Hinglish chats and Hindi voice notes, handled by AI

If you run a salon in Bandra, a physiotherapy clinic in Saket or a coaching centre in Koramangala, you already know the pattern. Very few customers write pure English, and almost nobody types formal Hindi. They write Hinglish: Hindi words in Roman letters, English nouns in the middle, and spelling that changes from one message to the next. So the real test for any WhatsApp chatbot in Hindi and Hinglish is not whether it can translate a textbook sentence. It is whether it can read "bhaiya kal shaam 6 ke baad koi slot hai kya, hair spa ka" and turn it into a booking.

A typical evening inbox at an Indian service business looks like this:

• "Facial ka rate kitna hai? Aur Sunday open ho?"
• "Dr sahab kal OPD me hain? 11 baje ka milega?"
• "Mera beta class 9 me hai, maths ka demo kab hai"
• A 40-second voice note in Hindi, recorded in traffic, asking to move Saturday's appointment.
• "ok done" followed by a thumbs-up, which may or may not mean a confirmed booking.

Each of these needs a different reply, and most arrive after 8pm, when the front desk has gone home. That is the gap NextReply's WhatsApp AI receptionist for India is built to close: it answers the chat, asks the one missing question and writes the booking into your calendar.

Below are three example conversations of the kind an Indian business receives every day. They show what a well-configured AI should do at each step. Use them as a checklist when you test any tool, including ours.

1. A salon in Andheri West, 10:40pm

Customer: "Hi, kal hair colour karwana hai, global. Kitna lagega aur 5 baje ho jayega?"
AI: "Hi! Global hair colour starts at ₹2,800, depending on length. Kal (Saturday) 5 baje Priya available hain, about 2 hours lagenge. Book kar doon? Aapka naam bata dijiye."
Customer: "haan karo, Neha"
AI: "Done, Neha. Saturday 5:00pm, global hair colour with Priya. Address aur map link bhej rahi hoon. Change karna ho toh yahin message kar dijiye."

What went right: the AI quoted the price from the salon's own service list, read "kal" as tomorrow from context, checked a real stylist's calendar, asked for the name only once and confirmed in a structured line. It also kept the customer's mix of Hindi and English instead of switching to formal English.

2. A clinic in Gurugram, Sunday morning

Customer: "Dr Mehta Monday ko baithte hain? Pet me dard hai 3 din se"
AI: "Dr Mehta Monday ko 10am se 1pm OPD me hain. 10:30 ya 11:15 khaali hai. Kaunsa time chalega? Agar dard bahut zyada hai ya bukhar bhi hai, please abhi nearest emergency jaiye ya clinic ko 0124-XXXXXXX par call kijiye."

What went right: it answered the scheduling question and did not give medical advice. The symptom triggered a safety line written by the clinic, not by the AI. Our OPD booking guide goes deeper into this pattern.

3. A parent asking a coaching institute

Customer: "Sir JEE foundation batch ka fees structure bhejo. Weekend batch hai?"
AI: "Namaste! Class 9 JEE Foundation: ₹48,000 per year, 2 instalments. Weekend batch Saturday-Sunday 9am-12pm. Free demo class is Sunday 9am. Aapke bache ka naam aur class bata dijiye, main demo seat book kar deta hoon."

What went right: the reply answers both questions in one message, offers a concrete next step and books the demo. For more on this vertical, see our post on coaching demo bookings.

Modern language models handle Hinglish far better than the keyword bots Indian businesses tried a few years ago. Those old bots matched exact phrases, so "rate" worked and "rates kya hai" failed. A language model reads meaning, which matters for these everyday patterns:

• Spelling drift. "Kitna", "kitne", "ktna" and "kitna hai bhai" all mean the same thing. So do "kal", "kl" and "kal ko".
• Script mixing. Some customers switch to Devanagari mid-chat ("ठीक है, 5 बजे"). The AI reads both scripts in one conversation.
• Time words. "Saade paanch" is 5:30, "paune chaar" is 3:45, "subah" and "shaam" set am or pm, and "parso" means the day after tomorrow.
• Politeness levels. You can tell the AI to use "aap", never "tum", and to address older patients respectfully.
• Two questions in one line. "Rate kya hai aur parking hai?" gets both answers, not just the first.

The tricky one is "kal". In Hindi it means both yesterday and tomorrow. In a booking chat it almost always means tomorrow, but NextReply's confirmation line always names the weekday and date ("Saturday, 17 October"), so the customer catches any mistake before it reaches your calendar.

Voice notes are the part most Indian businesses worry about, and with reason. Customers record them while driving, in a crowded market or while holding a child. A Hindi voice note chatbot has to cope with noise, speed and mixed vocabulary.

