Ask in another language and the citations get thinner: a 15,942-citation multilingual audit
Non-English queries surface fewer citations and are routed to language-appropriate sources far less often — translation alone is not a multilingual strategy.
Key numbers
- Audit: 20 English mental-health questions under two prompt conditions, a subset translated into six further languages — 1,140 responses, 15,942 citations and 1,713 unique domains recorded.
- Citations are heavily concentrated: the ten most-cited domains account for 43.6% of English citations, with government, commercial health and academic sources split at roughly 22% each.
- Platforms differed little in citation volume but sharply in consistency and source-type preference; explicitly requesting sources shifted the mix only modestly.
- Non-English queries surfaced fewer citations and were routed to language-appropriate resources at significantly lower rates.
Method and sample
arXiv paper “Sources of Truth: A Multi-Platform, Multilingual Audit of Citations in AI Mental Health Information Queries” (2026-08-31): 20 mental-health questions run on ChatGPT, Perplexity and Google AI Overview under two prompt conditions, partly translated into six more languages, with all 15,942 citations classified into a nine-category typology using a human-validated classifier.
What it means for brands
The same question in another language produces a weaker source structure: fewer citations and fewer local-language sources. Multi-language markets need their own projects and their own measurements — English results do not predict other languages.
How SourceSight handles it
SourceSight runs one project per target market with prompts in that market's language, reports mentions and citations per language and market, and builds the off-site list from that market's platforms rather than an English list.
Limits
The sample is concentrated in a sensitive health domain, and one of the three audited platforms is outside our monitoring scope; the exact distribution of citations and source types does not transfer to other industries.
References
Run it on your own project
What the same prompts return across 10 engines is only knowable by running them.
Related research
What Chinese generative search actually cites: 8.3% brand selection, and app vs web cite different sources
Across 160,860 citation-level records, the same platform surfaced systematically different source sets on its app and web interfaces.
Sources · 2026-10-06Incumbent advantage: with identical specs, a well-known brand is recommended 100% of the time
Three experiments on skincare: identical specifications produce near-total dominance for known brands — and less than a +0.1-star edge for a competitor breaks it.