An unfamiliar website looks convincing.
The design is polished. The logo is official-looking. The About page describes a serious mission. The article cites studies and uses a confident editorial voice.
A reader can spend ten minutes inspecting the site and become more impressed by precisely the features the publisher controls.
A professional fact-checker often does something different.
They leave.
That movement is the heart of lateral reading.
What lateral reading is
Lateral reading is a source-evaluation strategy in which a reader leaves an unfamiliar page and searches across the open web to investigate the source, organisation, author, claims and surrounding evidence.
Instead of evaluating credibility primarily from information supplied by the source itself, the reader asks what independent sources say about it.
This can involve opening multiple tabs, checking institutional or professional records, tracing citations, looking for reputable reporting, identifying ownership or funding and comparing how knowledgeable sources describe the organisation.
It is called lateral reading because the reader moves sideways across sources rather than vertically down one page.
Where the protocol came from
Sam Wineburg and Sarah McGrew studied how professional fact-checkers, historians and university students evaluated online information.
In their 2019 expert-novice study, the fact-checkers characteristically left unfamiliar sites quickly and investigated them through other sources.
Historians and students were more likely to remain on the original site, reading its pages and evaluating surface features.
The fact-checkers reached more warranted conclusions more efficiently in the tasks studied.
That observed expert practice became a foundation for civic online reasoning instruction.
The core protocol
A practical lateral-reading sequence can be expressed as six moves.
1. Pause before deep reading
Do not invest heavily in the unknown source before learning what it is.
The goal is not to ignore content. It is to postpone content-level trust until basic source orientation exists.
2. Leave the page
Open a new tab or search window.
This step matters because the original source controls its own self-presentation.
3. Investigate the source
Search for the organisation, author or publisher.
Look for independent descriptions, professional affiliations, ownership, expertise, funding, editorial standards and relevant history.
4. Prefer evidence about the source over repetition of the source
Ten websites copying the same biography are not ten independent confirmations.
Search for sources with their own evidence and accountability.
5. Trace important claims upstream
When possible, locate the original study, document, dataset, speech, image or reporting from which the claim came.
This overlaps with information provenance and source criticism.
6. Return with a changed question
Only after external orientation should the reader decide how much weight to give the original page.
The question becomes not "Does this page look credible?" but "What do I now know about who is behind it, what evidence it uses and how it fits with better-established sources?"
Lateral reading reverses a familiar literacy habit
Traditional close reading teaches people to attend carefully to the text in front of them.
That is indispensable in many contexts.
Online source evaluation creates a different problem.
A sophisticated misleading site can use all the surface signals associated with seriousness: references, professional design, technical language, testimonials and institutional imagery.
Reading more closely inside that controlled environment can deepen exposure without improving source knowledge.
Lateral reading does not reject close reading.
It changes the order.
First establish who is speaking. Then decide whether deeper reading is warranted.
The protocol depends on source independence
The method works only if the outside sources are better than the unknown source.
This sounds obvious and is easy to violate.
A person can leave a questionable page and land on a network of sites that repeat the same unsupported claim.
Search results can be influenced by popularity, optimisation, personalisation, language, geography and coordinated amplification.
A lateral reader therefore needs a second-order skill: the ability to evaluate the sources used to evaluate the source.
Lateral reading is recursive.
Search engines are part of the measurement environment
The protocol developed in an open-web environment where search engines functioned as gateways to external information.
That means search infrastructure is not neutral background.
The query a person writes affects what appears.
Ranking affects which sources are inspected.
Knowledge panels, snippets and AI summaries can create an impression of consensus before the reader reaches underlying sources.
A study of lateral reading conducted in one search environment may not reproduce identically in another.
Researchers should document the browsing tools participants used.
AI browsing changes the protocol again
An AI assistant can now perform something that resembles lateral reading by summarising external sources.
That can be useful.
It can also hide the very provenance work the protocol was designed to make visible.
If the model says, "independent sources describe this organisation as…", the user needs to know which sources, how independent they are and whether the model represented them accurately.
For AI-mediated lateral reading, the protocol should therefore add a rule:
Do not outsource source inspection completely to a synthesis layer. Open the evidence behind the synthesis.
Teaching evidence is encouraging but incomplete
Educational studies have tested civic online reasoning lessons based on lateral reading.
McGrew and Breakstone's 2023 study embedded instruction across ninth-grade biology and world geography.
Students showed significant improvement in online-source evaluation over the school year.
The authors also emphasised that performance still left substantial room for improvement and that strategy use depends on procedural knowledge, substantive source knowledge and reading comprehension.
This is an important evidence boundary.
Lateral reading can be taught.
That does not mean one lesson creates durable information literacy.
A protocol is different from a checklist
Many traditional website-evaluation tools ask users to score features such as currency, relevance, authority, accuracy and purpose while remaining on the page.
Lateral reading challenges that orientation because a misleading publisher can manipulate many internal cues.
The protocol is therefore less a list of traits than a change of epistemic position.
The reader stops asking the source to certify itself.
That is why the method can feel surprisingly simple while producing a profound change in behavior.
How researchers can measure lateral reading
There is no single universal lateral-reading score.
