The page looks official.
It has a clean logo, a serious typeface, references at the bottom and an About page written in institutional language.
A reader can spend ten minutes inside it and learn almost nothing about whether it deserves trust.
The decisive move may be to leave.
Source criticism is the disciplined evaluation of a source’s origin, production, context, purpose, dependence and fitness for a particular claim.
In digital environments, that often means investigating the source from outside the source itself.
Definition
For this Knowledge Library:
source criticism is the practice of evaluating who or what produced information, under what conditions, from which underlying materials, with what degree of independence and accountability, and whether that source is appropriate evidence for the question at hand.
The practice concerns provenance before it concerns agreement.
A source does not become credible because it supports the reader’s view, nor unreliable because it challenges it.
The Nordic term carries historical weight
In Sweden, källkritik is not a recent response to social media. It has roots in historical method and a long educational life.
The term is now used far beyond archival history, including school education, journalism, digital competence and public resilience.
That makes source criticism unusually important for the Institute’s Northern European authority layer.
But familiarity can also make the practice sound simpler than it is.
A remembered checklist is not the same as competent evaluation in a live information environment.
Classical source questions still matter
Traditional source criticism asks questions about authenticity, time, dependence and tendency or interest.
Is the source genuine?
How close is it to the event or process being described?
Is it independent, or derived from another source?
What interests, incentives or perspectives may shape it?
These questions remain useful.
Yet digital information complicates each one. Copies detach from origins. Screenshots lose metadata. Many outlets repeat the same upstream report. AI-generated text can appear original while synthesizing sources that are invisible to the reader.
The classical logic survives, but the methods have to evolve.
Lateral reading changes the direction of attention
Wineburg and McGrew’s 2019 study compared professional fact-checkers, historians and university students evaluating unfamiliar websites.
Fact-checkers tended to leave the site quickly and open new tabs to investigate who was behind it and what independent sources said. Historians and students more often stayed on the site, reading vertically and attending to features the publisher controlled.
The fact-checkers’ strategy became known as lateral reading.
The finding should not be inflated into a universal law of source evaluation. The study involved 45 participants in a specific expert-novice design.
Its practical importance is that source criticism on the open web often improves when readers stop asking a site to certify itself.
Looking professional is cheap
Visual quality used to carry some information about institutional resources.
That cue has weakened.
Templates, stock photography and generative tools make polished presentation available to almost anyone. AI can produce references, charts and confident prose at low cost.
This does not make design meaningless.
It makes design a weaker proxy for provenance.
Source criticism therefore has to move from surface cues toward traceable relationships: original publication, author identity, institutional accountability, evidence chain, correction history and independent corroboration.
Corroboration is not counting links
Ten search results can still be one source.
News outlets may republish a wire service. Blogs may cite the same press release. AI-generated pages may copy one another or echo a shared underlying document.
Corroboration requires independent evidential pathways where possible.
The question is not how many pages repeat the claim.
It is whether distinct sources with relevant competence and access converge for reasons that are not merely inherited from the same origin.
This is where source criticism meets epistemic autonomy: independence among sources changes the informational value of agreement.
Digital source evaluation can be taught
Research following the lateral-reading work suggests that explicit instruction can improve students’ online evaluation practices.
McGrew and Breakstone’s 2023 cross-curricular work describes civic online reasoning around three practical questions: who is behind the information, what is the evidence, and what do other sources say?
The educational evidence is encouraging but should remain bounded.
Interventions vary in duration, population, assessment design and transfer. Improved performance on source-evaluation tasks does not mean students become globally resistant to misinformation.
Recent 2026 work by McGrew and Reynolds continues to examine how students reason about credibility and corroboration during lateral-reading lessons, underscoring that the quality of reasoning matters beyond whether a new tab was opened.
Source criticism is not claim verification
A trustworthy source can make an error.
An untrustworthy source can publish something true.
This is why source criticism and verification must remain distinct.
Source criticism asks whether the source deserves weight.
Claim verification asks whether the specific proposition is supported.
A good process often uses both.
When stakes are high, a strong institutional source should still point to inspectable evidence. When a weak source makes a consequential claim, finding the original evidence can sometimes rescue the claim even if the intermediary is poor.
Primary and secondary sources are roles, not quality labels
“Primary source” is often used as praise.
That is a mistake.
A primary source is closer to the object or event under study. It can still be biased, incomplete, forged, mistaken or methodologically weak.
