Back

Asymmetry between warmth and clinical substance in multilingual consumer health AI

Ariel, D.; Grumberg, L. R.; Supakul, S.; Wannasri, S.; Mitchnik, I. Y.; Lev, A.; Ariyamethanon, W.; Agbarieh, M.; Miari, S.; Laban, G.; Hasid, B.

2026-05-14 health informatics
10.64898/2026.05.09.26352813 medRxiv
Show abstract

The same patient question can yield different clinical quality across languages. Across 504 forum-derived patient queries in six languages and four chatbots, language-matched clinicians rated responses on five clinical dimensions (1,008 ratings; 5,040 dimension scores). Patient language outweighed chatbot identity across the four clinical-substance dimensions (composite language partial {superscript 2} 0.275 vs chatbot 0.035; robust to investigator-rating exclusion: {superscript 2} 0.260) but not for empathy ({superscript 2} 0.029): clinical substance was language-associated; warmth was relatively preserved. Catastrophic safety ratings ranged 4.3-fold by language (3.6% English, 15.5% Thai and Hebrew); 62% of catastrophic ratings exceeded the English baseline (descriptive disparity). Failures were systematic and silent: none of 24 stroke responses conveyed time-criticality framing, none of 24 CO-poisoning responses challenged the familys stress framing, and 120 sentinel responses contained no confident errors. Warmth did not discriminate clinical danger (response-level empathy AUC = 0.49): consumer health AI can deliver fluent, caring tone with degraded clinical substance.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

1
npj Digital Medicine
97 papers in training set
Top 0.1%
33.9%
2
BMJ Open
554 papers in training set
Top 3%
7.0%
3
Scientific Reports
3102 papers in training set
Top 16%
6.5%
4
Frontiers in Digital Health
20 papers in training set
Top 0.1%
5.0%
50% of probability mass above
5
Journal of the American Medical Informatics Association
61 papers in training set
Top 0.7%
3.8%
6
Journal of Medical Internet Research
85 papers in training set
Top 1%
3.7%
7
PLOS ONE
4510 papers in training set
Top 47%
2.1%
8
Nature Communications
4913 papers in training set
Top 47%
2.1%
9
The Lancet Digital Health
25 papers in training set
Top 0.3%
1.9%
10
Philosophical Transactions of the Royal Society B
51 papers in training set
Top 3%
1.8%
11
PLOS Digital Health
91 papers in training set
Top 2%
1.5%
12
Healthcare
16 papers in training set
Top 1%
1.3%
13
JAMA Network Open
127 papers in training set
Top 3%
1.3%
14
Nature Human Behaviour
85 papers in training set
Top 3%
1.3%
15
Journal of Biomedical Informatics
45 papers in training set
Top 1%
1.3%
16
eBioMedicine
130 papers in training set
Top 3%
0.9%
17
eLife
5422 papers in training set
Top 55%
0.8%
18
Journal of Personalized Medicine
28 papers in training set
Top 1%
0.8%
19
BMJ
49 papers in training set
Top 1%
0.8%
20
Clinical and Translational Science
21 papers in training set
Top 1.0%
0.8%
21
Computers in Biology and Medicine
120 papers in training set
Top 4%
0.8%
22
Annals of Internal Medicine
27 papers in training set
Top 0.9%
0.8%
23
eClinicalMedicine
55 papers in training set
Top 2%
0.8%
24
JAMIA Open
37 papers in training set
Top 1%
0.8%
25
Med
38 papers in training set
Top 0.8%
0.8%
26
Nature Medicine
117 papers in training set
Top 5%
0.8%
27
BMC Medical Informatics and Decision Making
39 papers in training set
Top 3%
0.7%
28
Journal of Clinical Epidemiology
28 papers in training set
Top 0.6%
0.7%
29
Frontiers in Psychiatry
83 papers in training set
Top 3%
0.7%
30
Journal of General Internal Medicine
20 papers in training set
Top 1%
0.7%