How Can Dairy Farmers Use Large Language Models (LLMs)?

Introduction

Since the breakthrough of generative artificial intelligence (AI) and large language models (also known as LLMs) a few years ago, it seems as though it’s all that people talk about. They’re showing up in phone apps, online search tools, email programs, and even recipe books. Whether we realize it or not, generative AI is already playing a role in our daily lives. But what exactly is “generative AI”? And more importantly, how can dairy producers use it in a practical way to help manage their farms, stay up to date on research, and make informed decisions?

Generative AI

Generative AI refers to computer programs that can create new content based on patterns learned from large amounts of data. LLMs are a specific type of generative AI that focuses on understanding and generating human language. They’ve been trained on text from books, websites, articles, and more, allowing them to answer questions, explain complex topics, and summarize information in plain language. Think of an LLM like a well-read assistant that can help you find, understand, and organize information, but one that you still need to guide carefully. 

Although many farmers may have already used an LLM to help draft an email, come up with meal ideas for the week, or plan a family trip, the same type of technology can be used for farm-related purposes. It can be used to summarize new dairy research or extension bulletins, draft farm protocols or Standard Operating Procedures (SOPs) based on University recommendations, translate technical research articles into summaries, and help compare multiple articles or viewpoints on a topic

Shorter and concise: There are many LLMs from different companies. Some are built into search engines, and others are standalone tools. Many offer free basic versions. You don’t need to choose one company or pay to benefit from them. The idea is simple: you type a question or prompt, and the model generates a clear, conversational answer.  

Financial viability drives nearly every management decision in dairy farms. LLMs can also help evaluate the economic side of new technologies by organizing cost information, estimating labor savings in dollar terms, calculating simple payback periods, or comparing financing options. For example, you might ask, ‘If I’m currently paying 3 full-time employees $18/hour for parlor work and this technology could reduce labor needs by 15%, what are my potential annual savings?’ However, remember that LLMs may not have current equipment prices or region-specific costs, so always verify numbers with vendors and your financial advisor before making investment decisions.

A farm example: Short on labor in the dairy

Let’s suppose you’re a dairy farm manager overseeing an operation with about 2,000 lactating cows in a freestall system. Your days are busy and your mornings start early. Over the past year, you’ve had trouble hiring and retaining milkers. The current crew is stretched thin and turnover is higher than you’d like. 

You’re not looking for the newest gadget just for the sake of having technology. You’re trying to answer a very practical question:

“What technologies could realistically help us reduce labor or improve workflow efficiency in the next 6–12 months, and which options are worth pricing out for a 2,000-cow herd like mine?”

This is the kind of question where large language model tools can help you quickly organize information and connect high-level research to your day-to-day decisions. 

Step 1: Use ExtBot to identify the options

You start with ExtBot (https://extension.org/tools/extbot/ ↗️), the Extension Foundation’s LLM tool that is tuned to Extension-style content and university-based resources. Instead of typing a vague query like “new tech for dairies,” you give ExtBot a focused, farm-specific prompt: 

“I manage a 2,000-cow freestall dairy in the Midwest. We’ve been short-staffed in the milking parlor and need to reduce labor pressure without hurting milk quality or cow comfort. 
Please outline realistic automation or digital tools that large herds use to reduce parlor labor or improve milking workflow, and list pros, cons, and key considerations for a herd like ours.” 

Because ExtBot has been trained on Extension materials, it responds in a structured, practical way. Instead of a long, unorganized list of links, it might group options into categories like: 

• Cow flow and sorting tools 
  • Examples: Automatic sort gates to pull cows for vet checks, hoof trimming, or special groups. 
  • Pros: Saves time chasing and identifying cows; reduces disruption in the parlor. 
  • Considerations: Barn layout changes, integration with existing herd software. 
• Computer vision and sensor systems 
  • Examples: Camera-based systems for lameness scoring, body condition scoring, or heat-stress monitoring. 
  • Pros: Continuous monitoring without needing someone to stand in the alley. 
  • Considerations: Lighting, camera placement, data quality, and cost. 
• Milking-related technologies 
  • Examples: Parlor automation upgrades, rotary systems with better ergonomics, or (longer-term) robotic milking. 
  • Pros: Potential for major labor savings over time. 
  • Considerations: Very high capital investment and major changes in management. 
• Workforce and management tools 
  • Examples: Digital checklists, protocol apps, tools for training and standardizing milking routines. 
  • Pros: Improved consistency, easier training, better communication. 
  • Considerations: Adoption depends on people and follow-through. 