A high-quality systematic review may provide a better answer to a broad evidence question than one primary study.
Source criticism therefore asks what role the source plays in the inquiry.
Original legislation is primary for what a law says. A peer-reviewed meta-analysis may be better for what interventions tend to do. A witness account can be primary for lived observation but insufficient for population prevalence.
Fitness is relational.
Sweden’s current education evidence makes the issue concrete
Skolverket’s 2026 analysis of ICILS 2023 describes digital competence as including source criticism and safe, responsible action in digital environments. The report draws on responses from about 1,560 Grade 8 teachers, alongside school leaders and IT coordinators at roughly 150 participating schools, and notes substantial variation in how digital competence is taught.
That evidence does not test one source-criticism intervention.
It establishes a current Swedish educational context in which source-critical competence remains a live and unevenly developed practice area.
This is a substantive Nordic connection, not keyword localization.
AI creates a provenance problem
When a user asks a generative system for “sources,” several things may happen.
The model may retrieve real documents. It may cite a secondary summary. It may provide an inaccurate citation. It may synthesize across material without exposing the path.
The user therefore needs to distinguish answer provenance from source provenance.
Where did the claim in the generated answer come from?
Where did the underlying evidence come from?
Are those the same thing?
A clickable citation improves inspectability, but the reader still has to check whether the cited document supports the sentence attached to it.
Source criticism can become performative skepticism
There is a failure mode in which people learn the vocabulary of source criticism but use it selectively.
They demand perfect provenance from disfavoured claims while accepting screenshots and anonymous posts that support their own side.
They ask about “bias” only when they dislike the conclusion.
That is not source criticism. It is motivated filtering with scholarly vocabulary.
The discipline has to be symmetrical enough to turn back on sources we want to believe.
Measurement and assessment
Source-evaluation research can use live-web tasks, think-aloud protocols, scenario judgments, credibility ratings and performance-based assessments.
Each captures different operations.
A checklist quiz may test declarative knowledge about source criteria. A live-web task can show whether people actually investigate provenance. Think-aloud data can reveal reasoning but may alter the process being measured.
Assessment should therefore specify whether it measures knowledge of principles, enacted evaluation, final accuracy or transfer across contexts.
These are not interchangeable outcomes.
Source criticism has two clocks
Some source judgments can be made quickly.
A page lacks an author. A domain is impersonating another organization. A quotation has no traceable origin.
Other judgments require slow reconstruction.
An archival document may be genuine but produced for strategic reasons. A peer-reviewed paper may be methodologically weak. A dataset may be authoritative for one population but unsuitable for another.
This gives source criticism two clocks: rapid triage and deeper appraisal.
Confusing them creates errors in both directions. A fast heuristic can save attention when the stakes are low. The same heuristic can become dangerous when it substitutes for detailed evaluation in a high-stakes claim.
Lateral reading is a strategy, not a religion
The lateral-reading research associated with professional fact-checkers is valuable because it identifies a counterintuitive expert behaviour: leaving a page in order to understand the page.
Instead of reading an unfamiliar site vertically from top to bottom, fact-checkers often open new tabs, search the organization, trace claims and inspect independent coverage.
Wineburg and McGrew’s 2019 study compared 10 professional fact-checkers, 10 historians and 25 Stanford undergraduates. The fact-checkers’ lateral strategies helped them evaluate unfamiliar online sources more effectively in the studied tasks.
The result should not be inflated.
Lateral reading is not proof that the top search result is correct. Search engines have their own ranking biases. Coordinated misinformation can contaminate external sources too. In specialized domains, expert appraisal may require reading deeply within the source rather than leaving it quickly.
The useful principle is broader: source evaluation should not be trapped inside the source’s self-presentation.
Primary, secondary and upstream sources
The instruction to “find the primary source” is often good.
It can also be oversimplified.
A primary study may be the correct source for a method or measured result, while a high-quality systematic review may be the better source for the state of evidence. A government dataset may be primary for reported statistics but not for a causal explanation. A press release may be primary evidence of what an institution announced while remaining weak evidence that the announcement is true.
Source criticism therefore asks primary for what claim?
This claim-relative approach is more precise than a universal hierarchy in which primary always outranks secondary.
Independence matters as much as number
Three articles can look like corroboration while all tracing back to one press release.
Ten websites can repeat a statistic while none links to the original dataset.
