Skip to content

Evaluating Sources

Principle · All principles
Evidence · 13 claims (7 for, 6 mixed) · 28 studies (12 causal, 5 review, 3 quant-synthesis, 3 qualitative, 3 theoretical, 1 associational, 1 design), q1–q4 · 1 of 28 report an effect size · 6 claims rest on one study

Conditional relationship

This page owns one judgment: whether to trust a source, and on what grounds. For an adolescent or adult learner who is a novice at judging unfamiliar sources, and whose starting response is to judge a web page by its own features (an official-looking logo, a .org or .edu domain, professional design, an "About" page), teaching a move that takes the reader off the page (search what independent sources say about who is behind it, then decide) and practising that move on live sites across several lessons is expected to produce better credibility judgments of new, unfamiliar sites (near-transfer, a procedure applied with judgment) than the learner's ordinary instruction. The gain has been measured at an immediate posttest and, in one small randomised pilot, five weeks later; nothing here measures it later than that, or in what learners do when they browse, share or cite outside class.

A claim tests part of this. Three studies of teaching professional fact-checkers' moves (lateral reading) compare taught classes with untaught ones: a district-wide high-school field study with matched controls, a randomised pilot in four university course sections, and a single-group study of eleventh graders (Civic Online Reasoning Instruction Improves Evaluation [+M]). None reports an effect size. What the evidence does not show: that the off-page move beats a checklist when each is taught in a single session (in one library session with nursing students the two did not differ, and both groups improved), that the gains last beyond five weeks, or that the same instruction helps judge research articles, policy documents or a primary source with no outside coverage. For those kinds of source no claim here tests any design.

The relationship is conditional on the kind of source (an organisation's website or a news story can be checked against what others say about it; a single historical document or a dataset often cannot); whether the learner has open web access and knows which outside sources are themselves reliable; the number of sessions; and whether the judgment is scored on reasons or only on the verdict. Learners who already check sources off the page have less to gain (an expertise-reversal pattern carried from cognitive-load research, not tested here).

How this sits beside the converted siblings. Epistemic Cognition holds the general relationship (teach the standard for one kind of evidence explicitly, have learners apply it to real claims and justify each judgment; gains stay with the standard taught) and its evidence across source evaluation, critique of studies and argumentation. This page is that model applied to source credibility, and adds what a designer of source-evaluation teaching needs that the general page does not: which move to teach, the dose and progression across sessions, what to do without open web access, and how to read a learner's trust judgment. Inquiry-based Learning uses source evaluation inside an investigation; Scaffolding and Fading holds how coaching is reduced, which the default design below uses and does not repeat.

Default design

Each step is a design proposal. A claim tests the whole package of teaching fact-checkers' moves over several lessons against no such teaching; it does not test any step on its own, so each step's evidence status is stated beside it. The earlier page's guidance (real source comparisons, explicit criteria, justification, contingent coaching) is kept and made concrete.