Within a couple of minutes, ExtBot has turned a broad problem (“we’re short on labor”) into a shortlist of realistic options that deserve a closer look. You’re now ready to move from “what exists?” to “what does the research actually say?” for one specific technology area. 

Step 2: Choose one technology and provide a research paper

From ExtBot’s overview, you decide to look more seriously at computer vision systems (CVS) for automated monitoring. You’ve heard claims about cameras that can detect lameness or body condition, but you’re not sure how mature the technology is or whether it has been tested at scale. 

You ask ExtBot for some recent research articles discussing the topics of computer vision and it brings you to the following article: 

Menezes et al. (2024). Artificial intelligence for livestock: a narrative review of the applications of computer vision systems and large language models for animal farming (Animal Frontiers; https://doi.org/10.1093/af/vfae048 ↗️). 

The article is long and detailed. It covers technical aspects of CVS and a wide range of applications. You do not have time to carefully read and interpret every section, but you would like to know: 

Step 3: Ask NotebookLM to translate the paper into farm-level insights

You upload the PDF of the Menezes et al. article  into NotebookLM (https://notebooklm.google.com). Now you can “chat” with the paper directly. You ask a targeted question: 

“I manage a 2,000-cow free-stall dairy and I am evaluating computer vision systems to help with lameness, body condition, and heat-stress monitoring. Summarize what this paper says about how accurate and ready these systems are for commercial dairy use, AND provide 5–7 practical takeaways for a farm like mine, in simple, non-technical language.” 

 NotebookLM reads the paper and returns a concise summary focused only on what matters for your situation. For example, it may highlight that: 

• CVS has shown promising accuracy for lameness scoring using video of cows walking, with performance comparable to trained observers in several studies. 
• Systems for body weight and body condition scoring using 2D or 3D images are improving and can reduce the need for manual scoring but still require good camera placement and calibration. 
• Heat-stress related applications (e.g., monitoring respiration rate or behavior) are emerging, but many are still at pilot or research-herd stage. 
• Many studies have been conducted in more controlled conditions than a typical commercial 2,000-cow free-stall operation, so robustness in the “real world” remains a key question. 
• Integration with herd management software and reliable individual identification (tags, pattern recognition, or both) are critical for making the data truly actionable. 

It then turns those findings into plain-language, farm-level takeaways, such as: 

1. Start where you already spend a lot of time. 
   If staff spend many hours per week visually checking cows for lameness, CVS for lameness detection is likely to provide value sooner than more experimental applications. 
2. Check that systems were tested in herds like yours. 
   Prioritize technologies that have been evaluated in large freestall herds, not only small or highly controlled research barns. 
3. Plan for data integration and alerts. 
   Systems are most useful when they can flag specific cows and send alerts into the software and workflows your team already uses. 
4. Expect a learning curve. 
   Even good systems require time to adjust camera placement, refine alert thresholds, and train people on how to respond to the information. 
5. Treat this as a pilot project first. 
   For a 2,000-cow herd, starting with one barn or group to validate the system before full-herd rollout is a practical way to manage risk. 

Instead of you having to decode a dense review article, NotebookLM extracts and organizes what matters for your type of herd and your specific problem.

Step 4: From information to an actionable plan

By combining ExtBot and NotebookLM, you can now move from “ideas” to an actionable plan. In this example, the farm manager might decide to: 

• Narrow down technology priorities to: 
  • An automatic sort gate for vet and hoof-trimming pulls, and 
  • A CVS system focused on lameness detection, where the evidence and potential labor savings are strongest. 
• Prepare a focused list of questions for vendors and key advisors, such as: 
  • In what type and size of herds has your system been validated? 
  • How are individual cows identified and how are alerts delivered? 
  • What training and support are included in the first year? 
  • Can we run a 6–12 month pilot in one barn before committing to full-herd installation? 
  • Decide what changes in treatment, trimming schedules, or grouping would follow from new information. 
  • Estimate realistic labor savings and wellbeing improvements from earlier detection. 

