Farmers Using AI Advice With a Helping Hand Harvested More Than Twice as Much
Robauto Team · Published · Last updated
New studies from Kenya, Uganda, the Philippines and India show AI farm advice changes how farmers solve problems, and harvests jump up to 2.29x when trained people help them act on it.
The farmers who got AI advice plus a trained person to help them use it harvested more than twice as much as their neighbors. That's the headline from a new field study in Kenya and Uganda, published by the agronomy nonprofit iSDA on October 21, 2026. A separate randomized trial with almost 2,000 rice farmers in the Philippines and India, posted in early October, points the same way. AI advice changes what farmers do when a problem shows up, and it pays off most when people stay in the loop.
This is a good-news story with a useful twist. The AI isn't replacing the extension worker or the experienced neighbor. It gives them better, plot-specific answers, and they help farmers act on those answers on time.
What happened in Kenya and Uganda
iSDA's Virtual Agronomist is an AI advisory tool that gives each farmer recommendations for their own field, using local soil, weather, and crop information. During the 2025 short rains season, iSDA followed 605 farmers growing maize, sorghum, and sunflower in Meru and Tharaka Nithi counties in Kenya and in Bulambuli and Sironko districts in Uganda (iSDA).
Farmers fell into three groups: 214 who didn't use the tool, 197 who used it with support from a lead farmer (a trained farmer from their own community), and 194 who used it with intensive help from iSDA technicians. Instead of just asking farmers how much they harvested, field staff did crop cuts. They harvested a 10 m by 10 m square in the middle of each plot, dried and weighed it, and scaled it to a yield per hectare.
The harvest numbers
After adjusting for crop, country, and soil nutrient levels, farmers using the tool with a lead farmer's help harvested 1.56 times as much as non-users, or 56% more. Farmers with technician support harvested 2.29 times as much, about 1.5 times the lead-farmer group. The gains held in all four crop-and-country combinations, and farmers' own reported harvests told a similar story (agriRxiv preprint).
It also wasn't an easy season to prove anything in. iSDA notes that Meru had received less than a third of its usual rainfall by November 28, 2025, and that eastern Uganda saw late rains followed by flooding. The AI-supported farmers still came out ahead everywhere the study looked.
| Group (Kenya and Uganda, 2025 short rains) | Farmers | Adjusted yield vs. non-users | Nitrogen used (kg/ha) | Source |
|---|---|---|---|---|
| No AI advice | 214 | 1.00x | about 28 | iSDA |
| AI advice + lead farmer | 197 | 1.56x | about 47 | iSDA |
| AI advice + technician | 194 | 2.29x | about 54 | iSDA |
Was it just more fertilizer?
That's the fair question, and iSDA raises it itself. Supported farmers did use more nitrogen. But the team's early analysis suggests nitrogen alone didn't explain the gains, especially for technician-supported farmers. It points to better management across the season: better weeding, better crop protection, and advice applied accurately and on time.
Did it put more money in farmers' pockets?
Where support was strongest, yes. iSDA reports gross margins, meaning crop revenue minus fertilizer cost, which isn't the same as profit. Sunflower growers in Uganda went from about $233 per hectare without the tool to $439 with lead-farmer support and $628 with technician support. Kenya maize margins rose from $158 to $436 per hectare with technician support. In Kenya sorghum and Uganda maize, margins were higher on average but the gap was small enough that it could have been chance, and iSDA says so plainly.
The randomized trial in Asia: better decisions first
The Kenya and Uganda study is observational, so it's strong evidence but not a controlled experiment. That makes a second study a good companion. Economists Ali Aouad, Alp Sungu, Niyazi Taneri, and Canberk Ucel ran a pre-registered randomized controlled trial with 1,935 smallholder rice farmers in the Philippines and India over one production cycle (SSRN).
Farmers in the treatment group got an AI advisor on WhatsApp or Messenger, built with local experts. The control group got a similar chatbot that didn't talk about farming, so any difference came from the advice itself rather than the novelty of chatting with an AI. Deliberate practice changes rose by about a fifth (0.10 standard deviations), the quality of farming operations improved (0.08 SD), and farmers shifted toward integrated pest management (0.11 SD). Crop protection and problem response were also the most common topics in the chat logs.
