If you’re building AI visibility tools for the Indian market, stop treating English prompts as the default. A recent study by depra.ai analyzed 480 AI answers collected from India on August 14, 2026, and found that asking the same buying question in Hinglish—Hindi and English mixed in Latin script—significantly alters brand recommendations. The impact varies wildly by engine: Perplexity’s brand list changed by 23.4 points beyond normal rerun noise, while ChatGPT barely budged at 2.2 points.

The Methodology Behind The Metrics

The researchers didn’t just eyeball the differences. They tested 10 buying questions (5 skincare, 5 fashion) in both English and Hinglish across ChatGPT, Gemini, and Perplexity. Each question was asked 8 times per language, resulting in 480 total answers. Crucially, they established a baseline called 'rerun noise,' measuring how much brand lists change when you simply ask the same question twice. This allowed them to isolate the true 'language effect' from random algorithmic variance.

Perplexity Is The Most Volatile

Perplexity emerged as the most sensitive to language shifts. While it was the most consistent engine on reruns (68.3% brand overlap), its recommendations dropped to 44.9% overlap when switching from English to Hinglish. This suggests that for developers and marketers relying on Perplexity’s Sonar API, Hinglish queries trigger a completely different retrieval logic. In contrast, ChatGPT showed minimal change, with a 2.2-point gap that the study notes is statistically weak and driven primarily by skincare prompts.

Source Citations Drop In Hinglish

Beyond brand names, the study uncovered a critical infrastructure issue for Gemini users. Gemini attached source links to 90.0% of English answers but only 41.3% of Hinglish answers. Furthermore, Hinglish responses were 33% shorter on average. This means that if your SEO strategy relies on being cited in AI answers, you might be invisible to Hinglish users on Gemini even if you rank well for English queries. The study didn’t determine why this happens, but the data is clear: the citation pipeline breaks down for mixed-language inputs.

Key Takeaways

  • Perplexity’s brand recommendations shift by 23.4 points when switching from English to Hinglish.
  • Gemini shows source links for 90% of English queries but only 41.3% of Hinglish queries.
  • ChatGPT remains relatively stable, with only a 2.2-point change in brand lists across languages.
  • Mass-market Indian D2C brands like The Derma Co gained visibility in Hinglish, while premium international brands like La Roche-Posay lost ground.
  • Raw data and analysis code are available in the public IndicGEO repository on GitHub.

The Bottom Line

If you’re tracking AI visibility for India, your English-only monitoring stack is blind to half the market. Developers need to build dual-language tracking pipelines immediately, because the algorithms behind Perplexity and Gemini are clearly treating Hinglish as a distinct retrieval context, not just a language variant.