fg-zettelkasten

Exposure to low-credibility online health content is limited and is concentrated among older adults

3 min read

Exposure to low-credibility online health content is limited and is concentrated among older adults

Summary

This paper combines a US national survey (n=1,059) with roughly four weeks of passive web and YouTube tracking in late 2023 to measure actual exposure to low-credibility online health content. The authors find that such exposure is rare in aggregate — about 3% of web health visits and 10% of YouTube health views — but heavily concentrated, with the top 10% of users accounting for 77% of exposure. Adults aged 60+ are disproportionately exposed, even controlling for overall browsing volume, and exposure correlates with poorer headline discernment, false cancer beliefs, conspiracism, and consumption of dubious political news. Crucially, referrals to low-credibility health sites come overwhelmingly from other low-credibility sites rather than search engines or social media, suggesting habitual cross-domain consumption rather than incidental encounters.

Key Contributions

  • Extends the digital-trace misinformation literature from political news into health, testing whether age-based vulnerability generalizes across domains.
  • One of the first studies pairing linked survey data with passive web and YouTube traces specifically for health misinformation.
  • Introduces a scalable hybrid pipeline (embeddings + fine-tuned GPT-4o + manual coding) to classify health content and rate credibility, validated against human coders.
  • Maps referral pathways into low-credibility health sites, with platform-design implications: search and social are not the main gateways.
  • Validates trace-based exposure measures by linking them to survey misperception and discernment outcomes.

Methods

The authors linked YouGov’s Pulse Panel survey data to RealityMine passive metering of 9.7M URLs across devices over ~4 weeks per participant. Health relevance was classified using YouGov tags plus OpenAI text embeddings; 1,155 health domains were hand-coded for credibility (78 flagged as low-credibility, Krippendorff’s α = 1.00). YouTube health videos (~3,900) were rated via a hybrid manual + fine-tuned GPT-4o procedure (κ = 0.89–0.97). Survey instruments measured cancer headline discernment, false risk-factor beliefs, conspiracism, ideology, and demographics. Preregistered OLS regressions with robust standard errors and survey weights were supplemented by logistic and negative binomial models, plus temporal referrer-sequence analysis.

Findings

  • 13% of respondents visited any low-credibility health domain in the 4-week window; mean = 0.5 visits, max = 122.
  • Exposure concentration: top 1% of users = 37% of all low-credibility exposure; top 10% = 77%.
  • Adults 60+ were substantially more likely to be exposed than 18–29 year olds, robust to controls for total health and total web activity.
  • Poorer cancer headline discernment and stronger false-risk-factor beliefs predicted greater exposure, with effects strongest among older adults.
  • Older conservatives and older heavy consumers of right-leaning partisan news showed the highest exposure.
  • Referrals came mainly from other low-credibility sites (especially for 60+), not from search engines or social media.
  • On YouTube, absolute low-credibility exposure did not differ by age, but it constituted a larger share of older adults’ overall lighter YouTube health diet.
  • Dubious political news consumption and low-credibility health web visits independently predicted low-credibility YouTube health exposure, indicating cross-platform clustering.
  • A media literacy intervention from a companion experiment produced no significant downstream effects on exposure.

Connections

This paper directly extends the digital-trace tradition that has documented older adults’ heightened engagement with untrustworthy political content into the health domain, complementing platform- and content-focused work on health misinformation ecosystems like Rothut2026-or and Bollenbacher2026-vz. Its finding that low-credibility consumption clusters across topics and platforms resonates with cross-domain perspectives on problematic information consumers seen in Efstratiou2026-ij. The null result for a media literacy treatment also speaks to ongoing debates about intervention efficacy explored in Scalco2026-bd and Song2025-yh.

Podcast

A research-radio episode discusses this paper: Listen

Graph View

Backlinks