Checking Sources: Lateral Reading and Tracing Claims

How fact-checkers judge an unfamiliar source by leaving it, how to trace a claim upstream to its earliest version, and why evidence differs from a report of evidence, after Caulfield and Wineburg's Verified and Booth's Craft of Research.

A claim reaches you on a page you have never seen. The scholar's instinct is to read it closely and weigh its argument. With an unknown source that is the wrong first move. This page sets out three habits that come before close reading: read laterally, trace the claim upstream, and keep evidence apart from reports of evidence. They come from Mike Caulfield and Sam Wineburg's Verified (2023), from the Stanford study behind it, and from The Craft of Research.

Reading vertically and reading laterally

Sam Wineburg and Sarah McGrew asked 45 experienced internet users to judge live websites while thinking aloud. There were 10 historians with doctorates, 10 professional fact-checkers and 25 Stanford undergraduates, and the topics included bullying, the minimum wage and teacher tenure. Historians and students read vertically. They stayed on the site to judge it and were often swayed by features that are easy to manipulate, such as official-looking logos and domain names. Fact-checkers read laterally. After a quick scan they left, opened new tabs, and read what other sources said about the site. The authors' summary is one line: "Fact checkers, in short, learned most about a site by leaving it."

The fact-checkers had two companion habits. They took their bearings, charting a plan before diving into unfamiliar content. And they showed click restraint, reading the whole page of search results before clicking any one. Established as a finding about these 45 people and these tasks; its limits are below.

Close reading comes second

The site's advice for a great book runs the other way: read the original slowly (How to Read a Great Book). The two do not clash, and the study says why. Its authors grant that close reading belongs in any thoughtful curriculum, then add a condition: "the close reading of a digital source when one doesn't yet know if the source can be trusted is neither efficient nor effective." With a great book you already know who wrote it and why it has lasted. With an unknown page, that is the first thing to find out. Understanding exactly, the grammar of the matter, still applies. It simply comes after you know who is speaking.

From four moves to SIFT

Caulfield's open textbook, Web Literacy for Student Fact-Checkers, teaches four moves: check for previous work, go upstream to the source, read laterally, and circle back when you get lost. Verified builds on a sequence called SIFT, and the sections of its first chapter name the steps: stop; investigate the source; find better coverage; trace claims, quotes and media to their original context. The names carry the method. Pause, find out who is behind the source, see what others report, and follow the claim back.

Its second chapter covers cheap signals: first impressions, web addresses, the .org ending, nonprofit status, numbers and links. Each is easy to fake, so none settles anything alone. Peirce warned against settling belief by authority (Peirce on the Fixation of Belief). Lateral reading does not hand your belief to an authority. It asks who stands behind a claim, then sends you upstream to what they rest on.

Go upstream

Caulfield points out that most news pages carry no original information. They are "reporting on reporting": a story about a story about a study. Trace a claim until you reach the earliest version you can find, such as the study, the speech, the photograph in its first setting, or the quotation in the book it came from. At each step, note what changed. The site's Canon grades quotations this way: attested in the primary text, reported by someone else, paraphrase, or doubtful. A quotation you can find only in other people's posts is, at best, reported.

Evidence and reports of evidence

The Craft of Research, by Wayne Booth, Gregory Colomb and Joseph Williams, later joined by Joseph Bizup and William FitzGerald, gives a section of its chapter on reasons and evidence to "Distinguishing Evidence from Reports of It" (section 7.4 in the fifth edition). In this page's terms, evidence is what was observed or measured; a report is someone's account of it. You rarely hold the evidence itself. You hold a report, often of another report. A news story about a trial reports a press release, which reports a paper, which reports measurements. Each layer can drop a caveat, round a number or swap the population. The page on Reading Nutrition Evidence asks the matching question of any study: was the main outcome the one that improved?

The same structure turns up far from libraries. A name cache on a network is a stored report of an old answer, and Stale Information and Dependable Reports shows how such a stored claim can outlive the truth it recorded.

Where this could be wrong

The strongest objection. The founding study was small and staged: 45 people, chosen sites, a researcher watching. Its authors say so. They call it exploratory, note that being observed makes the setting artificial, and grant that a different sample of sites might have given different results. Lateral reading also leans on reputation, and reputation is a crowd's testimony. A respected outlet can be wrong, and the top search result is not a judgment of truth.

The best reply. Lateral reading is triage, not a verdict. It tells you which sources deserve close reading and which claims to trace. The trail ends upstream, at the evidence, where reputation stops counting. The page on Weighing Testimony shows how historians weigh the witnesses themselves.

What would settle it. Studies of trained and untrained readers judging real sources in ordinary settings, which the authors themselves call for. Until then the method is Provisional, however sensible it looks.

Try this

Check five online claims by leaving the page and reading what others say about each source. Record what you found and how long each check took.
Trace three widely shared quotations to their earliest source. Grade each as attested, reported, paraphrase or doubtful.
For one question, list five sources and mark, for each, what it observed itself and what it reports others observed.
These are the practice steps for the Levels on lateral reading, tracing claims and evidence (the Levels path). Time your first check and your fifth.