The mango saplings turned black, so a nursery owner sent photographs to a chatbot

Shiv Kumar Yadav watched dark patches climb the stems of his young mango trees in Varanasi district, with white fungal growth spreading across the leaves. He photographed the damage and sent it to FarmerChat, a free chatbot built by the nonprofit Digital Green. Within days the infection had stopped moving.

So he took photographs and described what he saw, and the app answered with a diagnosis — a fungal disease — and a treatment plan to follow step by step. "Within a few days, the spread of the disease was under control," Yadav said. "It helped me save a large number of saplings and reduced my losses."

Uttar Pradesh grows roughly a quarter of India's mangoes, and the Langra variety that made the region's reputation originated near Varanasi, staying stubbornly green-skinned even when ripe. Anthracnose and powdery mildew move fastest through humid orchards, which is precisely when a grower most needs someone who has seen it before.

That someone is largely absent. India's public agricultural extension service has about one officer for every 1,162 farm holdings, against a recommended ratio of one to 750 on irrigated land, and more than half of the centrally sanctioned extension posts sit empty. By most estimates the service reaches under seven percent of the country's farmers. FarmerChat was built into that gap.

The organisation behind it began in 2006 as a research project inside Microsoft Research India, and its co-founder, Rikin Gandhi, spent close to six months in villages in Karnataka testing ways to teach farming before landing on the one Digital Green still uses: short videos filmed with local farmers rather than visiting experts. Some 8,000 of those videos exist now, in more than fifty languages. The chatbot, released in October 2024 and built with the San Francisco company Gooey.AI, extends the same instinct — the answer should arrive in the language the question was asked in. Farmers type, speak a voice note, or send a photograph; speech recognition comes from Bhashini, the Indian government's language platform.

Close to a million farmers now use it. In Mirzapur, bitter gourd grower Jyoti Devi raised her weekly harvest to between 20 and 30 kilograms. In Bihar, potato farmer Anu Kumari read the rainfall forecast and held off sowing until the storm had passed.

Digital Green has published 1.4 million of these farmer questions and answers as an open dataset, along with a benchmark for speech recognition in languages that rarely receive any. The cost of persuading one farmer to adopt a new practice, the organisation reports, has fallen from thirty-five dollars to under one.