CoChat

Evolving Health Information–Seeking Behavior in the Context of Google AI Overviews, ChatGPT, and Alexa: Interview Study Using the Think-Aloud Protocol

Open access
Wardle, Claire, Urbani, Shaydanay, Wang, Eric · 2025 · Journal of Medical Internet Research

Published version of a preprint

The DOI matches its Crossref record; 96 of the 96 references it cites were found in a registry or an index; 14 of the 44 sources its claims cite were checked for support.

Abstract

We conducted in-depth qualitative research with 27 participants (ages 19-80 years) using a think-aloud protocol. Participants searched for health information across 3 platforms—Google, ChatGPT, and Alexa—while verbalizing their thought processes. Participants integrated AI tools into their broader search routines rather than using them in isolation. ChatGPT was valued for its clarity, speed, and ability to offer broad overviews, but participants often questioned its source transparency and accuracy.

Overview

CoChat Summary

TL;DR

This qualitative think-aloud interview study with 27 participants found that people do not simply “trust” or “reject” AI health tools; instead, they mix Google AI Overviews, ChatGPT, and Alexa into broader search routines depending on context, urgency, and perceived usefulness. Across platforms, participants valued speed and summarization but were wary of missing sources, bias, and limited or inaccessible responses, especially for more serious health questions.

Key findings

  • Of 27 participants, 23 (85%) reported favorably on some aspect of ChatGPT’s formatting or efficiency, while 14/27 already used it as part of their search routine.
  • Google’s AI Overviews were often treated cautiously: about one-third of participants commented on them, 5/27 explicitly skipped them, and no participant reported stopping their search at the overview.
  • Alexa was seen as useful mainly for low-stakes tasks: 17/27 (63%) said they would not typically use it for the study questions, while 8/27 (30%) said short voice responses were convenient for simple questions.
  • Utility often outweighed trust: 15/27 (56%) said they would still use AI-generated information if it helped direct further investigation, and participants frequently described AI tools as “starting points” rather than final authorities.

Methods

  • Qualitative interview study using a think-aloud protocol with 27 participants aged 19-80 years, conducted in summer and fall 2024.
  • Participants searched for two health prompts on three platforms: Google, ChatGPT 3.5, and Alexa; one prompt was a standardized St. John’s wort question and the other was a personally relevant health query.
  • Transcripts (370 pages) were analyzed by the three researchers using reflexive thematic analysis, with 34 codes developed and organized around two research questions.

Limitations

  • The paper states the sample was small, recruited through loose personal-network connections, and not longitudinal; it warns the findings cannot be generalized to a wider population.
  • The think-aloud protocol may have produced a Hawthorne effect, with participants changing behavior because they knew they were being observed.
  • The paper notes that the technologies changed during data collection, so findings may quickly become outdated; this is an evident limitation given platform updates and evolving AI Overviews/ChatGPT source-linking behavior.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.2196/79961

Cites

96 of 96 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

11 Valid · 2 Not valid · 17 Not checked · 109 Own results and interpretations

76 more claims

Create a free account to see all 139 claims and to verify the 8 references still open.

See all claims — free account
  • Probably not a substantial claim. 36
  • It may state two separate claims as one. 29

Create a free account to see these claims.

Citations

Cites

Research works printed but not linked: 3

  1. Hovland, CI, 1953, Communication and Persuasion; Psychological Studies of Opinion Change, xii
  2. Amazeen, MA, 2023, Int J Commun, 17, 23
  3. Oh, P, 2023, Research Handbook on Artificial Intelligence and Communication

Cited by

See all 43

Metadata

Authors
Wardle, C., Urbani, S., & Wang, E.
Affiliations
  • Cornell University (US)
  • Brown University (US)
Corresponding author
Wardle, Claire
Venue
Journal of Medical Internet Research
Published
2025-10-07
Volume
27
Pages
e79961
ISSN
1438-8871 (ISSN-L)
OpenAlex
W4414280203
Cited by
43
References
94
Citations by year
2026: 43
FWCI
53.84
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Journal of Medical Internet Research (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Health Sciences › Health Professions › General Health Professions › Health Literacy and Information Accessibility
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Information Systems › Information Retrieval and Search Behavior
Keywords
health information seeking, AI Overviews, ChatGPT, voice assistants, Amazon Alexa, large language models, search behavior, trust in AI, credibility assessment, perceived usefulness, multilingual users, digital health literacy, platform selection, data provenance, voice recognition, cultural relevance, Google search results, user demographics
MeSH terms
Generative Artificial Intelligence, Adult, Aged, Aged, 80 and over, Artificial Intelligence, Female, Humans, Interviews as Topic, Male, Middle Aged, Internet, Qualitative Research, Young Adult, Information Seeking Behavior

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.