The LLM tools have not made the decision for you. Instead, they have: 

• Helped you identify options relevant to your situation. 
• Translated a complex research article into clear implications for your herd. 
• Supported a more structured discussion with your team and vendors. 

How to ask better questions (Prompting Tips)

The results you get from an LLM depend heavily on how you ask your question. It’s an LLM area that is gaining a lot of traction in the computer science community right now called “prompt engineering”. You don’t need to be an engineer, but phrasing matters when prompting an LLM. Hence, the usefulness of LLM tools depends heavily on how you ask your questions. A few simple strategies make a big difference: 

• Be specific about your context. 
  Include herd size, housing type, region, and your main constraint (labor, facilities, cash flow, etc.). 
• State your goal clearly. 
  Ask for “3–5 options,” “a short list of pros and cons,” or “practical steps for the next 6–12 months” rather than “tell me everything.” 
• Ask for plain language and structure. 
  Phrases like “Explain in simple terms,” “Use bullet points,” or “Focus on what I can actually do on my farm” help shape the answer. 
• Use pre-selected documents when possible. 
  Upload specific extension bulletins or journal articles to NotebookLM and ask, 
  “What are the main takeaways of this document for a 500-cow tie-stall herd in northern Wisconsin?” 

Trying a few different versions of a question is normal; a small change in wording often leads to much more helpful responses. 

Use LLM for…	Example Prompts
Staying current on research	“Summarize new Extension bulletins on mastitis prevention.”
Understanding complex topics	“Explain how SARA impacts farm profitability in plain English.”
Comparing management strategies	“Compare pros and cons of different pre-fresh diets.”
Drafting or translating content	“Write a short SOP for monitoring fresh cows based on Extension recommendations.”
Employee education	“Create 5 quiz questions about fresh cow care for farm staff training.”

Treat LLMs as tools, not as final authorities

Finally, it is important to remember what LLMs are and are not:

• They are very good at summarizing, organizing, and rephrasing information. 
• They are not perfect, and they can: 
• Be out of date if new research or prices have recently changed. 
• Occasionally “hallucinate” details that sound plausible but are not actually correct. 

For that reason: 

• Use LLMs as a starting point for learning and planning, not as the last word. 
• Double-check critical details (prices, dosages, legal requirements, etc.) with trusted sources. 
• Involve your veterinarian, nutritionist, and other advisors in any major decision that affects animal health, food safety, or large capital investments. 

Take-home message

When used thoughtfully, tools like ExtBot and NotebookLM can help dairy producers:

• Turn complex and rapidly evolving research into farm-specific insights. 
• Clarify options when facing a concrete dilemma or challenge 
• Prepare better questions for advisors and vendors, leading to more confident decisions. 

It’s important to understand that LLMs don’t “know” things, they predict likely answers based on patterns in their training data. That means they can sometimes make mistakes. To use them responsibly, always verify important information with trusted sources, avoid using AI outputs as a replacement for professional recommendations or scientific data, and treat the tool as a starting point, not the final word. You can also restrict the answer to come from trusted sources by either indicating the type of online sources allowed or by directly providing the sources. Generative AI and LLMs can be valuable tools for farmers who want to stay informed and make sense of new research. They can save time, increase access to information, and help translate complex science into practical ideas. Used wisely, they can become part of a modern farmer’s toolkit for lifelong learning and better decision-making. 

 Rather than replacing your experience or judgment, LLMs can become another practical tool in your management toolbox to help you stay current, ask better questions, and make informed decisions for your cows, your people, and your business. 

Authors

Ariana Negreiro, Doctorate – Dairy Science at the University of Wisconsin, Madison.

Joao Dorea, Associate Professor – Precision Agriculture & Data Analytics at the University of Wisconsin, Madison.