The World Bank has long recognized the critical importance of agricultural extension services – ranging from training and data to technology transfer – which make up the second-largest share of its agriculture portfolio. Yet farmers have often been slow to adopt the very methods and tools these services are designed to deliver—limiting their own productivity and the sector’s potential to create jobs.
That’s in large part because they depend on limited numbers of extension agents: the field advisors responsible for providing them with data, training and advice. Most countries have just one extension agent for every 1,000 to 2,000 farmers.
Digital agriculture has long held the promise of filling this gap through bundled services that integrate weather forecasts, market prices, mobile payments, and agronomic advice. Still, in spite of the $1.5 billion invested in mobile phones and other technologies over the past decade, the uptake remains low. In many Sub-Saharan African countries, for example, fewer than 10 percent of all farmers use digital agricultural technology. Some solutions have scaled, such as a digital agricultural advisory service launched by the Government of Odisha that has successfully reached 7 million farmers. Sustained engagement, however, is still hampered by gaps in digital and language literacy, relevance of content, cost, and trust issues.
The AI advisory revolution
Generative AI is poised to overcome these persistent barriers. And indeed, the evidence from early projects is compelling:
Digital Green, a technology non-profit, has created an AI-powered chatbot that has reached 250,000 farmers and extension workers across five countries, supporting 40 crops in six languages. Innovative Solutions for Decision Agriculture (iSDA) has produced a similar tool called Virtual Agronomist serving over 200,000 plots in seven African countries, boosting profits by up to 4.7 times through personalized agronomic advice delivered via WhatsApp.
Kisan e-Mitra AI Chatbot, a voice-based, AI-powered chatbot launched by the Government of India, has resolved over 8.2 million queries from farmers in multiple local languages regarding direct benefit transfer, a government program that sends subsidies and welfare benefits directly into citizens’ bank accounts.
What makes these AI advisory systems uniquely effective?
Generative AI directly addresses challenges of digital literacy, language limitations, and relevance of information by offering farmers voice-based interfaces in the languages they use every day. These systems help them weigh different options, offer visual recognition to diagnose crop diseases, and provide highly relevant personalized advice. Hybrid tools that blend AI with human advice not only boosts impact but slash costs, bringing extension down from around $5 per farmer each year to under $1.
This dramatic drop is driven by a decline in AI costs – by some estimates down 240-fold in just 18 months. Today, delivering services costs about $1.50 per farmer per year, mainly for indirect expenses such as registration and adoption.
Figure 1: Key Generative AI capabilities and their applications. Source: authors.
This efficiency gain is enabling transformative scale. For example, Kenya’s digital climate advisories now serve 4.8 million farmers – 62 percent of the total – with coverage expected to rise to 91 percent, or 7.4 million, by 2030.
Beyond the advice they provide, such systems generate valuable market intelligence by recording farmers’ questions on planting, pests, and prices, creating demand-driven insights that also benefit providers, suppliers, and researchers.
Risky hallucinations
While the promise of Generative AI is immense, the risks for smallholder farmers are equally real. GenAI models are probabilistic, meaning they can confidently provide incorrect answers, commonly referred to as “hallucinations”. These are particularly dangerous in agricultural contexts where a flawed recommendation about pesticide use, fertilizer dosage, or planting time can result in yield loss, legal violations, or even health risks, not to mention loss of trust in the technology.
These risks are amplified when AI is used without any human oversight. To mitigate this, most current deployments include a “human-in-the-loop” model to ensure extension agents or agronomic experts examine results before they reach farmers.
Managing these risks requires multiple layers of scrutiny, including robust feedback mechanisms such as in-app ratings systems, conducting rigorous “red teaming” to reveal vulnerabilities, and clear governance frameworks executed through legal, marketing, and technology teams. Finally, benchmarking agricultural AI language tools is essential to make rapid, objective comparisons between models, chatbots, and voicebots.
Foundation for scale and path forward
Scaling AI advisory services requires digital infrastructure, data services, regionally specific AI tools and widely accessible channels – and above all, buy-in among farmers. As more data is gathered on farmers and their assets, the costs of building and deploying new models rise—but so do the benefits.
Figure 2: The Data-to-Value Curve for AI Agricultural Advisory Services. Source: authors.
Further, successful deployment of AI agricultural advisory at scale will require collaboration across sectors, linking global AI providers, agricultural research organizations, AgTech innovators and public and private institutions.
Generative AI offers a powerful way to scale decades of extension work, delivering personalized, timely advice to smallholder farmers regardless of literacy, language, or location. The World Bank and Gates Foundation are joining efforts to support extension services, leveraging their respective strengths.
This November, building on our long-standing collaboration, we will launch a comprehensive report with public and private partners exploring AI’s potential across the agri-food system. We invite governments, development partners, researchers, and tech providers to join us in transforming agricultural extension into a driver of rural prosperity.
Source : World Bank