NextReply understands voice notes as part of the conversation. The AI turns the audio into text, reads the meaning and replies in writing, in the language the customer used. A typical note sounds like: "Haan hello, main Rakesh bol raha hoon, Saturday ka appointment tha na 11 baje ka, usko Monday kar do agar ho sake, same time." The AI understands it as a reschedule request for Rakesh's Saturday booking and, on NextReply, passes it to your team with the booking details and the Monday 11am preference, so nobody has to replay the voice note.

Where voice notes need care:

• Heavy background noise. Horns and loud music can drop a word. When the date or time is unclear, the AI should ask a short follow-up ("Monday 11am, sahi hai?") rather than guess.
• Names and amounts. Proper names can be misheard. The confirmation message repeats the name so the customer can correct it.
• Very long notes. A two-minute note that mixes a complaint with a booking request should go to your team. The handoff rule covers this.
• Other Indian languages. Many customers in Chennai, Kolkata or Pune message in Tamil, Bengali or Marathi. NextReply replies in the customer's own language, but test your main local language with real samples before you go live.

Photos work the same way. A customer can send a screenshot of a reference hairstyle or a photo of a prescription page, and the AI can read it as context. For anything clinical, the AI passes the chat to staff.

No language model is perfect, and Hinglish adds ambiguity. The fix is not a smarter model. It is clear rules. These are the ones we recommend for Indian businesses:

1. Confirm in a fixed format. Service, date with weekday, time, staff name and price. Free-flowing chat is fine, but the final line should be structured.
2. Never invent prices. The AI only quotes what is in your service list. If a customer asks for "bridal package with mehendi" and you have not listed it, the AI says your team will share the quote and alerts you.
3. Hand off complaints and refunds. "Pichli baar bahut kharab cut hua" is not a booking chat. It goes to a human straight away.
4. Hand off clinical questions. Clinics should list trigger words and always route symptoms, reports and medicine questions to staff.
5. Let customers ask for a person. "Kisi insaan se baat karao" should always work.

NextReply sends an instant owner notification when a handoff happens, so a person can step in from the WhatsApp Business app on their phone. Read how this works on the features page.

The best default for India is simple: answer in the language and script the customer used. Hinglish in, Hinglish out. Devanagari in, Devanagari out. English in, English out. Switching a Hinglish customer to stiff English feels like talking to a call centre, and switching an English-speaking corporate client to Hinglish feels too casual.

You can adjust this. A premium dental clinic in South Mumbai may want English replies with a few Hindi courtesies. A neighbourhood salon in Lucknow may want warm Hindi with "ji" everywhere. NextReply lets you set the personality and tone, then the AI applies it across every language it replies in.

One practical tip: add Hindi aliases to your service list. "Threading" is also "eyebrow" and "bhauhein". "Cleanup" and "safai" mean the same thing at many salons. "Root canal" is often typed as "RCT". The more of your customers' real words you add, the fewer clarifying questions the AI needs to ask.

You do not need a developer. The three-step setup looks like this for an Indian business:

1. Connect your number. With WhatsApp Coexistence you keep your existing WhatsApp Business app and the same number. Your team still sees every chat on the phone.
2. Add services, hours and FAQs. Put in prices in rupees, durations, staff names, OPD timings or batch times, parking, GST on invoices and your cancellation rule. Add the Hindi and Hinglish words your customers use.
3. Connect Google Calendar. The AI books into real availability, so two customers never get the same slot. See appointment booking for the details.
4. Test with your own history. Copy 20 real messages from last month, including two voice notes and one angry customer. Send them to your number from a friend's phone and check each reply.
5. Go live after hours first. Many owners start by letting the AI handle nights and Sundays, then widen it. Our after-hours replies page explains that rollout.

NextReply plans are priced in US dollars, the same in every country: Starter $29, Pro $59 and Business $99 a month, for up to 500, 1,500 or unlimited bookings, and every plan starts with a 7-day free trial. At checkout, your card or bank may show an estimate in rupees, and local taxes may apply. Full details are on the pricing page.

Meta charges separately for some WhatsApp messages. Since July 2025 it bills per delivered template message (marketing, utility and authentication) instead of per conversation. Free-form replies inside the 24-hour customer service window, which is how most booking chats run, have not carried a template fee. Meta has also moved Indian businesses to rupee billing during 2026, and it has announced further pricing updates, so check Meta's pricing page for current rates before you budget for template campaigns.

Policy matters too. Since 15 January 2026, Meta's WhatsApp Business Solution Terms bar general-purpose AI chatbots on the platform. NextReply is a business assistant: it answers questions about your services, books appointments and hands over to your staff. That is the use Meta's terms allow.