Farmers with the advisor were also 7 percentage points more likely to report avoiding crop-residue burning, according to a write-up of the study and the authors' posts (Green Thumb Hobbies). The trial found no significant change in yields or costs in that single season, and the authors' takeaway is a practical one: AI advice helps most with decisions farmers can act on using what they already have, especially when a problem is urgent. Bigger changes, like new equipment, need more than information.
| Study | Who | Design | Key positive result | Source |
|---|---|---|---|---|
| Virtual Agronomist field study | 605 farmers, Kenya and Uganda | Crop-cut yields, adjusted comparison | 1.56x to 2.29x harvest with human support | iSDA |
| AI advisory RCT | 1,935 rice farmers, Philippines and India | Pre-registered randomized trial, one season | About a fifth more deliberate practice changes, better pest management | SSRN |
Why the human part matters so much
Put the two studies together and a clear pattern shows up. The AI gives a fast, specific answer. A person helps turn that answer into the right action on the right day. In iSDA's results, more hands-on support brought better outcomes at every step.
That matters because there aren't enough advisors to go around. iSDA cites Kenya's extension policy target of one officer per 600 farmers by 2029, and Ugandan figures of one officer per 1,800 farmers against a recommended one per 500. Its stated goal is for the tool to support extension workers, not replace them, by giving each farmer advice for their own plot and giving the worker a nutrient plan they can check, adapt, and explain on the spot.
The Robauto takeaway: be the answer people can act on
Farmers in the rice trial reached for the AI advisor when trouble showed up, not on a schedule. Customers are starting to behave the same way with AI assistants. They ask when they have a problem, and the assistant pulls together whatever clear, trustworthy information it can find. Your business gets the same opportunity the agronomists had: be the source with the specific, usable answer.
- Answer the urgent questions plainly. Shipping, returns, refunds, hours, and how to reach a real person should be easy to find in plain text.
- Be specific to the situation. The best advice in these studies was plot-specific. Product pages that state sizes, compatibility, and limits help an AI give a precise answer.
- Keep the human path visible. The biggest wins came with human support. Make your support contact and escalation route obvious so an assistant can hand people off.
- Make it machine-readable. Structured data, a clean llms.txt, and agent-friendly endpoints help AI tools read you correctly the first time.
{
"topic": "AI farm advice with human support",
"studies": [
{"name": "iSDA Virtual Agronomist", "farmers": 605, "countries": ["Kenya", "Uganda"], "season": "2025 short rains", "yield_vs_nonusers": {"lead_farmer_support": 1.56, "technician_support": 2.29}, "source": "https://www.isda-africa.com/posts/how-much-human-support-does-ai-advice-need/"},
{"name": "AI advisory RCT", "farmers": 1935, "countries": ["Philippines", "India"], "result": "practice changes +0.10 SD, integrated pest management +0.11 SD, no significant yield change in one season", "source": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7572761"}
],
"takeaway": "AI advice works best when it is specific and a person helps people act on it",
"brand_action": "Publish clear, specific, machine-readable answers and a visible human support path",
"cta": "https://robauto.ai/scan"
}Want to know how easily AI assistants can find and use your answers? Scan your site free at robauto.ai/scan.
Sources
- iSDA: How Much Human Support Does AI Advice Need? (Oct 21, 2026)
- agriRxiv: Yield and economic impacts of an AI-enabled Virtual Agronomist advisory system
- SSRN: AI Advisory for Smallholder Farmers: A Randomized Field Experiment in the Philippines and India
- Green Thumb Hobbies: The First Randomized Test of an AI Farm Advisor
Where can an agent get this data live?
Robauto MCP — Agent Intelligence & Brand Discovery Infrastructure. A public, stateless Streamable HTTP endpoint at https://mcp.robauto.ai/mcp, callable with no key. Setup and the full tool list are in the Robauto MCP directory.