Skills
Searching and checking what you find
Lateral reading is the one verification habit with randomised trial evidence, and no published trial has tested whether the gains outlast a school term.
Everything measured about how people judge online sources points the same way. In the largest US assessment of the skill, high school students were shown an anonymous Facebook video shot in Russia and asked whether it was good evidence of voter fraud in the United States; 52% called it strong evidence (Stanford History Education Group, 2019). Adults fare little better at the judgement a results page demands constantly, telling a paid placement from an unpaid one. One remedy has real trial evidence: lateral reading, which means leaving a page to find out who produced it. It produced large gains in a randomised trial and a cluster-randomised trial. Nothing published shows those gains lasting beyond a school term or appearing in unprompted browsing. For a practical example of software used to turn activity into operational metrics, see this reference.
Students fail source evaluation at every level tested
The Stanford assessment used short written tasks: say who is behind a site, whether a video shows what it claims, whether a page is news or advertising.
90% of students received no credit on four of the six civic online reasoning tasks, and fewer than 3% of all responses earned full credit.
Direction: Decrease. Strength of evidence: Strong.
Caveat A convenience-weighted school sample, with mastery rubrics that may understate partial competence.
The failures were failures to ask who paid: 96% did not consider that fossil fuel industry ties might undermine a climate website's credibility, and about two-thirds could not tell news from sponsored content (Stanford History Education Group, 2019).
An earlier study by the same group collected 7,804 responses across 56 tasks in 12 states in 2015-16, from middle school to six universities including highly selective ones, and found weak evaluation at every level (Wineburg and colleagues, 2016). It is an unreviewed working paper counting responses rather than students, so it corroborates rather than confirms.
Lateral reading is the technique with the trial evidence
Lateral reading is what professional fact-checkers do: rather than scrutinising a page, leave it and find out what other sources say about its author or publisher.
61% of students taught lateral reading made a correct trustworthiness judgement on at least one problem using the strategy, against close to 0% of controls.
Direction: Increase. Strength of evidence: Strong.
Caveat Sections were matched rather than individually randomised, and there was no follow-up.
Six one-hour lessons for 271 students, against 228 controls, produced significant growth in the ability to judge the credibility of digital content.
Direction: Increase. Strength of evidence: Strong.
Caveat One urban district, and the accessible record reports significance without an effect size, so magnitude should not be quoted.
Taught inside ordinary subject lessons, the same approach moves scores by a real but modest amount.
Mean civic online reasoning scores rose from 2.25 to 3.75 out of 9 (F(1,572) = 299.91, p < .001), leaving students answering fewer than half the questions correctly.
Direction: Increase. Strength of evidence: Mixed.
Caveat No randomised control group, so maturation and testing effects cannot be ruled out.
Nothing shows the habit outlasts the term
This is the part usually left out. The college trial had no follow-up, and the classroom trials measured outcomes at the end of instruction, on tasks that told students to evaluate something. Whether a student who learned to open a second tab in March still does it in September, unprompted, on material they want to believe, is unpublished. Claims that a short course inoculates people against misinformation go past anything these studies tested. The skill can be taught in weeks; its durability is unmeasured — still better evidence than exists for most of what is taught with technology in classrooms.
Adults misread the results page
The commercial structure of a results page matters most often and is checked least.
Only 52% of UK search engine users correctly identified sponsored results as paid-for placements in 2025, and a further 37% felt confident about spotting advertising but answered incorrectly.
Direction: Decrease. Strength of evidence: Strong.
Caveat A survey scenario rather than observed behaviour, and the 2018 comparison of 48% used a differently formatted page.
In the same report, 72% of UK adults said they feel confident judging whether online information is true or false, while 11% admitted clicking results they simply like the look of, up from 6% in 2024 (Ofcom, 2026). The two figures come from different question batteries and should not be combined, but the pattern matches the gap between confidence and measured skill elsewhere.
AI summaries changed what a search result is
The page people were taught to read is being replaced by one that answers the question itself.
Users clicked a traditional search result on 8% of visits to pages carrying an AI summary versus 15% without, only 1% clicked a link inside the summary, and 26% ended the session versus 16%.
Direction: Decrease. Strength of evidence: Strong.
Caveat Observational, and searches that trigger summaries differ systematically, so this is not a causal estimate.
The summaries are themselves unreliable at saying where information came from. Testing eight AI search tools on 1,600 queries, the Tow Center for Digital Journalism found them wrong on more than 60%, with Grok-3 wrong on 94% (Columbia Journalism Review, March 2025). That test was narrow and tied to model versions that have since changed, so no such figure means anything without a model name and a test month. Meanwhile 41% of UK adults said they had met misinformation online in the previous four weeks in June 2025 (Ofcom, Online Nation 2025) — suspicion rather than verified exposure, and worth reading alongside who sees and shares false content.
The short version
- Fewer than 3% of responses earned full credit on source evaluation tasks across 3,446 US high school students (Stanford History Education Group, 2019).
- Lateral reading — leaving the page to check who is behind it — is the only search skill with randomised evidence that teaching works.
- None of those trials followed students afterwards, so persistence and transfer to everyday browsing are unmeasured.
- Just over half of UK search users identified sponsored results as paid placements in 2025, while 72% felt confident judging online information (Ofcom, 2026).
- On Google visits with an AI summary, 8% of users clicked a traditional result against 15% without one (Pew Research Center, 2025).