Lateral reading: the fact-checker habit that still works when fakes are perfect

Almost everything you were taught about judging online information assumes you can learn the truth by looking harder at the thing in front of you. Check the domain. Look for an author. See whether it looks professional. Examine the image for flaws.

Generative AI has made that approach close to useless. A model can produce a polished site, a plausible author biography, a confident tone and a flawless photograph in seconds. Every surface cue can now be manufactured.

There is a method that does not depend on surface cues at all. It was discovered not in a lab but by watching the people who check facts for a living, and it has the rare distinction in this field of having been tested in experiments. It is called lateral reading, and its first instruction is to stop looking at the thing in front of you.

The study: historians against fact-checkers

In 2017, Sam Wineburg and Sarah McGrew of the Stanford History Education Group ran a small observational study that has since shaped how media literacy is taught. They sampled 45 individuals: 10 Ph.D. historians, 10 professional fact checkers, and 25 Stanford University undergraduates. Each was asked to evaluate live websites and search for information on social and political issues while the researchers watched.

The historians and students were intelligent, educated and experienced with sources. They did poorly. In the authors’ words, historians and students often fell victim to easily manipulated features of websites, such as official-looking logos and domain names. They read vertically, staying within a website to evaluate its reliability.

The fact-checkers behaved differently. As a later summary describes it, fact-checkers quickly left the original web pages and opened new tabs for searching, instead of sticking to the landing pages and reading through them. They looked up who was behind the site, what others said about it, and whether the claim appeared anywhere reputable.

The result was lopsided. This technique made fact-checkers obtain more accurate assessments of the credibility of online information in a shorter time, compared with other people. The paper’s subtitle captures it: reading less and learning more.

The finding that PhD historians, professional evaluators of sources, were outperformed is the important part. Expertise and intelligence did not help. Method did.

Why it works, and why it works against AI

Vertical reading asks: does this look trustworthy? Lateral reading asks: what does the rest of the world say about this source?

The first question can be gamed by whoever made the page, because they control everything on it. The second cannot, because the answer lives elsewhere: in other outlets’ coverage, in reference sources, in records the creator does not own.

That is exactly the property you need against synthetic content. A deepfake can be pixel-perfect. It cannot make reputable news organisations report an event that did not happen, cannot put a fabricated study in a journal’s archive, and cannot create a ten-year history for an organisation invented last week. The fake can be flawless and the context still will not match.

It is the same logic as the security advice to hang up and call back on a known number. Do not evaluate the message. Go around it.

The method in four moves

The most widely taught version is SIFT, devised by the digital literacy educator Mike Caulfield, who describes lateral reading as reading across many connected sites instead of digging deep into the site at hand. The acronym stands for Stop, Investigate the source, Find better coverage, and Trace content to the original context.

Stop. Before you read closely, react or share, pause. Notice if the content has produced a strong emotion, since that is what manipulative content is built to do.

Investigate the source. Open a new tab. Search for the name of the site, organisation or person, not on their own pages. One library guide suggests excluding the site itself from results so you see what others say about it. Thirty seconds usually tells you whether this is a newspaper, a lobby group, a satire site or something that did not exist last month.

Find better coverage. Do not try to verify the claim using the source that made it. Search for the claim itself. If something significant really happened, outlets with reputations to lose will have covered it. If the only sources are obscure or all trace back to one post, that is your answer.

Trace to the original. Screenshots, clips and quotes arrive stripped of context. Find the full video, the whole study, the complete quote. Much misleading content is genuine material cut to mean something else, which no deepfake detector would ever flag.

For images and video, the equivalent move is a reverse image search: find where the picture first appeared and what it was originally said to show.

Can it be taught? Yes, and less than you would hope

Here the evidence is encouraging and also sobering, and both halves matter.

Several studies show that instruction works. In one, published by Harvard’s Misinformation Review, 87 college students completed four one-hour online modules on these techniques. Students improved significantly from pretest to posttest, engaging in lateral reading more often post intervention. The researchers note that the modules had to tackle common misconceptions directly, such as the belief that a dot-org domain makes a site trustworthy.

But the size of the gain is modest. A controlled study of a classroom fact-checking curriculum found that students who received it were more likely to read laterally and accurately assess trustworthiness than controls. However, on average, students who received the curriculum still only read laterally and accurately assessed one out of four posttest examples of online content.

One in four, after a course designed to teach exactly this. A review of the field notes that students start with minimal use of such strategies, so one in four is real progress. It is still a reminder that knowing the technique and using it are different things. Lateral reading costs effort, and in the moment most people do not spend it.

What is not known

Whether it works against AI-generated content specifically. The research was done on websites and claims, mostly before convincing synthetic media existed. The logic transfers, and there is no direct trial showing that lateral readers are fooled less often by deepfakes.

Whether the gains last. Most studies test people shortly after instruction. Durable change in everyday browsing has not been shown.

What happens when the wider web is polluted. Lateral reading assumes the surrounding information is mostly sound. If search results fill with AI-generated pages repeating the same false claim, “find better coverage” gets harder. That is a reason to lean on sources with reputations and track records, not just on volume.

How it interacts with AI search summaries. Asking a chatbot “is this true?” feels like lateral reading and may not be, since the summary can carry its own errors. Going to the underlying sources still matters.

Making it a habit you will actually use

The obstacle is effort, so ration it.

Trigger on stakes and emotion. You do not need to verify everything. Verify what you are about to share, act on, pay for or get angry about.

Give it thirty seconds, not thirty minutes. The fact-checkers in the original study were faster than the historians, not slower. One search on the source, one on the claim.

Use the sources you already trust. Decide in advance which outlets and reference sites you consider reliable, so that “find better coverage” is a quick check and not a research project.

Be most careful with what you agree with. Content that confirms what you already believe is the content you are least likely to check and most likely to pass on.

Teach it by doing it aloud. If you have children, colleagues or parents who send you dubious links, show them the two searches. The research suggests modelling the behaviour is part of what makes instruction work.

Common questions

What is lateral reading?

It is a way of evaluating online content by leaving the page and searching for what other sources say about it, instead of studying the page itself. The term comes from a study by Sam Wineburg and Sarah McGrew of how professional fact-checkers work.

Does lateral reading work better than checking a website carefully?

In the original study of 45 people, professional fact-checkers who read laterally reached more accurate judgements in less time than PhD historians and Stanford undergraduates, who stayed on the page and were misled by features such as official-looking logos and domain names.

What does SIFT stand for?

Stop, Investigate the source, Find better coverage, and Trace content to the original context. It is a four-step version of lateral reading devised by Mike Caulfield.

Can lateral reading be learned?

Yes, but gains are modest. Students improved significantly after four one-hour modules in one study, yet in a controlled trial those who took a fact-checking curriculum still read laterally and judged accurately on only one in four test items.

The bottom line

Looking harder at a piece of content tells you less every year, because everything on the surface can now be generated. Looking around it still works: who is behind this, who else reports it, and where did it come from. Professional fact-checkers beat professional historians using nothing more than a second browser tab. The technique is simple, the evidence for it is real, and its main weakness is that people do not bother. Save it for the things that matter, and then actually do it.