A University AI Called CropWizard Answers Farm and Garden Questions From Extension Guides, and It Shows You Where Each Answer Came From

Ask a general-purpose chatbot why your tomato leaves are curling and you will get a confident answer. What you usually will not get is a clear way to check where that answer came from, or whether it applies to a garden in the Sonoran Desert rather than one in Ohio.

A research team at the University of Illinois Urbana-Champaign has been building an AI assistant that tries to close that gap. It is called CropWizard, and instead of answering from whatever it absorbed during training, it looks up material in a large library of Cooperative Extension publications and agricultural research, then cites the documents it used so you can read them yourself.

CropWizard came out of AIFARMS, short for Artificial Intelligence for Future Agricultural Resilience, Management, and Sustainability, a national AI institute for agriculture housed in the university’s Center for Digital Agriculture. A profile of the institute published Oct. 29 by The Daily Illini, the University of Illinois student newspaper, described CropWizard as one of the most broadly accessible technologies the group has developed.

Built on Extension, not the open internet

According to AIFARMS executive director Jessica Wedow, CropWizard was built on Extension documents from around the United States combined with generative AI. Extension is the network of land-grant university experts who write the plain-language fact sheets on soil, pests, irrigation and planting dates that many gardeners already rely on.

The project’s documentation describes the scale. The help pages say CropWizard consults more than 400,000 selected online publications, including Extension resources from U.S. land-grant universities and a growing set of open-access research papers. The project’s own overview page gives a larger figure, more than 650,000 Extension publications, research papers and technical documents. The two pages do not explain the difference, which may reflect a collection that has grown over time.

The Extension portion spans more than 55 land-grant universities. North Carolina State University contributes the most, with 80,851 documents, followed by the University of Illinois with 20,606. The University of Arizona is on the list of largest contributors with 10,189 documents. That detail matters for desert gardeners, because Arizona Extension material is written for the conditions here: alkaline soil, extreme summer heat, low-desert planting calendars and irrigation schedules that look nothing like those in the Midwest.

The research collection includes open-access journals from Springer Nature and Wiley, such as Agronomy Journal, Crop Science and Horticulture Advances, along with a list of the licenses under which those papers are shared.

How it works

CropWizard uses a technique called retrieval-augmented generation. When you type a question, the system first searches its document library for relevant passages, then hands those passages to a large language model to compose an answer. Responses include numbered references, and a Sources button at the bottom of each answer opens a list that links to the original documents online.

The documentation says users can also see some of the work behind an answer, including how many relevant document chunks were retrieved and whether any specialized tools were called. Those tools include a pest detection model that analyzes uploaded photos and a seasonal characterization tool for field trial data. The project’s overview page says other tools can pull weather information for a location, analyze farm shapefiles, read spreadsheets and search the web.

There is also a setting that will interest gardeners outside the Corn Belt. In the chat settings, the document library is organized into groups by university for Extension material and by publisher for research papers. All documents are searched by default, but users can switch groups on and off to narrow the search. In principle, that means an Arizona gardener could limit answers to University of Arizona Extension material.

You can ask questions in plain text or upload photos and ask about them. The help pages suggest that the system can serve as a virtual agronomist for farming advice, a research assistant for finding the most relevant publications on a topic, or simply a search engine for basic agricultural information. Its example questions lean toward field crops, such as crop rotation in corn and nutrient needs for wheat, but the document library also covers gardening, lawn care, tree care, weed control, fruits and pest management.

Accurate is not the same as useful

Wedow told The Daily Illini that one of the biggest challenges was making sure the information was precise and relevant to the people using it.

“Is the information you get not only accurate, but timely and actionable?” she said. “The answer could be very accurate, but the feasibility of that is not necessarily something that a farmer could do, so that’s a component.”

That is a useful test for any gardener using an AI tool. An answer can be technically correct and still be wrong for your yard, your budget or your season. Being able to open the source document and check whether it was written for your region is one of the more practical safeguards available.

The CropWizard team has also turned that concern into a research program. Its project site lists two benchmarks, AgMMU and MIRAGE, both described as published in the NeurIPS Datasets and Benchmarks Track in 2026, that test how well vision-language AI models answer agricultural questions and handle expert-style conversations. The site says this work led in 2025 to the AI AgriBench Benchmarking Consortium, led by CropWizard principal investigator Vikram Adve and the Center for Digital Agriculture, with founding members including Bayer Crop Science, John Deere, Microsoft, the Extension Foundation and Kissan AI. The consortium’s stated goal is to give farmers, policymakers and the public trustworthy ways to evaluate AI agricultural advice tools.

Who pays for it, and what comes next

CropWizard was launched as a research project of AIFARMS, which is funded by the U.S. Department of Agriculture’s National Institute of Food and Agriculture. The project also lists support from Intel, from Amazon Web Services through the Amazon-Illinois Center on AI for Conversational Experiences, and from the University of Illinois system. It runs on Illinois Chat, a university chatbot platform whose funders include the Center for AI Innovation at the National Center for Supercomputing Applications.

The software is open source. The project says the code and documentation are free for research, teaching and other noncommercial uses, while commercial use requires a paid license.

The Daily Illini reported that the National AI Institute core grant to AIFARMS ends this year. Wedow said the organization is applying for other federal grants, foundation support and university funding to continue some of its work. The article did not say what the funding change will mean for CropWizard specifically.

Trying it yourself

CropWizard’s chat page is at chat.illinois.edu/cropwizard, and its help documentation is on the Illinois Chat documentation site. A good first test is a question you already know the answer to, such as when to plant garlic in the low desert or how often to water a young citrus tree in November. Read the answer, then open the sources it cites and see whether they came from Arizona or from somewhere with a very different climate.

CropWizard is not a substitute for a local Master Gardener or a call to the county Extension office. But for anyone tired of AI answers that cannot show their work, it offers a different approach: an assistant built on the same fact sheets gardeners have trusted for decades, with every answer pointing back to the page it came from.