  1. Start with a judgment, not a lesson. Before any teaching, give learners two or three unfamiliar sources on one topic (one polished but unreliable, one plain but reliable) and ask which to trust and why, in writing, with a confidence rating. Five to ten minutes. This records the starting response and makes the problem visible. Untested proposal; the earlier page suggested "comparing a polished but unreliable site against corroborated sources"; the pretest-then-teach form follows the studies' designs.
  2. Teach one move first: leave the page. Model, thinking aloud, opening a new tab and searching the source's name (with words such as "funding" or "who runs") before reading the page closely, then deciding. Ten to fifteen minutes of demonstration on two sites. Evidence: the taught package in Civic Online Reasoning Instruction Improves Evaluation [+M]. Domain names and design are taught as weak signals, never as the test: in both one-session comparisons learners kept leaning on domain and design after training (One-session checklist (CRAAP) instruction did no better than the alternatives it was compared with [~M]).
  3. Practise on live sources where the verdict matters. Pairs or individuals judge new sites tied to a question the class is working on (a claim for an essay, a decision, a debate motion), so a wrong verdict would change the conclusion. Three or four sources per lesson, 20–30 minutes. Untested as a separate step; all the studies used live or realistic sites.
  4. Require a reason with every verdict. The learner writes what they found off the page, who is behind the source, and why that warrants trust or not; a verdict without a reason does not count. Carried from a different task: Prompting learners to self-explain improves understanding and problem solving on immediate tests by a moderate average amount, with little evidence yet on delayed or classroom gains. [+S] concerns explaining worked examples and texts in mathematics and science, not source judgments.
  5. Coach the weakness the attempt shows. Read the reasons, not only the verdicts: a learner who stayed on the page needs the move modelled again; one who searched but trusted an unreliable outside site needs a list of reference sources for that topic; one who reached the right verdict for a wrong reason needs a case where that reason misleads. Give the least help that gets the next attempt right. Carried from one-to-one tutoring of mathematics and science: Contingent scaffolding improves learning more than fixed or absent support. [+M]; untested for source evaluation or at class scale.
  6. Spread it over several lessons, and fade the teacher's part. A default arc, based on the doses in the studies (two 75-minute lessons at university, six 50-minute lessons in high school, eight lessons in eleventh grade): early lessons, teacher models each move on a new kind of source and learners practise with a prompt card (who is behind this? what do others say? what is the evidence?); middle lessons, learners work in pairs without the card and compare verdicts and reasons with a model answer shown afterwards; late lessons, individual judgments on unseen sources of mixed kinds, including some that look poor but are reliable, with help only on request. Plan at least four lessons of 45–75 minutes, two or more weeks apart in total span, embedded in a subject course rather than as a stand-alone unit. The lesson counts come from the claim; the three-stage fading, the four-lesson floor and the spacing are untested proposals (Scaffolding and Fading holds the fading model).
  7. Move a learner on when the reasons, not only the verdicts, are right. A learner is ready for unsupported work after correct verdicts with an off-page reason on three consecutive unfamiliar sources; a learner who already does this on the first-lesson judgment skips to step 8. Untested proposal; skipping guidance for those who already succeed follows Instructional guidance that helps novices can become redundant or counterproductive as expertise grows. [~M], carried from cognitive-load studies.
  8. Check later, on new sources, scoring the reason. Several weeks after the last lesson, give unseen sources and score both the verdict and whether the reason rests on outside evidence. Untested as a design step; the randomised pilot tested five weeks after its lessons, and the eleventh-grade study used constructed-response tasks.

When a condition is not met: with one session only (a library visit), teach step 2 and one practice round, expect modest gains, and say so; without open web access, supply a packet of outside documents about each source and teach learners to look for them; for research articles, teach the standards for judging a study separately (see the Epistemic Cognition page), since gains in one did not carry to the other.

Fitting the design to a situation

The three facts that most change the decision:

