Can an AI chatbot give reliable crop advice?
AI chatbots like ChatGPT are gaining traction across industries – but can they also support farmers with crop advice? Are such tools already being used in agriculture? How accurate and in-depth are their responses, and could they eventually replace agronomists? Future Farming investigated the current state of AI-based advisory tools and explored existing sector-specific models and chatbots.
Sometimes it seems as if you are no longer part of the game unless you are using or applying AI (artificial intelligence). That impression quickly arises from many email newsletters and promotional talks about new machines and equipment. And yes, there’s a good chance that the smartphone you’re reading this article on already contains various AI features—such as analyzing photos to identify which weed plant is developing in a field.
Previously, special apps were needed for that, but thanks to Apple Visual Lookup and Google Lens, any (modern) smartphone can do it. And thanks to AI chatbots—ChatGPT being the first and probably the best known, alongside many others such as DeepSeek AI, Google Gemini, Microsoft Copilot, and Le Chat by Mistral AI—virtual assistants are becoming available to everyone. An assistant that can also provide crop advice. Is that useful? At first glance, all are free, but there are certainly 2 (major) caveats.
Chatbots are a form of generative AI
Just like AI itself, chatbots have existed for much longer. ChatGPT and other AI chatbots use so-called large language models (LLMs). These are a type of AI capable of generating and understanding human language. They are trained on data and text. Meta, the parent company of Facebook, Instagram, and WhatsApp, recently made headlines because it intends to use unprotected user data to train its AI models.
This is something Casper de Jong, lecturer and researcher at Saxion University of Applied Sciences, recently warned about during an inspiration session on AI in the manufacturing industry. “If you use a free version of an AI chatbot, you automatically agree that it can use your input to train itself. Be aware of that when sharing a business plan for editing, translating, or summarizing. This doesn’t apply to paid versions.”
Trim Bresilla, robotics engineer and researcher with the Vision+Robotics program at Wageningen University & Research (WUR), explains on https://visionrobotics.nl that AI chatbots are a form of generative AI (GenAI). “With a few words or a question as input, ChatGPT uses AI to generate text. More generally, a GenAI model uses data, numbers, or words to generate an output. It’s important to realize that the output is always slightly lower in quality than the original input, since generative models cannot invent data themselves. That’s why AI-generated data must always be combined with real data. Still, models are improving rapidly, and especially since last year, the quality and reliability of the output have improved significantly.”
A good example is the question ChatGPT would have answered a year ago: “If it takes 25 minutes to dry 5 T-shirts in the sun, how long does it take to dry 6 T-shirts?” Back then, the answer was “30 minutes.” Now the answer is “probably also about 25 minutes” (“assuming they can all hang properly under the same conditions”).
Are chatbots used by crop advisors?
As a farmer or consumer, you may already have been using AI and chatbots unconsciously for quite some time—on your smartphone, while online shopping, or for translation help. Since its launch in 2022, AI chatbot ChatGPT has sparked tremendous curiosity, leading to a flurry of discussions and articles. But are AI chatbots already being used professionally, and if so, for what?
Initially, Boerderij intended to test ChatGPT for crop advice together with several arable farmers and their crop advisors, but that turned out to be a step too far—or too early. Most interviewed advisors reported their farmers were not yet enthusiastic. And where interest did exist, fieldwork took precedence, and the reaction remained limited to a phone call.
Ardon Verschoor, arable advisor at Van Iperen: “Until I was asked, I hadn’t used AI chatbots for technical content. Mostly due to unfamiliarity and questions about the reliability of the information. I’ve since done some experimenting. But when I asked ChatGPT about a crop protection product for onions, it recommended something unsuitable. I do see potential for automating office work and assisting with subsidy applications.”
Independent advisor Gerard Meuffels has not seen AI chatbot use in practice and notes that many farmers don’t know what to expect. “The only things I’ve seen are weed recognition apps like Bayer MagicScout and Pl@ntNet.”
Driven mainly by curiosity
Luc Remijn, arable advisor at Delphy, an international agricultural consultancy, has tried several prompts out of curiosity and “because everyone is talking about it.” “For example, ‘bean fly onions’ for a description of the insect and developmental stages of wireworms—info also found online at Best4Soil. Due to new internal policies, we’re no longer allowed to use ChatGPT, so I am switching to Microsoft Copilot.”
Gert Sterenborg from Maatschap Sterenborg in the Dutch village of Onstwedde experimented a bit but finds the answers too generic and vague. “The (agricultural) knowledge seems limited and insufficiently tailored to soil type and conditions. I do think that could improve by guiding the prompt, for example, with ‘You are now my crop advisor.’ What I do already use it for is to formulate my thoughts and as a text generator for subsidy applications.”
The (agricultural) knowledge seems limited and insufficiently tailored to soil type and conditions
Ted Vaalburg of Vaalburg VOF in the Dutch village of Zuidschermer uses it as a search engine and for generating text. “I speak my query out loud. The result is just as reliable as Google’s, but I do not have to type. It is more convenient.” Gerben van Dueren den Hollander from Geduho in Oud-Beijerland also uses it for convenience. “I have used it, for example, to calculate the number of seed potatoes needed per hectare. When you change one variable, you immediately get a new result.”
Already big in CEA
Subsectors where AI already plays a major role include greenhouse horticulture and vertical farming, often referred to as Controlled Environment Agriculture (CEA). As part of the public-private partnership AGROS, Wageningen University & Research and partners are working on a fully autonomous greenhouse. The project’s first phase shows AI can match crop experts, grow strong crops, and deliver profitable yields.
WUR researcher Guido Jansen says: “Autonomous growing only works with known situations. What happens when something unexpected occurs? That’s when a grower is still needed to make the right decisions.” Colleague Rick van de Zedde: “Ideally, a grower could give the system a task in natural language, like ChatGPT. In other words, the grower wouldn’t need to program, just ask: ‘Over the past years in this compartment, is there a relationship between disease pressure and climate?’”
Validation remains necessary
Most examples show that validating the generated output remains (for now) necessary. Jeroen Rheinfeld, professor of agricultural law at the University of Groningen, recently proposed a mix of artificial and agricultural intelligence: “automated where possible, but always keep your head in the game.” Pieter de Wolf, researcher in open-field crops at WUR and project leader of the Farm of the Future in Lelystad, refers to enhancing human intelligence and emphasises that artificial intelligence cannot do without social intelligence.
“Automated where possible, but always keep your head in the game.”
Sector-specific AI chatbots and models may help ease skepticism and concerns about reliability. Several examples are listed in the box: “Number of specific agricultural applications is increasing.”
Join 17,000+ subscribers
Subscribe to our newsletter to stay updated about all the need-to-know content in the agricultural sector, two times a week.
