Researchers studied ChatGPT health advice without the developer’s involvement by collecting 202,590 donated conversations from 1,252 users in India, Nigeria, Brazil and Pakistan (December 2022 to February 2026); an AI judge applying a published rubric rated roughly one in five health conversations high-risk.

SocioLens project,

Claim LD08-SL-C23 (v1.0)

Provenance and status

Claim type: Estimate
Publication status: Eligible for publication, with qualifiers: preprint
Verification state: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Does not license: A population estimate for all users, or a route that scales without paying donors
Limitations: Read in arXiv version 1 (13 October 2026); donors are a paid convenience sample of crowdworkers; risk dimensions and anchors are taken from Kempermann et al. (2026) and applied by an AI judge, with the authors’ own high-risk cut-off; scores measure risk, not health outcomes; the intent classifier was validated (κ = 0.78); data are released on OSF
Source (as cited): Roy Chowdhury & Garimella, “How Safe Is ChatGPT Health Advice? Evidence from Real User Conversations in Four Global South Countries,” arXiv:2610.18465, 2026
Source type: Preprint (recorded as “Preprint”)
Institutional position: Independent academic (Rutgers University) and an independent researcher
Maturity of the evidence: Emerging evidence
Recorded date: 16 October 2026
Currentness / correction: This is the current version. No correction or concern is recorded on its sources.

Sources this version cites

Roy Chowdhury & Garimella, “How Safe Is ChatGPT Health Advice? Evidence from Real User Conversations in Four Global South Countries,” arXiv:2610.18465, 2026 (Preprint)
Read the original

Version

This page shows version v1.0. Its record never changes, even when newer evidence arrives; the site may add a note to the page or withhold a figure not cleared for public use.

This pinned version is also the current version.

Versions: v1.0 · current

Raw record: every recorded field, with its coded values

All 25 canonical Claim fields, verbatim, with the coded values derived from them. The plain-language labels above are read from these.

Claim type (coded): ESTIMATE (as printed: “ESTIMATE”)
Publication status (coded): ELIGIBLE (PREPRINT) (as printed: “ELIGIBLE WITH PREPRINT QUALIFIER”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY (PREPRINT)
Source type (coded): PREPRINT
Maturity (coded): EMERGING

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD08-SL-C23	—
2	PUBLIC-FACING CLAIM	Researchers studied ChatGPT health advice without the developer’s involvement by collecting 202,590 donated conversations from 1,252 users in India, Nigeria, Brazil and Pakistan (December 2022 to February 2026); an AI judge applying a published rubric rated roughly one in five health conversations high-risk.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	B — Data and access	—
5	MATURITY	EMERGING	states: EMERGING
6	SOURCE	Roy Chowdhury & Garimella, “How Safe Is ChatGPT Health Advice? Evidence from Real User Conversations in Four Global South Countries,” arXiv:2610.18465, 2026	source_ids: SRC-SL-ROY-CHOWDHURY-GARIMELLA-ARXIV-2610-18465
7	SOURCE TYPE	Preprint	classes: PREPRINT
8	INSTITUTIONAL POSITION	Independent academic (Rutgers University) and an independent researcher	—
9	SCOPE	ChatGPT users in four countries who donated their full conversation histories	—
10	DESIGN	Consent-based, paid data donation using the product’s export function; rubric-based risk rating	—
11	SAMPLE / DATASET	202,590 conversations from 1,252 users; 15,555 health conversations	—
12	MODEL / TOOL / VERSION	ChatGPT (studied); Gemini 3.1 Pro Preview as the risk judge	—
13	COMPARATOR	Ratings with and without the user’s own context	—
14	MAIN RESULT	About one in five health conversations rated high-risk; in a context-aware subsample of 5,657 conversations, adding the user’s context raised the high-risk share by 13–17 percentage points, which the authors say does not extrapolate	—
15	INFERENTIAL TARGET	Whether data donation can give independent access to real AI use	—
16	DOES NOT LICENSE	A population estimate for all users, or a route that scales without paying donors	—
17	VERIFICATION STATE	CONTENT VERIFIED PREPRINT	state: CONTENT_VERIFIED_PRIMARY · qualifiers: PREPRINT · verified_by: UNKNOWN
18	LIMITATIONS	Read in arXiv version 1 (13 October 2026); donors are a paid convenience sample of crowdworkers; risk dimensions and anchors are taken from Kempermann et al. (2026) and applied by an AI judge, with the authors’ own high-risk cut-off; scores measure risk, not health outcomes; the intent classifier was validated (κ = 0.78); data are released on OSF	—
19	CONTRADICTORY / SUPPORTING SOURCES	LD08-SL-C22	—
20	GEOGRAPHY / LANGUAGE	Multi-country	—
21	DISTRIBUTIONAL LEDGER	Studies users in four lower- and middle-income countries	—
22	R01 / FIELD-MAP LINKS	LD-08, D3	—
23	WHAT WOULD CHANGE THIS CLAIM?	Evidence that donation samples match the platform’s user population	—
24	PUBLICATION STATUS	ELIGIBLE WITH PREPRINT QUALIFIER	state: ELIGIBLE · qualifiers: PREPRINT
25	RECORDED DATE	2026-10-16	—