  • What kind of source, and what follows from the judgment. An organisation's website or a news story can be checked laterally; a research article is judged by its methods; a primary document with no outside coverage is read closely. If unstated, ask: what will the learner have in front of them when they decide what to trust, and what will they do with it (cite it, share it, act on it)?
  • How many sessions there are. The tested doses are two to eight lessons; one-session teaching of either method produced modest gains that did not differ by method. If unstated, ask: how many lessons, over what span, and is there a later task where the learner must judge new sources?
  • What the learner already does and knows. Learners who already leave the page have less to gain; learners who leave the page but consult unreliable sites need to know which reference sources are trustworthy in the topic. If unstated, ask: given an unfamiliar site, what does the learner do first, and can they name a reliable reference source for this topic?
If the brief says… Then change… Basis
adolescent learners in a school subject (government, history, science) Embed six or so 50-minute lessons in the subject's own topics, with teacher preparation beforehand; practise on live sites about the unit's questions Civic Online Reasoning Instruction Improves Evaluation [+M] (high-school field study with matched controls, and eleventh graders without a control)
adult learners at university Two or more 75-minute lessons inside a writing or critical-thinking course, with a check on unseen sites several weeks later Civic Online Reasoning Instruction Improves Evaluation [+M] (randomised pilot, four sections; five-week posttest)
adult learners re-entering study, or adult basic education Start from sources they already use (health, benefits, news, job information); teach the off-page move on those first; keep reading load of practice sites manageable and model searching step by step Untested; the claim's adults are university students. From the earlier page's target learners
child learners (primary age) Use a curated set of sources and a single question ("who made this, and what do others say about them?") asked with the teacher; do not expect the full routine Untested; no claim here tests source-evaluation teaching with children
Only one session (a library visit, an orientation) Teach one move (leave the page and search the source's name) and one practice round rather than a long checklist; tell the course teacher what to follow up One-session checklist (CRAAP) instruction did no better than the alternatives it was compared with [~M] (first-year business and nursing students; one session of either method, modest gains, no difference)
The existing course teaches a checklist (CRAAP or similar) Keep the checklist questions as prompts for what to look up elsewhere, not as on-page criteria; add practice on polished but unreliable sites where the checklist misleads One-session checklist (CRAAP) instruction did no better than the alternatives it was compared with [~M] (learners kept using domain and design cues after training); the re-purposing is untested
Goal is judging research studies, not websites Teach the standards for judging a study (valid methods, replication) as a separate target, with small-group critique and redesign of flawed studies; assess source credibility separately Collaborative critique-and-redesign instruction shifts students from a value-free view of objectivity toward a procedural one [~M] (one psychology course; source-credibility judgments did not change)
online-self-paced, no teacher present Short recorded demonstrations of the off-page move; practice items where the learner commits a verdict, a reason and a confidence rating before a model answer and the fact-checker's search are shown; a delayed check on new sites Untested; all studies were taught in classes
workplace-clinical learners (e.g. nurses and health information) Practise on the health sites and summaries they will meet, including misinformation sites; add judging study methods if they read research One-session checklist (CRAAP) instruction did no better than the alternatives it was compared with [~M] (nursing students, one 110-minute session); the multi-session design is untested in this setting
No open web access, locked devices, or an offline exam Supply a packet of outside documents about each source (reviews, a reference entry, a fact-check) and teach learners to look for them; or shift the target to the reasoning inside the source Untested; the claim's discussion names this constraint, and no study tests it
Learners already check sources off the page on the first judgment Skip the modelling; give harder cases (reliable sources that look poor, unreliable ones with good coverage elsewhere) and peer critique of reasons Instructional guidance that helps novices can become redundant or counterproductive as expertise grows. [~M] (cognitive-load studies, carried to this skill)
High stakes, or the skill is graded Grade on unseen sources with constructed responses, scoring the reason and the verdict separately; a verdict alone can be right for the wrong reason Untested; constructed-response tasks were used in the eleventh-grade study

Seven of the twelve rows cite a claim. The two learner rows and the one-session, checklist and clinical rows rest on studies of source evaluation (the last three on one-session comparisons, carried to a multi-session design where the row says so); the research-study row carries a study of judging research; the expertise row carries a cognitive-load review. The other five are untested proposals.

Observation, state and explanation

What can be observed is a verdict on a named source, the reason given, the confidence stated, and what the learner did before deciding (stayed on the page, searched, which outside sources they opened, how long it took). "Is a critical consumer of information" or "has digital literacy" are inferred states. A correct verdict does not show a sound method: it can come from agreeing with the source's conclusion, or from a domain-name rule that happens to fit.

Response Plausible explanations Observation that could change the interpretation
Trusts a site because it looks professional or has a .org domain Knows no other method; knows to search but did not have the time; the topic agrees with what they believe Allow time and search, and give a site with the same look whose conclusion the learner disagrees with; if they search only then, the issue is motivated, not lack of method.
Leaves the page, then trusts what an unreliable site says about it Does not know which outside sources are reliable for this topic; took the first search result Ask which of three outside sources they would rely on and why; if they cannot rank them, teach reference sources for the topic.
Distrusts everything ("you can't trust anything online") Blanket scepticism as a shortcut; did not see the difference between sources; avoiding a contested topic Give two sources on a low-stakes topic with clearly unequal backing; if the learner now ranks them with reasons, the response was topic-specific.
Correct verdicts in class, back to on-page judgments in their own assignments The move is tied to the class task; it costs time when the stakes feel low; the assignment did not ask for it Check the sources and reasons in a later assignment without prompting, then with one prompt; compare.
Reliable verdicts on websites, poor judgments of a study's methods Only the taught standard was learned; judging a study needs different knowledge Assess both on matched material at the same time and treat them as separate targets.

