Why does the AI reply in the wrong language?

Sometimes you write in Hindi and the AI answers in English, or you write in English and suddenly get a reply in another language. This is usually easy to fix.

Why does it happen?

1. Mixed signals. If your message mixes languages, such as Hinglish, or includes English technical terms, the AI may choose English.
2. Earlier conversation. If the chat started in English, the AI may continue in English even when you switch.
3. Custom instructions or memory. A preference saved earlier, such as "always reply in English" or a note that you speak Hindi, can override your current message.
4. Uploaded content. If you upload an English document and ask a question in Hindi, the AI may answer in English.
5. Location or account settings sometimes influence the default language.
6. Romanised Hindi. When Hindi is typed in English letters, the AI may not recognise it as Hindi.

Simple fixes:

1. State the language directly: "Please reply in Hindi" or "Answer only in English."
2. Specify the script: "Reply in Hindi using Devanagari script" or "Reply in Hinglish using English letters."
3. Specify the style: "Use simple spoken Hindi, not formal or Sanskritised Hindi."
4. Set a default in custom instructions or personal preferences, such as "Unless I say otherwise, reply in the same language I use."
5. Check memory. If the AI keeps choosing a language you do not want, review saved memories and instructions.
6. Start a new chat if the conversation's earlier language keeps pulling replies back.
7. In voice mode, specify the language at the start of the conversation.

For translation tasks, be explicit about both languages: "Translate this English email into formal Marathi." For bilingual output, ask clearly: "Give the answer in English, followed by a Hindi summary."

If you need accurate Indian-language output for official or public use, such as notices, legal documents or marketing for a large audience, have a fluent speaker review it. AI quality in Indian languages is good and improving, but phrasing can sometimes sound unnatural or overly formal.

Language control is mostly about being explicit. Once you tell the AI clearly, it usually follows reliably.