President Trump has gone all in on his contention that the 2020 election was stolen from him. Republican leaders, for their part, refuse to oppose him or the Hail Mary legal cases he is feverishly pursuing in every state he can. But the real battle to delegitimize this election is being waged on social media, where for the past week Trump supporters, and everyone else, have seen a barrage of claims about supposed election fraud — claims that, even after the president steps down, will leave tens of millions of Americans convinced that Trump was robbed.

You’ve probably seen the most basic of these claims, which focus on individual cases of alleged fraud: the blind woman in Nevada who said her mail-in ballot had been stolen and filled out by someone else; the election worker in Georgia captured on video angrily crumpling up a piece of paper that Trump supporters insisted was a ballot. (It wasn’t.) But the more common, and more effective, claims have been those that purport to be dispassionate statistical analyses documenting election fraud — or at least the possibility of fraud — by just presenting the numbers.

These tweets, Facebook and blog posts, and website pieces — bearing titles like “There Is Undeniable Mathematical Evidence the Election is Being Stolen” and “The Statistical Case Against Biden’s Win” — started appearing almost as soon as the election results did. They purported to document implausibly high turnout in individual states, and highly suspect election outcomes. When you dug into the numbers, the “evidence” of fraud turned out to be neither undeniable nor “statistical.” Yet that’s done little to stop the parade of misinformation.

In some cases, the numbers at issue were simply invented out of whole cloth. One chart that made the rounds on Twitter last week included what were supposedly voter-registration numbers in various swing states next to vote totals from Tuesday’s election. They were meant to show that in these states the number of votes cast had been larger than the number of registered voters. The numbers were wrong (they were from 2018, a midterm election, when registration numbers were much lower), and the claim was ludicrous. But you can still find people retweeting it today. Similarly, conservative operative Harlan Hill — who was, all on his own, a veritable fount of misinformation about election results — tweeted last Wednesday that Wisconsin’s voter turnout had risen from 67 percent in 2016 to 89 percent in 2020. This, also, was untrue. (Wisconsin’s turnout this year is around 72 percent.) But before that claim was flagged by Twitter as potentially false, it was featured by Donald Trump Jr. in a Facebook post, and retweeted by Eric Trump with the tagline “Absolute fraud!”

Flat-out false numbers are, of course, easier to debunk. So the more artful way of presenting dubious numbers has been to simply omit any context, and to suggest the numbers are self-evidently outrageous. When former federal prosecutor and Trump ally Sidney Powell appeared on Maria Bartiromo’s Fox News show Sunday, she said her group had identified “at least 450,000 votes” in “key states” where the ballot had been marked for Joe Biden but for no other candidate. This, Powell argued, was clearly problematic, while Bartiromo, for her part, said it merited a “massive government investigation.” The only problem with this analysis, as many people have pointed out since, is that “undervotes” — ballots where voters don’t vote in all the races listed — are surprisingly common in the United States, so much that in 2016, close to 2 million more votes were cast for president than for Senate candidates. The 450,000 number, then, isn’t especially surprising.

But to know that, you would have to know how common undervoting is, and most Americans — quite reasonably — don’t. (Bartiromo, in this case, did not inform them.) Omitting context thus transforms an unremarkable statistic into one that seems like a sign that something fishy is going on. The same is true of turnout numbers. In a recent piece for the Trumpist site the National Pulse, Steve Cortes — a senior adviser to the Trump campaign — claimed the 84 percent registered-voter turnout in Milwaukee this year “def[ied] reasonable expectations.” What Cortes failed to mention was that registered-voter turnout in Milwaukee in 2016 was 80 percent. So the increase to 84 percent, far from defying expectation, was predictable, given that turnout in Wisconsin as a whole was up about 5 percent. Hill, meanwhile, was at it again for Minnesota. He mockingly said it was “totally believable and consistent” (meaning it wasn’t) that certain Minnesota counties showed preregistered-voter turnout numbers that were above 90 percent. What he didn’t mention was that an apples-to-apples comparison to 2016 showed that same turnout number was almost 91 percent for the entire state (and was higher than that in certain counties).

In other words, this year’s turnout numbers are both believable and historically consistent.

Of course, these claims can be debunked reasonably easily (I spent a good chunk of last week on Twitter doing it), since the relevant data is literally just a few clicks away. But once these claims escape, as it were, into the wild (or onto Facebook, which has far more reach than Twitter), it’s hard to limit their spread, particularly because social media is perfectly designed to amplify them. Hill, for instance, had his tweets casting doubt on the legitimacy of Wisconsin’s turnout numbers picked up by both Trump brothers (with their millions of followers), and then retweeted by Wall Street Journal op-ed columnist Kimberley Strassel, who said the 89 percent registered-turnout number was “not feasible.” The number was not just feasible, it was unsurprising. Her tweet was nonetheless picked up by Sean Hannity and featured on his show on Fox. This chain of transmission doesn’t just spread misinformation; it also, in some sense, launders it, making it seem more respectable and, therefore, more plausible.

By this point, you’re probably a little weary of all these numbers, a little tired of trying to keep the various claims straight. And that, in a way, is the point: No one’s really trying to prove these claims, because they can’t; there is no meaningful statistical evidence of election fraud. What they’re doing is trying to create uncertainty, and doubt, to make it seem like something — even if not the particular thing they’re talking about — is wrong. They’re playing on our instinctive sense that where there’s so much smoke, there must be fire somewhere. But there is no fire. There are just lots of toxic, billowing clouds of nothing, blown here and there by a remarkably efficient smoke machine.