Same buyer question, different language. Does AI recommend the same brands?

I'm Zia, co-founder of Depra AI.

If you sell in India, do you check what AI recommends in Hinglish as well as English?

We collected 480 responses from ChatGPT, Gemini and Perplexity using 10 paired buying questions in skincare and fashion, with eight runs per language. We compared the language difference with each engine's normal rerun variation. In this sample, the language gap exceeded that variation on all three engines.

One finding: Gemini cited a source in 72 of 80 English answers and 33 of 80 Hinglish answers. That is 90% versus 41.3%. A citation rate is different from a brand recommendation rate.

This is a small study from one collection window, not a claim about every category. The method, responses and analysis code are public:

https://depra.ai/research/hinglish-vs-english-ai-shopping

https://github.com/ifham001/IndicGEO

We're launching Depra on October 12. It tracks buyer questions in English and romanised Hinglish, keeps the answers and sources, and helps teams choose what to work on next.

Preview: https://www.producthunt.com/products/depra-ai?launch=depra-ai

Have you seen a different set of brands when asking the same question in another language? Share the question and what changed. I'd also welcome criticism of our method.