Evidence for Finding good sources today #

Every substantive claim on the Finding good sources today page is checked against current research. Here is each claim, how well today’s evidence supports it, and the sources. The full, de-duplicated source list lives on the references page.

Supported · moderate evidence — Choosing which sources and materials to study from is part of the planning (forethought) phase of self-regulated learning, not a trivial preliminary.

Zimmerman locates task analysis and strategic planning — including selecting the resources and methods one will use — within the forethought phase of self-regulated learning, so the page’s treatment of source selection as a deliberate strategic step is consistent with the standard SRL framework.

Sources: Zimmerman (2002), Becoming a Self-Regulated Learner: An Overview, Theory Into Practice — https://doi.org/10.1207/s15430421tip4102_2 · full reference ›

Supported · moderate evidence — Evaluating the credibility of online sources is a genuine, learnable skill, and most people (including educated adults and students) do it poorly by default.

Wineburg & McGrew compared PhD historians, Stanford undergraduates and professional fact-checkers evaluating online information and found that the historians and students were frequently fooled, while trained fact-checkers were far more accurate — evidence that source evaluation is a distinct, trainable expertise rather than a given. Findings come from a modest expert/novice sample.

Sources: Wineburg & McGrew (2019), Lateral Reading and the Nature of Expertise, Teachers College Record — https://doi.org/10.1177/016146811912101102 · full reference ›

Supported · moderate evidence — Cross-checking a claim against other independent sources, rather than judging a page on its own internal cues, is an effective way to assess credibility.

The ’lateral reading’ strategy identified by Wineburg & McGrew — leaving a page to consult what other sources say about it before trusting it — distinguished expert fact-checkers from less accurate evaluators, supporting the page’s advice to cross-check important points against several independent sources.

Sources: Wineburg & McGrew (2019), Lateral Reading and the Nature of Expertise, Teachers College Record — https://doi.org/10.1177/016146811912101102 · full reference ›

Supported · strong evidence — Material that feels fluent, polished or confident can seem more credible or better understood than it actually is, so surface polish is an unreliable guide to quality.

The illusion-of-fluency effect — that ease of processing inflates judgements of comprehension and learning relative to actual understanding — is well established in cognitive and educational psychology, supporting the page’s caution that slick or confident sources can feel authoritative while being thin or wrong.

Sources: Dunlosky, Rawson, Marsh, Nathan & Willingham (2013), Improving Students’ Learning With Effective Learning Techniques, Psychological Science in the Public Interest — https://doi.org/10.1177/1529100612453266 · full reference ›

Supported · strong evidence — AI assistants and other generative language tools can produce fluent, confident output that is factually incorrect (hallucination), so their answers should be verified against trusted sources rather than taken as authoritative.

Hallucination — the generation of plausible but unfaithful or fabricated content — is a documented, well-characterised failure mode of large language models surveyed by Ji et al. (2023) and remains a recognised limitation in 2026, directly supporting the page’s advice to treat AI output as a starting point to verify, not a final authority.

Sources: Ji et al. (2023), Survey of Hallucination in Natural Language Generation, ACM Computing Surveys — https://doi.org/10.1145/3571730 · full reference ›

Supported · moderate evidence — Explaining or articulating your own understanding to another learner helps reveal gaps in that understanding.

Generating an explanation in one’s own words (self-explanation, and the act of retrieval involved in articulating knowledge) is a moderately well-supported study technique in Dunlosky et al.’s review; explaining to a peer exposes what one cannot in fact reconstruct, consistent with the page’s claim, though benefits depend on how the explaining is done.