A scientific review can include multiple studies from overlapping samples.
This makes source dependence an essential but often neglected dimension of source criticism.
Independent convergence generally carries different evidential weight from repetition through a citation chain.
Digital systems make this hard because copied language can detach from provenance. Generative AI compounds the problem by synthesizing repeated claims without always exposing whether they ultimately share one source.
Institutional authority is domain-specific
A source can be highly authoritative for one question and nearly irrelevant for another.
A central bank may be authoritative about the interest rate it set. That does not make every macroeconomic forecast it publishes certain. A company is authoritative about its own product policy but conflicted when evaluating the product’s social impact. A researcher may be a leading expert in one subfield and an ordinary commentator in another.
Source criticism therefore evaluates authority in relation to the claim, not prestige in the abstract.
This protects against both credentialism and anti-expertise.
Provenance in the age of generated content
Traditional source criticism often assumes that a document has an identifiable author or institution.
Generated content destabilizes that assumption.
A text may be produced by a model, prompted by one person, edited by another and published under an organizational account. An image may combine synthetic and authentic elements. A dataset may be summarized through an AI tool without the intermediate transformations being recorded.
Provenance becomes layered.
Who supplied the evidence? Which system transformed it? Which person accepted the transformation? Which version was published? What remains recoverable?
These are not merely technical metadata questions. They determine whether later readers can reconstruct the evidential chain.
Teaching source criticism without training suspicion
Poor source-criticism instruction can accidentally teach that every source hides manipulation.
The more useful goal is calibration.
Learners need practice recognizing when trust is warranted, when verification is needed and when a source’s limitations are ordinary rather than disqualifying.
A newspaper can be reliable without being infallible. A peer-reviewed paper can be useful without being definitive. A personal testimony can be essential evidence of lived experience while remaining insufficient for a population-level causal claim.
This claim-relative trust is more demanding than both gullibility and cynicism.
It is also more compatible with how real knowledge systems work.
Source criticism is also correction literacy
A source’s willingness to correct matters.
Version histories, errata, retractions and visible corrections provide information about how a knowledge-producing institution responds when evidence changes. The presence of a correction is not automatically a mark of unreliability. In many mature systems, correction is evidence that error-detection mechanisms exist.
Readers therefore need to distinguish a source that changes because it is accountable from one that silently rewrites its record. That distinction becomes increasingly important when AI systems quote cached or transformed versions of material whose original source has already been updated.
What the evidence does not show
Lateral reading is not a universal substitute for close reading.
Fact-checker expertise does not imply that every fact-checker judgment is correct.
Source reputation does not determine claim truth.
Teaching one digital-evaluation strategy does not guarantee transfer to all platforms, languages or high-stakes domains.
The Swedish ICILS-related analysis is descriptive and does not identify a causal teaching method.
And source criticism cannot infer malicious intent merely from disagreement, error or institutional affiliation.
What this changes for Inner Technology
The Institute treats source criticism as an outward-facing discipline of attention.
Instead of asking only, “What does this page say?”, the reader learns to ask:
Where did this come from? What does it depend on? Who can correct it? What independent trail exists? What is this source actually qualified to establish?
In AI-mediated environments, that sequence becomes part of human discernment because fluent content arrives faster than provenance can be felt.
Leave the page
The official-looking site is still open.
The reader opens another tab.
Then another.
They find the organization’s history, the original study, a correction, a second source that disagrees and a methodological note that changes the meaning of the headline.
The page itself did not become more or less beautiful.
The reader simply stopped asking appearance to do the work of provenance.
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. https://doi.org/10.1177/23328584231176451
- McGrew S, Reynolds EC. Discussing Credibility and Corroboration: Students’ Reasoning About Digital Sources and Claims During Lateral Reading Lessons. American Educational Research Journal. 2026. https://doi.org/10.3102/00028312261426323
- Skolverket. Lärares användning av digitala verktyg i undervisningen – En fördjupande analys av ICILS 2023. 2026. https://www.skolverket.se/sok-publikationer/publikationsserier/rapporter/2026/larares-anvandning-av-digitala-verktyg-i-undervisningen—en-fordjupande-analys-av-icils-2023?id=13369
- Cosgrove J, Cachia R. DigComp 3.0: European Digital Competence Framework, fifth edition. European Commission Joint Research Centre. 2025. https://doi.org/10.2760/0001149