Testing AI advisory services – insights from FarmerChat in India and Kenya

The International Food Policy Research Institute (IFPRI) and Digital Green are deepening their long-standing partnership to explore how artificial intelligence (AI) can strengthen agricultural extension and advisory services for smallholder farmers. The collaboration will now focus on user testing of Digital Green’s FarmerChat application as part of IFPRI’s Generative AI for Agriculture (GAIA) initiative.

FarmerChat is an AI-powered assistant developed by Digital Green that provides farmers with free, localised and climate-smart agricultural advice in their own languages, using text, video, voice and images. The tool is designed to expand farmers’ access to timely and trusted information on crop management, markets and climate resilience.

User testing of FarmerChat

The Phase II collaboration (2025–2027) focuses on structured, human-centred user testing of FarmerChat in India and Kenya, with particular attention to women, youth and underserved farmers.

The user testing includes several components:

• Real-world usability testing: Farmers and extension agents use FarmerChat in live field conditions, while researchers observe how they navigate the tool, what questions they ask and where confusion or friction occurs.
   
• Participatory feedback sessions: Facilitated co-design workshops and focus groups gather farmer input on clarity, trust, tone, cultural relevance and usability – with a focus on women and youth inclusion.
   
• Structured surveys and behavioural analytics: Survey instruments measure perceived trust, usefulness and usability, while backend usage data (query types, repeat engagement, follow-up questions) helps assess how advice translates into action.
   
• Agronomic and inclusivity review: Responses are evaluated for technical accuracy, contextual relevance and whether recommendations reflect the constraints smallholder farmers face (e.g. affordability, local input availability, labour realities).

Initial findings on FarmerChat usability

Preliminary observations show that farmers are most engaged when the advice is locally relevant and specific (e.g. taking into account the crop growth stage, local weather conditions or generally available inputs). Trust increases significantly when the advice aligns with farmers’ practical experiences and existing advisory messages.

Voice-based interaction is particularly important for farmers with lower levels of education. Female users benefit from simplified onboarding and an introduction to the tool supported by peers. 

The official results of this structured testing phase will be published once the analysis has been completed.

Broader user feedback on FarmerChat

Beyond the IFPRI collaboration, FarmerChat is used by over a million farmers across a wide range of countries, Eric Firnhaber, director of global communications at Digital Green, says. Ongoing internal monitoring and independent research provide insight into how the tool is being used.

The majority of queries fall into the following categories: 

• pest and disease diagnosis (often with photos)
• fertiliser and nutrient management
• crop stage-specific agronomic practices
• weather-related decision support
• livestock health (in mixed farming systems)
• market price and timing questions

Many farmers use FarmerChat in a diagnostic mode - describing symptoms or uploading images - with subsequent follow-up questions on treatment or prevention.

Limiting factors and challenges

As with any AI-powered advisory tool, there are several challenges. For example, connectivity may be limited. In low-bandwidth environments, uploading images or voice responses may be slower. Similarly, there are limitations regarding inputs. Farmers sometimes lack precisely the recommended inputs (e.g. a specific brand of fertiliser), meaning that the advice needs to be tailored more flexibly.

Another challenge is building trust. Some users are initially sceptical about the reliability of AI-generated advice. Repeated positive experiences and coordination with advisory services strengthen trust over time.

Digital literacy gaps can pose a challenge for certain target groups. Women and those new to smartphones may need support with familiarisation before they can use the tool independently.

Participatory user testing and collaboration with research partners such as IFPRI are crucial, particularly for addressing these challenges.

Ines Lechner, editor Rural 21

Visit the Digital Green website