Depending on the research question, studies can code:
- whether the participant leaves the original site;
- time until first external search;
- number and type of external tabs opened;
- quality and independence of external sources;
- whether ownership or authorship is identified;
- whether original evidence is traced;
- whether the final credibility judgment is warranted;
- written reasoning supporting the judgment;
- efficiency, such as time to a defensible conclusion.
Educational assessments often use rubrics that distinguish beginning, emerging and mastery-level reasoning.
These are performance indicators, not interchangeable psychometric scales.
Think-aloud methods can reveal process
Because lateral reading is behavioral, screen recording and think-aloud protocols are especially informative.
A final answer alone cannot show whether someone reached the correct conclusion through strong source evaluation, prior knowledge or chance.
Process traces can reveal query choice, tab switching, source selection and reasoning.
They also introduce reactivity.
Asking participants to verbalise their thinking may change pacing or attention.
Method choice should follow the research question.
Reliability is not one number for the whole protocol
A constructed-response assessment of lateral reading can have reliable scoring if rubrics and rater training are strong.
McGrew and Breakstone reported very high inter-rater agreement for scored constructed responses in their 2023 study.
That does not mean "lateral reading" itself has one reliability coefficient.
Reliability belongs to a particular operationalisation, sample and scoring procedure.
Behavioral logs, written judgments and classroom rubrics each have different measurement properties.
Validity and limitations depend on what the task claims to represent
A lateral-reading assessment has stronger construct validity when its scoring follows the epistemic work the protocol is supposed to capture: leaving the original page when warranted, identifying who is behind a source, comparing independent evidence, tracing claims toward stronger sources and reaching a conclusion proportionate to what the evidence supports.
A task becomes less valid when it rewards superficial proxies. Opening many tabs is not automatically better reasoning. Finding a negative search result is not equivalent to establishing unreliability. Speed can reflect expertise, but it can also reflect browser fluency, familiarity with a topic or a willingness to stop too early.
Criterion validity is also difficult because there is no universal gold-standard label for every live web source. Studies therefore need transparent expert coding, defensible source judgments, or outcome criteria tied to the specific task. Convergent evidence can come from process traces, warranted final judgments and comparisons with expert performance rather than from one behavioral count alone.
The method is most defensible when researchers state exactly what inference they are making from observed browsing behavior and what alternative explanations remain possible.
The method requires background knowledge without being reducible to it
A reader cannot recognise every credible organisation in advance.
That is precisely why lateral reading is useful.
But external evaluation still requires some understanding of what kinds of sources count as stronger evidence in a domain.
A medical guideline, a university press release, a preprint, a newspaper investigation and an anonymous forum post do not have the same evidentiary role.
The reader must know enough to distinguish them.
Lateral reading therefore complements, rather than replaces, domain knowledge.
Search skill can become a hidden confound
Someone may understand the idea of lateral reading yet struggle to formulate effective queries.
Another person may search quickly because they are highly experienced with web navigation.
If an assessment treats speed or number of external sources as direct evidence of discernment, it can accidentally measure search fluency.
Researchers should distinguish knowing that one should investigate externally from being technically efficient at doing so.
Both matter, but they are not the same capacity.
The protocol has accessibility and language boundaries
Open-web evaluation assumes usable search access, sufficient reading ability and enough high-quality information in the reader's language.
That assumption breaks in low-resource language environments, paywalled domains and contexts where trustworthy local sources have weak search visibility.
People using assistive technologies may also encounter friction in multi-tab browsing.
A method designed for epistemic empowerment should not quietly make browser fluency or English-language access a prerequisite for competence.
Adaptation matters.
Lateral reading can fail in genuinely novel situations
What if no reliable external source has yet discussed the page?
What if the event is unfolding in real time?
What if all available reporting traces back to one uncertain witness?
Then lateral reading may reveal not credibility but the absence of corroboration.
That is still useful information.
The disciplined conclusion may be "not enough is known yet."
A protocol for discernment must make room for suspension of judgment.
When the protocol is useful
Lateral reading is especially useful for:
- unfamiliar websites and organisations;
- viral claims with uncertain sources;
- advocacy material;
- health or science claims outside one's expertise;
- apparent news sites that may imitate journalism;
- images, quotations or statistics whose origins need tracing;
- classroom exercises in digital source evaluation.
It is less necessary when the source identity and evidentiary chain are already well established, though high-stakes claims may still justify verification.
Protocol fidelity can be observed directly
Because lateral reading is a sequence of actions, researchers can ask whether participants actually used the taught strategy rather than inferring it from the final answer.
Screen recordings, browser logs or instrumented tasks can show:
- how long the participant stayed on the original page;
- the first external query;
- which results they opened;
- whether they returned to the source;
- whether they traced evidence upstream;
- how the final judgment changed after external investigation.
This creates a distinction between outcome accuracy and protocol fidelity.
A participant can reach a correct judgment without using lateral reading. Another can use the protocol appropriately but face an information environment too ambiguous to resolve the question.
Both observations matter.
Query formulation is part of the epistemic act
Search is not a neutral window onto the web.
