Skip to content

← Thinking Clearly

Reading sideways

Chapter 11 · Where a claim comes from

In this chapter 4 sections
  1. What the professionals do
  2. Three questions
  3. Where a picture was first
  4. Nepal’s own checkers

On Sunday, 11 October 2026, the Budhi Gandaki came down through Gorkha black with mud. Houses went with it, and over the next days helicopters lifted more than a thousand stranded people out of the valley. By late morning on the first day a photo was going round Nepali Facebook: a man on top of a vehicle in floodwater, holding on. “After the terrible flood in the Budhi Gandaki,” the caption said, “a man climbed on top of a vehicle and saved his life.” A page posted it into a group of more than 700,000 members, and an account with 17,000 followers posted it too, after lunch. Hundreds of people reacted, and other pages and groups passed it on with the same words.

It was a real photo of a real man in a real flood. The flood was in Chile.

Three days later the checkers at NepalFactCheck.org published how they found out, and it took them two steps. First they searched for news of the Budhi Gandaki flood and looked at what pictures the reliable newspapers and the government’s offices were using. None of them had this one. Then they handed the picture itself to Google and asked where else it appeared. The answer came back at once: a CNN video from 9 October, two days before the Budhi Gandaki flooded, of a man clinging to his truck in a flood after torrential rain in Santiago, the capital of Chile. CNN’s English site had the same clip, and so did the British paper The Independent. Laid side by side, the frame and the viral photo matched, the man in the same place against the same water. The picture was older than the flood it was supposed to show.

What their report leaves out matters as much. It describes no study of the photo for clues, no peering at the mud, no argument about whether the vehicle looked Nepali, no closer reading of the post. They left it and went to see what the rest of the world said about it. That is the whole method, and it has a name. Reading a page from top to bottom, judging it by what it says about itself, is reading vertically. Leaving it to find out about it is reading sideways, or as the people who study it say, laterally.

What the professionals do

The name comes from a study done at Stanford University in the United States, in which the researchers Sam Wineburg and Sarah McGrew sat three kinds of skilled reader in front of a laptop and watched how they decided whom to believe. There were ten historians with doctorates, whose whole trade is judging old documents. There were ten professional fact checkers from well-known news and fact-checking organizations. And there were twenty-five first-year students at Stanford, the hardest university in the United States to get into at the time, all of whom spent at least four hours a day online. Each person talked aloud while they worked, and the screen was recorded.

One of the tasks was a website called minimumwage.com, open at an article arguing that raising the lowest wage a worker can legally be paid would make food dearer and jobs scarcer. The site looked like what it said it was. It had research reports and news, and its About page said it was a project of the Employment Policies Institute, a “nonprofit research organization” that sponsored “nonpartisan research” by economists at universities. What it didn’t say was that the institute was the work of a public relations firm in Washington that lobbied for the restaurant and hotel industries.

Most of the historians stayed on the site for minutes, reading it closely, as they would a letter in an archive, and so did many of the students. Students praised the layout and clicked wherever their interest took them. One spent two minutes playing with the site’s map of wages in different states. The fact checkers barely read it. One of them looked at the article for six seconds, clicked to the About page, followed it to the institute’s own site, glanced at its description and said, “This is profoundly not helpful.” She opened a new tab, searched for the institute’s name, skipped the top four results, and soon was reading that a New York Times reporter who went to visit had found no sign of an office. A little over two minutes in, she had the institute sized up. Only then did she go back to the article, and she knew what she was reading.

hint givenFact checkers51 sfound the public relations firm: 10 of 10Historians3 min 40 sfound the public relations firm: 6 of 10Students5 min 18 sfound the public relations firm: 10 of 250 min2 min4 min6 minaverage time to find who ran the site
Figure 11.1 The minimumwage.com task, as the Stanford study measured it. Each bar is a group’s average time to find whose project the site was. Anyone who hadn’t found it after five minutes was told it was paid for by someone else; below each bar, how many traced it all the way to the public relations firm.

On average the checkers had found the institute in under a minute. The historians took nearly four, and the students over five, most of them only after a hint. Every checker traced it on to the public relations firm. Six of the ten historians did, and ten of the twenty-five students.

Another task set two websites side by side, from two American organizations of children’s doctors with nearly the same name. One was the country’s main professional body for children’s doctors, with about 66,000 members and the field’s leading journal. The other was a small group that had split away from it in 2002, with a few hundred members and one full-time employee. Every checker judged the big one more reliable. On average they left the small group’s site within about half a minute, to find out who it was, and one had its measure in under two minutes by searching for its name. Half the historians couldn’t choose between the two, and many of them were reassured by the small group’s article having references to proper journals at the bottom. Sixty percent of the students chose the small group as the more reliable of the two. Its site, several of them said, was nicely laid out and had no adverts.

These weren’t careless people. The historians were the most careful readers in the study. They lost because they were careful in the wrong place. A page can say anything it likes about itself. Its name, its logo, its “.org”, its references and its tidy design all cost little to make, and anyone can have them. What a page can’t easily control is what everybody else says about it. So the quickest way to learn whether to trust a page is usually to leave it.

The study was small, forty-five people, and its authors said plainly that small samples can make differences look bigger than they are, and that people watched by researchers don’t always do what they normally would. But the gap between the groups was wide, and it ran the same way in task after task.

Three questions

The researchers turned what the fact checkers did into three questions that anyone can ask of anything on a screen. Who is behind this? What is the evidence? What do other sources say?

