Promises of protection: Where social media safety messaging and AI caregiver support fall short
· Medical Xpressby International Computer Science Institute
edited by Sadie Harley, reviewed by Robert Egan
Sadie Harley
Scientific Editor
Meet our editorial team
Behind our editorial process
Robert Egan
Senior Editor
Meet our editorial team
Behind our editorial process Editors' notes
This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:
fact-checked
preprint
proofread
The GIST Add as preferred source
Two new studies highlight how communication gaps can complicate efforts to protect vulnerable users of technologies such as social media and AI chatbots. Researchers will present the work at the 29th ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW 2026), held Oct. 1–14 in Salt Lake City.
One study—coming amid a swirl of lawsuits over the negative impacts of social media on youth—reveals ways platforms have fallen short in communicating about social media risks to children. The work is published on the arXiv preprint server.
"Social media companies are under enormous scrutiny right now over their impact on children, but an important part of that conversation is whether families and policymakers can actually understand what protections platforms are providing and whether those protections work," said Pamela Wisniewski, principal research scientist and leader of the Socio-Technical Interaction Research Group at the International Computer Science Institute.
"Our research shows that safety communications can often be selective, vague and difficult to verify. If we want meaningful accountability, platforms need to communicate much more transparently about what their safety features do, who has access to them and what evidence exists that they are effective."
A second study examines a different but increasingly important question: What happens when people begin relying on AI to help navigate some of life's most difficult and consequential situations? The researchers compared how chatbots, clinicians and peer caregivers responded to real questions from people caring for someone with Alzheimer's disease.
"Both studies are really about the potential gap between the promise of technological support and what people actually experience," Wisniewski said.
"With AI, that gap can be critical because people increasingly turn to these systems in moments of vulnerability. A response can sound helpful while still failing to recognize the real need behind the question."
How social platforms talk about youth safety
In the first study, the team examined how social media platforms communicate about online risks to children and their efforts to improve youth safety. Analyzing over 350 press releases and safety-related blog posts published by YouTube, TikTok, Meta and Snapchat over a five-year period, the researchers found notable gaps in the way platforms communicated about the availability of safety features, how they operate and their effectiveness.
For example, platforms sometimes described safety features as "rolling out," being tested or available only in certain regions, making it difficult to determine which protections were actually available to which users. Platforms also made claims that features were "improved," "stronger" or more effective without consistently providing baselines, evaluation methods or measurable outcomes.
The results showed that platforms communicated more about what young people see or whom they interact with, and less about how platforms collect or share young people's data, how children gain access to platforms, how youth-created content is circulated or how platforms may profit from or amplify that content.
By framing safety as a matter of individual behavior and parental oversight rather than platform design and business practices, researchers said platforms tended to communicate about these issues in ways that shift the responsibility for preventing harm away from the companies that structure these environments and onto young people and their families, even though many of the risks are shaped by decisions that users cannot see or control.
Researchers hope the findings can help inform guidelines for responsibly communicating about youth safety features.
When caregivers turn to chatbots
A second study, which received a CSCW 2026 Impact Recognition award, looks at how AI chatbots stack up against humans in offering advice to people caring for someone with Alzheimer's disease.
For this study, researchers analyzed responses to 85 real caregiver questions submitted by users on ALZConnected, an online community focused on dementia care. Using 28 measures of factors like linguistic style, emotional content and clinical content, they assessed responses to the same questions from four responder groups: peer caregivers, physicians, ChatGPT and CareGPT, a chatbot specially trained on peer-support discussions.
The results revealed a consistent pattern: peer caregivers provided time-anchored narratives and community grounding; clinicians provided concise, advice-forward, clinically bounded guidance; and chatbots produced a formal, verbose, highly structured "consultation" voice.
Notably, the analysis did not show much of a difference between CareGPT and ChatGPT responses, suggesting CareGPT's special training did not move the needle on making the chatbot responses seem more human.
Overall, chatbots scored higher than peer caregivers on politeness and emotion-support markers and were comparable to clinicians on informativeness and physician-alignment measures.
However, chatbot responses tended to be less narrative, less contextually situated and more narrowly focused on giving advice, while human responders, particularly peer caregivers, covered a broader range of issues, including emotional processing, legal and financial concerns and daily care.
"We need to be careful about equating a polished, supportive-sounding AI-generated response with meaningful support," Wisniewski said.
"The chatbots were very good at sounding helpful, but caregiving is messy, emotional and deeply contextual. Human caregivers brought lived experience and surfaced issues that the AI often missed. That matters, because as these systems become more embedded in health and caregiving, we risk designing for the appearance of support rather than the full complexity of what people actually need."
Technology's impacts on brain development
Later in October, Wisniewski will participate in a convening of experts examining how the use of digital technology affects brain development. She serves as a member of the planning committee for the National Academies of Sciences, Engineering, and Medicine workshop on "Digital Technologies and the Developing Brain: Insights from Neuroscience." The workshop will take place Monday, Oct. 26, and is open to the public.
Publication details
Renkai Ma et al, Analyzing Social Media Claims regarding Youth Online Safety Features to Identify Problem Areas and Communication Gaps, arXiv (2025). DOI: 10.48550/arxiv.2512.14965
Congning Ni et al, One Question, Four Voices: How Advice for Alzheimer's Caregiving Differs Between Caregivers, Clinicians, and Large Language Models, Proceedings of the ACM on Human-Computer Interaction (2026). DOI: 10.1145/3816980
Journal information: arXiv
Key medical concepts
Clinical categories
PediatricsChildren's healthHealthy agingPsychology & Mental health Provided by International Computer Science Institute Who's behind this story?
Sadie Harley
BSc Life Sciences & Ecology. Microbiology lab background with pharmaceutical news experience in oil, gas, and renewable industries. Full profile →
Robert Egan
Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile →
Citation: Promises of protection: Where social media safety messaging and AI caregiver support fall short (2026, October 5) retrieved 5 October 2026 from https://medicalxpress.com/news/2026-10-social-media-safety-messaging-ai.html This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.