From voice-based farming advice to daily weather updates, discover how ATREE piloted GaonSahayog and five lessons other NGOs can learn from its journey.

This case study, co-authored with ATREE, explores what the team built, how farmers responded and what other NGOs can learn from the experience.

A farming question, a voice message and a new possibility

In a village in Soreng, West Sikkim, three women used a WhatsApp chatbot to ask about a disease affecting their large cardamom crops.

Instead of typing, they simply asked their questions through voice messages.

The chatbot suggested removing and destroying the diseased plant parts, something they had not tried before. Excited by the suggestion, the women said they would try it when they returned home.

This simple interaction raised an interesting possibility: What if farmers could access relevant agricultural information in their own language, through a platform they already use?

That is the idea behind GaonSahayog, an AI-powered WhatsApp chatbot developed by the Ashoka Trust for Research in Ecology and the Environment (ATREE) for farming communities in Sikkim and Darjeeling.

GaonSahayog was built during the six-month Glific AI Chatbot Accelerator 2026, where 34 NGOs developed working AI chatbot pilots. ATREE received an honourable mention at the September close-out event.

Making agricultural knowledge more accessible

For farming communities in the Sikkim–Darjeeling landscape, access to locally relevant information can support decisions about crop diseases, soil preparation, weather and other livelihood challenges.

ATREE wanted to make this knowledge easier to access.

Using Glific, the team developed GaonSahayog, a multilingual WhatsApp chatbot that provides concise answers through text and voice.

Its knowledge base draws on scientific literature, government information, institutional expertise and questions raised by community members. It includes crop-specific information on orange, avocado, large cardamom and red cherry pepper, alongside resilient agriculture, market linkages and human–wildlife conflict.

Inputs from Sikkim University, the Spices Board of India and the Government of Sikkim helped strengthen the content.

Farmers can also share their location through WhatsApp to receive live weather information, supporting everyday farming decisions.

From building to testing: What farmers taught ATREE

ATREE conducted two rounds of community field testing in Sikkim and Darjeeling to understand whether people would use GaonSahayog, how they preferred to interact and where the chatbot needed improvement.

The first pilot focused on onboarding, multilingual interaction and access to agricultural information. The second tested an updated knowledge base, voice interactions, follow-up questions and live weather.The feedback revealed something important: most interactions were in Nepali, through voice messages.

Farmers found the chatbot easy to use and its responses relevant. Their questions also exposed gaps in the knowledge base.

For example, repeated questions about crop pests and weather helped ATREE identify where more detailed information was needed. Users also requested an image-query feature, which is planned for future development.

The team used this feedback to refine its FAQs, improve Nepali translations and strengthen its response testing.

Early results: Promising engagement, with room to improve

The pilots offered encouraging early signals.

• 90 contacts reached, with 55 actively engaging (approximately 61%)
• 70% of participants in Sittong provided very positive feedback
• Repeat interactions continued beyond the pilot period

AI evaluations also showed strong knowledge-base adherence (4.97/5) and prompt adherence (4.64/5). Ground-truth adherence was lower (3.86/5), highlighting the need to improve response accuracy.

Language presented another challenge. Some Nepali responses mixed in English words, and the initial Nepali voice output was unreliable. ATREE switched to the Gemini 2.5 Flash Preview text-to-speech model to improve the voice experience.

One particularly encouraging moment came from Dishan Gurung in Sittong, Darjeeling.

After one conversation, he wished the chatbot good night. He subsequently began using GaonSahayog each morning to check the weather and plan his day.

For ATREE, these interactions helped address an early uncertainty: would farming communities feel comfortable using an AI chatbot?

Seeing people ask questions, return for information and offer feedback motivated the team to keep improving it.

While longer-term adoption and agricultural impact remain to be assessed, the pilots demonstrated promising signs of usefulness and engagement.

Five lessons for NGOs exploring AI chatbots

ATREE’s experience offers practical insights for organisations considering chatbots for their own programmes.

1. Design around how people communicate

Most farmers preferred Nepali voice interactions. ATREE improved its voice capability rather than expecting users to rely on text.

The lesson: Start with your community’s language, accessibility needs and communication preferences, not just the technology available.

2. Good content matters as much as good AI

Agricultural questions are often highly specific, and generic answers may not be appropriate. ATREE invested in curating content, improving FAQs and evaluating responses against Golden Questions (test questions with known good answers).

The lesson: Reliable, context-specific information and response testing are essential to building a useful chatbot.

3. Treat pilots as opportunities to learn

Two rounds of testing helped ATREE identify gaps in agricultural content, translation and response quality.

The lesson: Test with real users, understand what isn’t working and refine the chatbot before expanding its reach.

4. Give people a reason to return

For at least one farmer, checking the weather through GaonSahayog became part of a morning routine.

The lesson: Think beyond initial adoption. Identify recurring needs that make your chatbot useful in everyday life.

5. Plan for adoption, not just launch

ATREE plans community outreach and training workshops to build awareness and help farmers use GaonSahayog.

The lesson: A functional chatbot is only the beginning. Awareness, onboarding and continued engagement are equally important.

What’s next for GaonSahayog?

ATREE plans to publicly launch GaonSahayog by the end of November 2026, following further improvements to its knowledge base and user experience. Future development will include image-based queries, information on government schemes and expanded content on human–wildlife conflict. The team will continue learning from real conversations and improving the chatbot based on community feedback.

What could a chatbot make possible for your programme?

ATREE started with a practical challenge: making agricultural knowledge more accessible to farming communities. Through GaonSahayog, the team built a working solution, tested it with users, identified limitations and improved it based on evidence.

For other NGOs, the challenge might be different, from answering programme-related questions to supporting field workers or making information more accessible.

The starting point remains the same: identify a real need, design around your users, test, learn and improve.

GaonSahayog was developed through the Glific AI Chatbot Accelerator, which supported NGOs in building and testing AI-powered WhatsApp chatbots for real-world programmes.

Read the AI Accelerator Impact Report here.

Curious about how a WhatsApp chatbot could support your programme?

Schedule a call with us and discover how your organisation can build, test and improve a chatbot designed around your community’s needs.