Who is behind this is the one the checkers asked first, before reading a word of the argument. It isn’t the same as who wrote it. A post on Facebook was written by whoever posted it, but the photo in it came from somewhere else. An article is signed by a doctor, but the site is run by an organization with a cause. The question is who made this and put it in front of you, and what they want. Often the fastest way to an answer is to type the name of the site or the organization into a search engine with a word like “funding” or “who runs”, or to look it up on Wikipedia. Students are often warned off Wikipedia, but more than one of the checkers in the Stanford study went straight to it, as a first stop to find out who someone was. A Wikipedia article on an organization usually says who founded it and why, and it links to its own sources, which can be checked in turn.

What is the evidence asks what the claim is resting on, and whether that thing is really there. A photo that says it shows a flood in Gorkha is evidence only if it was taken in Gorkha. A quotation is evidence only if the person said it, where and when the post says they did. Often the evidence turns out to be something from elsewhere, real but borrowed, and the honest question is not “is this real?” but “is this what it says it is?”

What do other sources say is the step people most often leave out, and it’s the one that saved the checkers. If something big has really happened, more than one reliable source will be reporting it, separately. If a site is what it claims, others will describe it the same way. If they describe it differently, or nobody has heard of it at all, that tells you something too.

The checkers had one more habit, about how they searched. When they typed a name into a search engine, they didn’t click the first result. One of them spent nearly twenty seconds reading the list of results and the lines of text under each before choosing anything, and skipped the first three, all from the group behind the lawsuit she was asked about. “I’d rather click on some press reports,” she said. The researchers called this click restraint. Sites work hard to get to the top of a search, so the top result isn’t necessarily the one most worth reading. In that task, finding who had paid for a lawsuit, the Stanford students were actually faster than the checkers. They clicked a likely result and took the first name they found, some from a site run by one side of the case, and many didn’t confirm it anywhere else. The checkers took longer, about half a minute on average, and their answers more often came from a source worth trusting, checked against a second.

None of this makes reading sideways a machine for doubting. It works just as well in the other direction. In the classroom lessons that came out of the study, students were shown againstmalaria.com, a site with a plain “.com” that students often take as a reason for suspicion. A little searching shows that the Against Malaria Foundation, a charity in Britain that provides long-lasting insecticide-treated nets, is well respected, has been written about by serious magazines, and is rated highly by an organization that judges how much good charities do with their money. Leaving the page can raise your trust as easily as lower it. The point is to stop guessing from the page’s looks.

Where a picture was first

Photos need the same treatment, with one more tool. A reverse image search takes a picture instead of words, and finds where else the same picture, or one very like it, appears online. Google Lens does it, and so do other services such as TinEye. On a phone it is usually a matter of pressing and holding the picture and choosing to search with it.

What you’re looking for is the picture’s first appearance. A photo of a flood that appeared online a year ago can’t be a photo of today’s flood. Some services let you sort the results with the oldest first. They come with a fair warning, though. TinEye’s own help pages say that the date it shows is the date its crawlers first found a picture, not the date it first appeared online, and that there may be a long gap between the two. So an old date is strong evidence that a picture is older than the claim. A recent date doesn’t prove the picture is new. And finding no match at all proves nothing either way.

That is what happened with the Chile photo. It wasn’t a fake. Nobody had painted the man onto the water. It was a true picture with a false label, which costs nothing to make. The flood in Gorkha was real too, and so was the fear. The post simply borrowed a more dramatic picture from another country two days earlier. We don’t know whether whoever first posted it knew where it came from, or whether those who shared it on did.

Nepal’s own checkers

When you can’t settle something yourself, someone in Nepal may already have. NepalFactCheck.org, which found the Chile photo, is run by the Center for Media Research – Nepal, a non-profit, together with MySansar, which calls itself the first blog in Nepali. It writes in Nepali, says it takes no side in politics, and says it follows the code of principles of the International Fact-Checking Network. By November 2026 it had published 684 checks. Anyone can send it a link, photo or video they suspect, through a form on its website, by Viber or WhatsApp, or by email, though it says plainly that it can’t check every request. A few Nepali news sites have fact-check desks of their own, such as the one at Techpana. Searching a claim’s key words together with “fact check” or “तथ्यजाँच” often turns up an answer in seconds, which is itself a kind of reading sideways.

Can a person learn to do this, or is it a knack that fact checkers simply have? The Stanford researchers tried it. In the spring of 2019, in a city school district in the American Midwest, about 500 students in a required government class took part. Half the students, in classes picked at random, had six lessons of 50 minutes on reading sideways, and the others carried on as usual. The students who had the lessons got clearly better at telling a trustworthy website from a dubious one, though even afterward they missed about half the points on the test. That trial was in the United States. We found no trial of the method in Nepal.

Reading sideways isn’t a full investigation, and it isn’t meant to be. It is a few minutes’ work, the kind the Nepali Times did in December 2000 when it found a film star’s interview and played the whole of it. Its job is to tell you, quickly, whether something deserves your attention at all, and how far to lean on it if it does. Most of the time the answer comes fast. The site is who it says it is, or it isn’t. The photo is from where it says it’s from, or it was somewhere else first.

Sometimes, though, what other sources say turns out to be a number. A hospital says a new treatment cuts the risk by a quarter. A poll says most people agree. A chart shows a line climbing steeply. Every source may report the same number correctly, and the number can still mislead, depending on what it’s counting and out of how many.

Found a mistake? Please tell us.