Generative AI and Agricultural Productivity: Evidence from Peanut Farmers in China

In many parts of the world, farming is a constant negotiation with uncertainty. A farmer must decide when to plant, how much water to give, and what to do when a leaf turns yellow, often without access to an expert standing right beside them. For decades, the solution has been to rely on local experience, neighbors, or government extension workers who travel from village to village. These channels provide vital knowledge, but they are often slow, expensive to scale, or too generic to fit the specific conditions of a single field. In recent years, mobile phones and the internet have begun to bridge this gap, offering instant access to weather reports and market prices. Yet, a new question has emerged: can artificial intelligence do more than just deliver static facts? Can a computer program actually converse with a farmer, understand the unique problem of a specific crop, and offer advice that feels like it comes from a seasoned expert? This is the promise of generative artificial intelligence, a technology capable of creating new, tailored responses rather than simply retrieving pre-written answers. If this technology can truly lower the barrier to expert advice, it could transform how small-scale farmers manage their land, potentially turning uncertainty into opportunity.

A team of researchers from Nanjing Agricultural University set out to test this promise in the real world, focusing on peanut farmers across six provinces in China. They gathered data from 977 households in 2025, a year that marked a significant moment for this technology in agriculture. The researchers wanted to know if using generative AI tools—applications that can chat with users and generate advice—actually changed how much land farmers planted, how much they harvested, and how efficiently they used their resources. To get a clear answer, they had to be careful. Farmers who choose to use new technology are often different from those who do not; they might be more educated, have better land, or simply be more willing to take risks. The researchers used a sophisticated statistical method to separate the effect of the technology itself from these other factors, ensuring that any improvements they found were truly linked to the use of the AI.

The results were encouraging, but specific. Farmers who used generative AI for their peanut crops planted significantly more land and harvested a higher yield per unit of area compared to those who did not. On average, users saw their peanut output rise, and they felt confident enough to expand their planting area. However, the technology did not seem to change the fundamental efficiency of the farm in a broader sense. When the researchers calculated total factor productivity—a measure of how well a farm converts all its inputs, like labor and fertilizer, into output—they found no significant difference between users and non-users. This suggests that while the AI helped farmers make better immediate decisions, it did not yet revolutionize the entire production system or change the underlying technology of how the peanuts were grown.

The study dug deeper to understand why these gains happened. The primary benefit appeared to come from better management of pests and diseases. Farmers using the AI could identify problems in their crops much faster and more accurately. Instead of waiting for a neighbor to notice a blight or guessing at the cause of wilting leaves, they could consult the AI, which helped them diagnose the issue and apply the right treatment sooner. This speed and precision reduced the amount of crop lost to disease. Additionally, the technology helped farmers use their labor more efficiently. By handling the mental work of searching for information and planning tasks, the AI freed up time for the farmers to focus on the physical work in the field. The study found that these improvements were concentrated in the area of pest control; the AI did not significantly change how farmers adopted new seed varieties, new fertilizers, or new harvesting machinery. The technology was proving most useful for the daily, recurring decisions that require quick, localized answers.

Not every farmer benefited in the same way. The study revealed that the advantages of generative AI were not automatic; they depended heavily on the farmer's own skills and circumstances. The yield gains were most pronounced among farmers who already had a higher level of digital literacy, meaning they were comfortable using smartphones and navigating digital tools. It also helped those who had previously attended technical training or exchanged knowledge with other farmers. These groups were better equipped to ask the right questions and interpret the AI's advice into action. Furthermore, the benefits were clearer for farmers with larger, less fragmented plots of land. For farmers with many small, scattered pieces of land, the technology did not produce the same boost in yield, suggesting that the scale of the farm matters when trying to translate better information into better results.

Beyond the fields, the researchers looked at the economic impact on the households. The increased yield and larger planting areas translated directly into higher income for the families using the technology. Interestingly, the AI did not seem to change how much farmers spent on fertilizers or pesticides, nor did it alter the price at which they sold their peanuts. The value came purely from growing more and losing less to disease. Looking toward the future, the farmers who used the AI expressed a strong desire to keep using it, particularly for tasks that happen often, like choosing seeds, managing water, and checking pest levels. They were less enthusiastic about using it for big, one-time decisions like buying heavy machinery or processing the harvest after the crop is gathered. This suggests that farmers see the technology as a powerful assistant for the daily rhythm of farming, rather than a replacement for major capital investments.

The findings offer a clear picture of where this technology stands today. Generative artificial intelligence is not a magic wand that instantly solves all agricultural problems or completely overhauls how food is produced. Instead, it acts as a highly effective decision-support tool that fills a critical gap: the need for timely, localized advice. By helping farmers spot problems early and manage their labor better, it can increase production and income, but only if the farmers have the skills to use it and the land structure to support it. The study suggests that the future of agricultural technology lies not just in building smarter algorithms, but in pairing them with human education and better infrastructure. If farmers can be taught to work effectively with these new digital tools, and if the land they farm is organized in a way that allows for efficient management, then generative AI could become a standard part of the toolkit for small-scale farmers, turning information from a scarce resource into a reliable ally.