Truth, Elections and the Questionable History of the Club Sandwich

In the public sphere of reportage, Duke Reporters’ Lab, which has tracked fact-checking trends for years, found that last year was the first time fact-checking department closures outpaced launches. And on the level of the individual, three-fourths of the fact-checkers surveyed by the International Fact-Checking Network reported being financially vulnerable or in crisis. If the field has had a famous debunker in the last two decades, it has been the Washington Post’s Glenn Kessler, who last year accepted a buyout and was not replaced. Should you want to pre-write a eulogy for the capital-T Truth, the material is all here. 

But fact-checking is not a static field. Like our immune systems, it’s capable of learning and enduring, as well as mobilizing to handle infections of mistruth, which if not treated tend to further polarize voters in political contests. Consider recent false claims from his opponents that Texas Senate candidate James Talarico had disappeared from the campaign trail for mental health treatment at an inpatient facility in Ohio. A high-profile fact-check quickly shifted the narrative from “Talarico is in trouble” to “his opponents look like liars.” (Talarico had the flu and was resting at home, which the New York Times proved by reviewing security camera footage.) 

Bad-faith punditry has a lessened effect when met with undeniable facts and data, as do the rumors that chip away at our trust in elections themselves. Consider a recent viral video in which a woman from Cook County, Illinois claimed that her voting machine was telling her that her chosen candidate was the less-popular choice. Nonpartisan fact-checks from places like PolitiFact quickly rushed to explain the less-salacious reality: She hadn’t selected enough candidates. Fact-checks can bring the temperature down and reduce our polarization; her town’s Republican Committeeman even rose to refute the claims on Facebook. So while the institutional layer of the media is wounded and we’re ushering in an age of slop and lies, any obituary for the Truth is likely premature.

“Ever since I was a little kid, I’ve had a thing about wanting to get stuff right,” says Jy Murphy, a freelance fact-checker for the Hearst stable of publications and others. “At this point it bleeds over as a kind of professional deformity.” 

Murphy tells a simple story set years ago, in which he was fact-checking a Popular Mechanics piece about the freight industry and encountered a confident sentence claiming that nobody double-stacks rail cars in the Eastern U.S. because the bridges were all built with too little clearance. Fact-checkers encounter these offhand assertions all the time: a just-so statement with the ring of a thing that everybody knows. 

If you were to Google this type of niche knowledge, you might not find anything, and so the fact-checker operates most often as a reporter of the reporting. But after calling the railroad lines themselves, Murphy learned that in the year since the writer had filed the piece, a considerable number of bridges had been raised to allow freight through. The writer’s claim was “either too pessimistic to begin with or had changed in that year,” says Murphy. There are few forces more powerful in the media than a stubborn person on the phone investigating a doubt.

It is precisely that thoroughness that makes peer-to-peer fact-checking matter more now in the age of A.I. Of course most large language models (LLMs) can’t yet make phone calls to freight lines, but if I ask Claude about Murphy’s rail bridges I may well still be told that nobody double-stacks rail cars in the Eastern U.S. That model can create a convincing, seemingly well-sourced sentence in a few seconds, but what it is not doing is actually checking the fact. 

Kelsey Lannin, a Seattle-based fact-checker, tells me that A.I is “very good at aping a confident intellectual on any given topic. But do any amount of work with it and you’ll quickly realize it’s a facade.” She says that it’s better to look at it like having an intern: “You give it an assignment, it gives you something back, and you [still] have to check it.”

A.I. boosters may tell you that we’re just getting started, and assure us that this is the worst A.I. will ever be. Nevertheless, A.I. is still failing to produce accurate citations in more than 60 percent of tests, according to a 2025 Tow Center study. Prominent critics say it has nuked the effectiveness of Google and is “destroying search, the internet and your brain.” Models create a texture of verification, but they aren’t doing the analog process that makes verification real. If you see a bird out of your window, you can grab an Audubon guide and verify it. Feed a picture of that bird to an LLM and that model is neither “seeing” the chickadee nor “reading” from a primary source specifically. Models produce the most statistically likely next words, but unless prompted, there isn’t a step where Claude or Gemini confirms the truth of its response against the truth of the world.

These individual hallucinations and absent citations might seem like drops in a bucket, but they add up to create a rising tide of mistrust in a society already wary of the truthfulness of mainstream media. “What grieves me,” says Murphy, “is how many people think there aren’t [trusted sources].” 

Still, it is encouraging to see the fact-checking infrastructure responding by adapting to, and building defenses against, this flood.

Sometimes, resistance looks like digging in. Inside the same Reporters’ Lab census that tallied closures, we learn that we’ve still got 437 active fact-checking projects around the world, which is just 27 below the all-time high and more than were operating in 2020. Many American newsrooms (North Dakota News Cooperative, CT Mirror in Connecticut and more) have agreed to work with Gigafact, an organization that publishes independently verified “fact briefs.” Wikipedia adopted a stronger A.I.-generated content policy earlier this year, banning LLM-generated content on the site, bar in certain circumstances. And as something becomes perceived to be scarce, it also becomes valuable: A 2026 Reuters study points toward an “increased demand for verification” from readers in a low-trust environment.

The most encouraging turn of resistance may be in redirecting the machine back at the onslaught itself. Full Fact, the U.K.’s independent fact-checking charity, is now using A.I. systems to “detect and classify claims across millions of sentences,” per Reuters. A newer version attempts to “prebunk” instead of debunk, creating a model that prioritizes the most potentially damaging narratives for review. A Joint Research Centre study published in Nature investigated and concluded that debunking and prebunking interventions were successful and effective in reducing misinformation’s threats.

Like any effort in inoculation, this effort will also need to be collective. At the end of last year, Scott Klein, then-publisher advocate at Newspack, predicted a “return to prioritizing open source collaboration, not as a luxury but as an existential imperative,” highlighting the need for a “show your work” ethos that will help journalism rebuild trust. Full Fact’s tools are being used now by more than 40 organizations across 30 countries, and importantly, the effort is mostly triage. Full Fact offers that the organization will “always keep human experts in the loop, partly to minimise the risks around bias and hallucinations that come with fully-automated systems but mostly because even the most advanced A.I. does not come close to human reasoning.”

Wait, you’re not a member yet?

Join the Reasons to be Cheerful community by supporting our nonprofit publication and giving what you can.

Join
Cancel anytime

Sierra Juarez, a fact-checker at NPR’s Planet Money and The Indicator, has another, simpler tip for resistance in the world of misinformation. “If you see something on your social media feed that your gut is telling you seems fishy,” she says, “try to find corroborating evidence in three different sources across different types of publications. Read broadly and try to identify common threads.” Bringing the fact-checker’s stubborn doubt to our daily engagement is available to anyone willing to take a minute before they share a post.

Years after my Wikipedia crime spree, I’m heartened that there are fact-checkers out there guarding the bridge and encouraged by the durability of this work. Reports of the Truth’s demise have perhaps been greatly exaggerated. It can be slow and unglamorous, but fact-checking’s collective response to changing times will continue to be a necessary governor, helping pace our accelerated information environment and dismantle false narratives that fuel polarization. “There are,” says Murphy, “good soldiers … a cadre of us out there trying to fight the good fight.”

The post Truth, Elections and the Questionable History of the Club Sandwich appeared first on Reasons to be Cheerful.

Source link