Field 22 references resolved against Field Map v0.2: LD-08 (LIVING_DEBATE); D3 (DOMAIN)

The sources as recorded, with their record IDs and coded source types:

Pinned Sources (field 6, version-specific)

“Roy Chowdhury & Garimella, “How Safe Is ChatGPT Health Advice? Evidence from Real User Conversations in Four Global South Countries,” arXiv:2610.18465, 2026” → Roy Chowdhury & Garimella (2026), “How Safe Is ChatGPT…” (SRC-SL-ROY-CHOWDHURY-GARIMELLA-ARXIV-2610-18465--v1.0) — Roy Chowdhury & Garimella, “How Safe Is ChatGPT Health Advice? Evidence from Real User Conversations in Four Global South Countries,” arXiv:2610.18465, 2026; class PREPRINT (explicit public provenance pin)

Cited by Living Debate versions

Who can know about AI? Epistemic dependence and jurisdiction over the field (v0.1) LD-08

Cite this page

SocioLens project. (2026, October 16). Researchers studied ChatGPT health advice without the developer’s involvement by collecting 202,590 donated conversations from 1,252 users in India, Nigeria, Brazil and Pakistan (December 2022 to February 2026); an AI judge applying a published rubric rated roughly one in five health conversations high-risk [Claim LD08-SL-C23, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld08-sl-c23/v1.0/

@misc{sociolens_ld08_sl_c23_v1_0,
  author = {{SocioLens project}},
  title = {Researchers studied ChatGPT health advice without the developer’s involvement by collecting 202,590 donated conversations from 1,252 users in India, Nigeria, Brazil and Pakistan (December 2022 to February 2026); an AI judge applying a published rubric rated roughly one in five health conversations high-risk.},
  year = {2026},
  date = {2026-10-16},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld08-sl-c23/v1.0/}},
  note = {Claim LD08-SL-C23, version v1.0}
}

The link points to this version, whose record never changes. For a finding, cite the original study as well: SocioLens summarizes research and does not replace it.

Research-of-record provenance

Materialized by sociolens-materializer 0.6.0 at 2026-09-28T20:00:00.000Z; normalization contract SocioLens — Claim Value Vocabulary v1.0.2 — Pre-Spike Materialization Contract (v1.0.2). Content seal SHA-256 15748f6eb17e0a4a228fc4521f283b2e08a18f42defff755fed0154e3093d67d.

Repository-authored inputs (not Drive exports)

SocioLens claim — LD08-SL-C23 v1.0 (v1.0) — written in the SocioLens repository by Project Lead (Claude, AI) on 2026-10-16; no Drive identity exists. Authority: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's choice in the SocioLens project chat on 16 October 2026 to build the LD-08 debate; a new Claim, first version, written from the project's LD-08 source notes, checked against full texts, official web pages or official abstracts. Block “ENTRY 23 — LD08-SL-C23” SHA-256 7ecd4963c10f75710c4ed2fd1795e89a482ab504be62b371091cf18f55dea6a8; fixture fixtures/authored/claims/LD08-SL-C23--v1.0.

Builder decisions (not Drive data)

VERSION = v1.0 — basis: Claim version taken from the repository-authored claim fixture version (SocioLens claim — LD08-SL-C23 v1.0); the first version of a new Claim of the LD-08 debate
CURRENT_POINTER = true — basis: repository-authored claim "SocioLens claim — LD08-SL-C23 v1.0" (Project Lead (Claude, AI), 2026-10-16); basis: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's choice in the SocioLens project chat on 16 October 2026 to build the LD-08 debate; a new Claim, first version, written from the project's LD-08 source notes, checked against full texts, official web pages or official abstracts
FIELD_6_SOURCE_REFERENCE = SRC-SL-ROY-CHOWDHURY-GARIMELLA-ARXIV-2610-18465--v1.0 — basis: field 6 of ENTRY 23 — LD08-SL-C23 cites "Roy Chowdhury & Garimella, “How Safe Is ChatGPT Health Advice? Evidence from Real User Conversations in Four Global South Countries,” arXiv:2610.18465, 2026"; the only field-6 segment of this ClaimVersion's repository-authored ledger block.
SOURCE_PROVENANCE_PIN = SRC-SL-ROY-CHOWDHURY-GARIMELLA-ARXIV-2610-18465--v1.0 — basis: explicit version-pinned public provenance of this ClaimVersion (Amendment v1.2 §6): the pinned field-6 Source record "Roy Chowdhury & Garimella, “How Safe Is ChatGPT Health Advice? Evidence from Real User Conversations in Four Global South Countries,” arXiv:2610.18465, 2026".