva, who had trailed Flávio Bolsonaro in the first round, won the runoff. Lula’s the first candidate since the two-round system began in 1989 to win after finishing second.

The two results look unrelated, though they answer the same question, which is whether the loudest voices online decide elections, and the answer I reached in my 2021 doctoral thesis, after studying, inter alia, four major crises in Sri Lanka, and 1 in New Zealand, was that they rarely decide anything on their own.

Carnival mirrors

I described Facebook and Twitter in that thesis as “carnival mirrors of polity and society, refracting and reflecting offline developments”, and the argument was that engagement follows the offline context rather than manufacturing it. The offline context in America this autumn was not hard to read, since Trump’s approval rating sat in the mid-thirties before the vote, economic worries were mounting in the campaign’s final stretch, and JD Vance has blamed the Iran war for the Republican losses, so the anger circulating on Twitter/X, whatever its algorithmic fuel, had to compete with a public whose anger pointed somewhere else.

Sri Lanka’s 2018 constitutional coup taught the same lesson in miniature, since over 60 accounts of the party backing Mahinda Rajapaksa’s unconstitutional appointment as prime minister tweeted the most, yet the 22 accounts of the party opposing it drew the highest engagement, leading me to conclude in my PhD thesis that “a high production volume is a poor measure of actual engagement”.

Lula’s army and the craft of the surge

The Brazilian case is more interesting still, because it shows people rather than money learning to work the machine. Alexandre Afonso, a political economist at Leiden University, Netherlands, joined a WhatsApp channel called O Exército do Lula (Lula’s Army), launched by the deputy André Janones on the night of the first round and, by its own count, 10 million strong within 3 weeks, and went through its 1,151 text messages and 104 voice notes. What he found was a plain playbook in which members were told to act at the same instant “to create that explosion”, since platforms such as Twitter/X reward sudden spikes over steady popularity, to flood hashtags until an opponent’s slogan fell off the trending list, to ignore the other side’s posts because even a rebuttal feeds the algorithm, and to type a neutral-sounding search into Google so that curiosity did the persuading.

Humans rather than bots did all this by hand, at almost no cost, which Afonso sets beside the finding by Adam Bonica, a professor of political science at Stanford University, and Jake Grumbach, an associate professor of public policy at UC Berkeley, that Democratic campaigns in America spend 45 cents of every dollar they raise on fundraising itself, with barely $ 13 of a $100 text-message donation buying ads. Afonso concedes, to his credit, that nobody can say whether the channel mattered to the result.

I recognise the craft, because Sri Lankans stumbled on it during the 2018 coup, when activists behind #riseuplk and Rajapaksa’s supporters behind #letmevote fought an “open, enduring contestation”, each “vying for attention and influence”, while a video of two young drama festival winners refusing to shake hands with 2 senior politicians allied to the President, though poorly recorded and without any post-production flair, drew over 187,000 views on a single Facebook page in under a day, and by far exceeded the combined interactions of nearly 60 official party Twitter accounts.

Mood, method and money

Outrage has not stopped working, though, because my own data point the other way, drawn from the 2019 presidential campaign, in which I studied 188 Facebook pages backing Gotabaya Rajapaksa and 63 backing Sajith Premadasa, and found that gossip pages produced 64,100 posts and meme pages 32,300, against 23,400 from Rajapaksa’s party account and 7,800 from the governing UNP’s. The administrators of those pages told me they were “interested in profit more than politics”, with some taking money to withhold a rival’s adverts, and Rajapaksa won.

Rajapaksa’s 2019 victory means rage cannot be written off as an electoral force, in Sri Lanka or anywhere else, though what 2026 shows, read through what I would call mood, method and money, is that engagement-driven amplification is neither sufficient nor reliably decisive, and that a few hundred million dollars cannot substitute for a favourable political context. Mood is the offline condition, and in Sri Lanka in 2019 it was fear after the Easter Sunday attacks, sharpened by revelations that the President and Prime Minister had been repeatedly warned and did nothing, which delivered an electorate to a strongman who promised security, whereas in America in 2026 prices and a war delivered the opposite. Method is how cheaply and cleverly people can synchronise, which Brazil and Colombo both show to be within anyone’s reach, the propagandist’s as much as the activist’s. Money, as Musk’s $ 366 million suggests, buys reach without buying belief.

The stakes for Sri Lanka

Sri Lanka blocked Facebook and other platforms during the anti-Muslim riots of 2018 and again after the Easter Sunday attacks in 2019. What I studied at the time, and since, suggests neither block touched the mood those platforms were reflecting, since memes mocking the ban drew heavy engagement, searches for circumvention apps surged, and the hate itself stayed online. A state that treats the mirror as the problem will keep reaching for a social media kill-switch, while the offline conditions that give outrage to its audience go unaddressed (think 2022’s Aragalaya, for example).

The Brazilian strategic playbook also travels in every direction, because hashtags can be used for anything, by anyone. The same synchronisation that lifted #NortheasternersWithLula can amplify hate against minorities in Sinhala across Facebook (and this is in fact happening even today). To wit, whoever reads America’s midterms as proof that algorithms no longer matter, or Brazil’s runoff as proof that catchy, clever hashtags can win elections will legislate, campaign and moderate for the wrong problem. That is a costly mistake for Sri Lanka, whose politics now plays out largely on Facebook, WhatsApp and Twitter/X, platforms whose algorithms and content rules are set by American companies with little to no regard for Sri Lankan voters, and lives.