These rows are untested proposals. Each probe is itself practice in judging sources, so record what was asked and keep the explanations open.

Evidence and qualifications

  • Civic Online Reasoning Instruction Improves Evaluation [+M]: three studies of teaching fact-checkers' strategies. Teachers given professional development taught six 50-minute lessons on lateral reading in high-school government classes; treatment students (n = 271) grew significantly more than matched peers (n = 228) in judging the credibility of digital content (coded q2; the abstract calls it matched-control, the authors' project page cluster-randomised). In a randomised pilot in four sections of a university critical-thinking-and-writing course (n = 67), two 75-minute lessons made treatment students significantly more likely to gain from a pretest six weeks before to a posttest five weeks after (q3). Sixty-eight eleventh graders given eight lessons improved on three of four constructed-response tasks, with no control group (q2). No effect sizes. The claim (merged with the former "Lateral reading improves source evaluation", which is now its alias) predicts better judgment of unfamiliar web content after several lessons; it does not predict durability beyond five weeks, a change in behaviour outside class, or benefit for a primary source with no outside coverage. Its rigour codes are r2 for the pilot and r? for the other two. Not yet checked against its sources.
  • One-session checklist (CRAAP) instruction did no better than the alternatives it was compared with [~M]: limits the relationship. A think-aloud study of ten historians, ten fact checkers and 25 undergraduates found those who stayed on a site were misled by logos and domain names and the fact checkers, who left it, judged better and faster (descriptive; it taught nothing). In one library session with first-year business students, CRAAP and six question words did not differ on quizzes; papers from the question-word sections scored higher on a rubric (2.81 against 2.28 of 4), possibly because of prior library instruction. In one 110-minute session with first-term nursing students, accuracy in the CRAAP group rose from 41% before to 59% five weeks later and never differed from a networked, lateral-reading-style framework; there was no untrained control. So one session of either method gives modest gains, and on this evidence the off-page move has not been shown to beat a checklist in a single session. The claim's title was narrowed on 2026-10-05 from "checklist-based evaluation instruction does not reliably improve online source evaluation"; its Discussion still carries strength tags from before, which it says overstate the evidence.
  • Collaborative critique-and-redesign instruction shifts students from a value-free view of objectivity toward a procedural one [~M]: 120 first-year psychology students, a 12-week course with three small-group critique-and-redesign tasks; appeal to valid methods rose from 2% to 38% of written evaluations and to replication from 24% to 55%, while judgments about source expertise and outlet credibility did not change significantly. Single group, no control. It limits this page from the other side: teaching how to judge a study did not improve judging a source, so the two are separate targets.
  • Prompting learners to self-explain improves understanding and problem solving on immediate tests by a moderate average amount, with little evidence yet on delayed or classroom gains. [+S]: a meta-analysis of 64 reports (g = .55) and studies in mathematics and science find that prompting learners to explain while studying improves learning. It is the basis, carried, for requiring a reason with each verdict; no study here tests reason-giving in source evaluation. Not settled: most entries are read from abstracts.
  • Contingent scaffolding improves learning more than fixed or absent support. [+M]: fully contingent one-to-one tutoring of fourth and fifth graders in long division beat fixed, partly contingent or no support at immediate and one-month tests (8 per condition); a review notes most evidence is one-to-one. It is the basis, carried, for coaching the weakness a learner's reasons show; untested in source evaluation and at class scale. The earlier page cited it for this use.
  • Instructional guidance that helps novices can become redundant or counterproductive as expertise grows. [~M]: a review of cognitive-load studies. It is the reason to let learners who already judge sources off the page skip the modelling; it says nothing about source evaluation or when an individual crosses over.