A query such as "Is Organisation X a scam?" can produce a different evidence field from "Organisation X ownership funding" or "Organisation X independent review."
Leading queries may amplify confirmation bias. Vague queries may return popularity rather than evidence. Search-result snippets can be mistaken for source conclusions.
Teaching lateral reading therefore benefits from teaching query discipline: search the entity, investigate ownership and expertise, look for independent descriptions, and avoid embedding the desired verdict in the query unless testing that framing deliberately.
External consensus can itself be manufactured
Lateral reading relies on finding what other sources say.
That creates vulnerability when many apparently separate sites derive from one press release, coordinated network or syndicated story.
The number of pages is not the number of independent evidence streams.
A sophisticated protocol should therefore look for source lineage as well as source count.
This is where lateral reading and information provenance meet.
Organisational use needs shared thresholds
A newsroom, school or research team can adopt lateral reading without requiring every person to improvise standards from scratch.
Teams can define which claims require primary-source tracing, when two independent confirmations are expected, how conflicts are escalated and how unresolved uncertainty is labelled.
Those procedures turn an individual browsing habit into institutional epistemic infrastructure.
They also make quality auditable.
The goal is not bureaucratic checking for its own sake. It is to make source evaluation reproducible enough that another person can understand why a judgment was reached.
Assessment should reward warranted uncertainty
A common educational mistake is to design every source-evaluation task with a hidden binary answer: credible or not credible.
Real information environments are often less tidy.
A strong lateral-reading assessment should sometimes allow the best response to be "insufficient evidence," "credible on this point but not that one," or "the source identity is clear but the claim remains unresolved."
If the rubric rewards certainty regardless of evidence, it trains performance rather than discernment.
The protocol can be taught as a sequence of increasingly difficult cases
A novice does not need to begin with the hardest possible source.
Instruction can progress from a clearly identifiable organisation, to a site with ambiguous ownership, to a claim whose source is credible but whose evidence is weak, and finally to live cases where independent sources disagree.
This sequencing matters because learners can otherwise confuse the method with a simple debunking game in which every unfamiliar site is secretly bad.
A mature curriculum should include trustworthy unfamiliar sources as well as misleading ones. The learner needs practice discovering that some sources deserve increased trust after lateral investigation.
That protects the method from becoming generalized suspicion.
Good lateral reading ends with an evidence statement
The final product should not be merely "credible" or "not credible."
A stronger conclusion states what was learned and what remains unresolved: who owns the source, what relevant expertise exists, whether the claim traces to primary evidence, which independent sources corroborate it, and how confident the evaluator is.
This makes the reasoning auditable and connects lateral reading to epistemic calibration.
The protocol is complete not when enough tabs are open, but when the reader can explain why the evidence now deserves a particular degree of trust.
What the evidence does not show
The expert-novice research does not show that professional fact-checkers are infallible.
Educational interventions do not show that lateral-reading gains automatically transfer to every domain or persist indefinitely.
Leaving a website does not guarantee better judgment if the external sources are weak.
Search rankings are not equivalent to epistemic authority.
The method does not replace domain expertise, close reading or primary-source analysis.
And a person who fails to read laterally in a test may lack procedural familiarity rather than general discernment.
What this changes for Inner Technology
Lateral reading is an unusually concrete Inner Technology practice because the inner and outer moves happen together.
Internally, the reader interrupts the seduction of first impressions.
Externally, they change the information environment by leaving the page.
The skill is therefore not "being skeptical."
It is knowing when to stop consuming one source's self-description and reorganise the evidence field around it.
Institutions can cultivate this through browser-visible modelling, repeated practice and assessment of reasoning rather than slogans about media literacy.
The tab that changes the judgment
The unfamiliar site is still open.
Nothing on it has changed.
But beside it are now three other tabs.
One identifies the organisation's funder.
One traces the quoted study to its original paper.
One shows that a respected independent source interprets the evidence very differently.
The page did not become less polished.
The reader became less dependent on the page to explain itself.
That is the lateral-reading move.
Research references
- Wineburg S, McGrew S. Lateral reading and the nature of expertise: reading less and learning more when evaluating digital information. Teachers College Record. 2019;121(11):1-40. https://doi.org/10.1177/016146811912101102
- McGrew S, Breakstone J. Civic online reasoning across the curriculum: developing and testing the efficacy of digital literacy lessons. AERA Open. 2023;9. https://doi.org/10.1177/23328584231176451
- McGrew S, Byrne VL. Who is behind this? Preparing high school students to evaluate online content. Journal of Research on Technology in Education. 2021;53(4):457-475. https://doi.org/10.1080/15391523.2020.1795956
- Wineburg S, Breakstone J, McGrew S, Smith M, Ortega T. Evaluating information: the cornerstone of civic online reasoning. Stanford Digital Repository. https://purl.stanford.edu/fv751yt5934
- Kamela C. Cut the CRAAP: replacing vertical evaluation with lateral reading. New Directions for Teaching and Learning. 2024;2024(180):49-58. https://doi.org/10.1002/tl.20608
- Stanford History Education Group / Digital Inquiry Group. Civic Online Reasoning resources and assessment materials. https://cor.stanford.edu/