Populations differ (US high-school and university classes on the open web; first-year business and nursing students in one library session; one psychology course), so do not rank these by effect label. None reports an effect size for source-evaluation teaching, none follows learners beyond five weeks, and two of the core studies have no control group.

The evidence by question

Further evidence, not yet read against this model

Claims the earlier page cited that are not used in the model above, and neighbouring claims found while rewriting it. None has been re-read against the model.

Objective and learner-valued goal

The designer's objective is usually that the learner, given an unfamiliar source, can say whether to trust it and why, on outside evidence, weeks after teaching. Elicit separately what the learner wants: to settle a question that matters to them (is this treatment safe, is this job offer real, who is right about a local issue), to finish an assignment with acceptable references, or not to be fooled in front of others. Agreement and divergence are observations. A learner who distrusts a source the evidence supports is not thereby a poor evaluator; ask for the reason.

For example, an adult returning to college may want to know whether a website offering student loans is legitimate, while the course objective is evaluating sources for a research essay. Using the loan site as the first practice case serves the learner's goal and teaches the move; the course objective is then checked separately on unfamiliar academic and news sources several weeks later. Record both, so that a correct verdict on the loan site is not read as the capability the essay needs.

What would revise this model?

The expectation should weaken if controlled comparisons with effect sizes find that several lessons of off-page evaluation do not improve judgments of new sources against ordinary teaching, or that gains fade within a few months. If multi-session checklist teaching matches multi-session lateral-reading teaching, the choice of move matters less than the dose and practice, and step 2 should say so. If learners who were taught evaluate sources no differently in their own assignments, the model should be limited to classroom tasks. If the doses in the studies prove larger than needed, the four-lesson floor should go. Do not protect the model by calling every failure too few lessons or too little background knowledge after the fact.

A better verdict on new sites, a better reason, durability over months, and changed browsing, sharing or citing outside class are separate claims. The evidence here reaches the first two, for web sources, at horizons up to five weeks. It establishes no optimal dose, no effect size, and nothing for research articles, primary documents or curated databases.

Claims

  • Inquiry-based Learning — inquiry quality depends on source quality
  • Annotating — annotation can focus attention on credibility signals, claims, and evidence
  • Clear Structure & Presentation — learners need explicit frameworks for evaluation, not just warnings about misinformation
  • Debate — debates become more rigorous when learners are accountable for source quality, not just rhetorical performance
  • Epistemic Cognition — the general model (teach the standard for one kind of evidence; gains stay with the standard taught) that this page applies to source credibility

Examples

Key Sources

  • Wineburg, S., & McGrew, S. (2017). Lateral reading: Reading less and learning more when evaluating digital information. doi:10.2139/ssrn.3048994
  • Breakstone, J., McGrew, S., Smith, M., Ortega, T., & Wineburg, S. (2018). Why we need a new approach to teaching digital literacy. Phi Delta Kappan, 99(6), 27-32. doi:10.1177/0031721718762419
  • Rapchak, M. E., Lewis, L. A., Motyka, J. K., & Balmert, M. (2015). Information literacy and adult learners. Adult Learning, 26(4), 135-142. doi:10.1177/1045159515594155
  • Afflerbach, P. A., & Cho, B. Y. (2010). Determining and describing reading strategies: Internet and traditional forms of reading. In Metacognition, strategy use, and instruction.
Page metadata
Field Value
type principle
id evaluating-sources
title Evaluating Sources
description For adolescent and adult novices who judge unfamiliar web sources by their own features, several lessons teaching and practising a move off the page (searching what independent sources say about who is behind it) are expected to improve credibility judgments of new sites up to five weeks later; one claim tests this package against untaught classes, though not against a checklist taught over the same number of sessions, nor for research articles or primary documents.
status review
canonical True
generated by claude/unspecified on 2026-10-05
trust unverified — no one has confirmed this page's content
sources 1 — rapchak-2015

Frontmatter is the page's machine-readable record — see the Schema & Guide.