Designing a Farm Emergency Plan Utilizing Artificial Intelligence
Noah J. Berning1,*, Shawn G. Ehlers1, William E. Field1
Published in Journal of Agricultural Safety and Health 31(4): 325-348 (doi: 10.13031/jash.16412). Copyright 2025 American Society of Agricultural and Biological Engineers.
1 Department of Agricultural & Biological Engineering, Purdue University, West Lafayette, Indiana, USA.
* Correspondence: njbernin@purdue.edu
The authors have paid for open access for this article. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License https://creative commons.org/licenses/by-nc-nd/4.0/
Submitted for review on 24 April 2025 as manuscript number JASH 16412; approved for publication as a Research Article by Associate Editor Dr. John Shutske and Community Editor Dr. Michael Pate of the Ergonomics, Safety, & Health Community of ASABE on 27 October 2025.
Citation: Berning, N. J., Ehlers, S. G., & Field, W. E. (2025). Designing a farm emergency plan utilizing artificial intelligence. J. Agric. Saf. Health, 31(4), 325-348. https://doi.org/10.13031/jash.16412
Highlights
Abstract. The ability of three artificial intelligence systems (ChatGPT, Microsoft Copilot, and Google Gemini) to generate functional Farm Emergency Plans (FEP) for a typical Midwestern row crop grain farm was evaluated. Four prompts, each of increasing levels of specificity, were used with the three AI systems, yielding twelve distinct FEPs. A rubric was developed to evaluate each of the twelve AI products against the findings of a review of relevant current literature including academic, government, not-for-profit, and insurance sources to identify essential and consistent components of a FEP.
Both ChatGPT and Microsoft Copilot were found to provide valuable starting points for developing FEPs when detailed prompts were provided, while Google Gemini results were less useful. However, none of the systems were capable of independently generating FEPs at the time of this study. Plans that were deemed as unreliable or incomplete enough for application were primarily due to the diverse nature of agricultural operations, limited resources on agricultural emergency preparedness, and the lack of maturity of current AI systems. Findings showed the essential need of using AI systems in collaboration with human guidance and input from other evidence-based sources to create effective FEPs. Similar results were confirmed in which the AI systems were prompted for emergency responses to three specific farm-related emergencies as part of the FEP: (1) flowing grain entrapment, (2) hazardous agricultural chemical spills, and (3) anhydrous ammonia exposure. The need for additional input was found to be essential. Outcomes were limited in scope to the particular type of farm selected for testing and the ability of the AI systems when they were queried on 30 September 2024; 12 February 2025; and 7 March 2025. Since AI systems rapidly continue to mature as they are “exercised,” further inquiries will, therefore, yield different outcomes, because AI has become more sophisticated and developed every day. It should also be noted that for “best practices,” the inquirer should provide AI with any resources that they have found and provide multiple inquiries to gain the best and most accurate results. This study demonstrated the potential that AI offers to agricultural producers, specifically in emergency preparedness and response, while emphasizing prompt development and user competency to verify AI outputs.
Keywords.Agriculture safety, Artificial intelligence, Emergency, Farm emergency plan, Farm emergency preparedness, Safety plan.
Artificial intelligence (AI) is being rapidly introduced into all aspects of agricultural production, including applications related to the use of autonomous equipment, chemical application, crop selection, marketing, and food safety (Talaviya et al., 2020; Vrochidou et al., 2022; Ozguven, 2023; Pandey et al., 2024; Sood et al., 2024). With increasing reliability, these systems will play even greater roles in decision making in all aspects of agriculture (Giattino et al., 2023), where there are large numbers of complex variables. However, as with other areas of AI application, concerns have been expressed over reliability in certain areas, ensuring ethical responses, and the reliability of outputs lacking expert human input (Yang et al., 2025). One area that has not been extensively explored in the literature is the application of AI to agricultural safety and health, including topics related to emergency management. Under current federal Occupational Safety and Health Administration (OSHA) workplace safety and health regulations (1910.38), agricultural production operations with 11 or more employees or that provide seasonal or migrant worker housing are required to have an emergency action plan (EAP) in place that is accessible to all employees (OSHA, 2002). The EAP, henceforth referred to as the farm emergency plan (FEP), is intended to provide action steps designed to protect employees in the event of an emergency. No published source was identified that clearly claimed the use of AI in developing these plans or specifically disallowed the use of AI in meeting regulatory requirements.
The three AI systems selected for use in this study were chosen due to heavy traffic and relative availability to users, making them the most likely systems to be used by farmers to generate FEPs. A zero-shot approach, meaning that the AI was not given any preliminary information from the researcher, was utilized to mimic, at this time, a farmer (or farm manager) approaching AI to create a novel FEP. This was done to determine if AI is currently a suitable resource for farmers to look to for the generation of an FEP. It is clearly recognized that additional queries will result in a maturing of the AI responses.
According to analysis by Similarweb, ChatGPT had 5.9 billion visitors in September 2025, ranking it third amongst programming and development software in this dataset (Similarweb, 2025). According to sentisight.ai (2025), Google Gemini “maintains a substantial user base of 275 million monthly active participants.” Gemini is also readily available to the public, as the Google search feature now features an “AI Overview” of whatever may be entered into the search engine by the user (Reid, 2024). Similarly, Microsoft Copilot has been introduced into various applications of the Microsoft 365 portfolio, such as Microsoft Word and Microsoft Outlook; again, making it a common tool the public may see (Spataro, 2023).
This study was conducted to compare farm emergency plans (FEPs) generated by AI systems to FEPs and templates previously developed by emergency management professionals, disaster relief non-profits, and governmental agencies. In doing so, the study looked to validate the accuracy, attention to detail, and reliability of the generated emergency plans as well as three specific event responses. Each of the three AI systems selected was provided with four prompts of increasing specificity to generate an emergency plan. Results were then compared to existing FEPs that have been used by agricultural businesses for decades. The goal of the study was to provide agricultural producers with a reliable strategy to develop, from credible online sources, a customized FEP that would enhance their operation resiliency before, during, and after a significant event.
Literature Review
Development of Farm Emergency Plans
A literature review was conducted to determine the current industry standards or best practices for FEP from a variety of sources, including university/academic faculty or departments, governmental agencies, not-for-profit disaster relief organizations, and insurance providers. Findings were used to evaluate the AI-generated FEPs. The following is a representative summary of findings.
Purdue University Extension’s Indiana PREPared program (Purdue Rural Emergency Preparedness) (2018) provided a publicly available template for farms to build their own FEP. The FEP template was a concise document, with various sections with fillable elements where users enter their specific information. The first section allows the user to record the primary contacts of the farm, including the owner, worker(s), neighbors, equipment dealer(s), chemical and fuel suppliers, and any other business or individual(s) directly involved with the operation. Additionally, there were sections for physical address; GPS coordinates, and other information that could be used to provide emergency responders the ability to timely locate the farm and other farm sites that may be in isolated locations, as many farm locations do not have a single 911 address associated with the site, other than the owner’s home. In some cases, emergency service agencies may not have “zoomed out” views of surrounding areas identifying local community and natural features which could impact emergency events such as chemical releases or field fires. The Purdue FEP template also contained sections to record any emergency contacts that may be essential, such as first responders, poison control, or utility companies. The template provided examples of various types of operations commonly found on mid-west farms/ranches and space for special considerations involved with the specific farm operations, such as on-farm fuel storage and manure storage. The final sections of the template dealt with mapping the layout of the farm and the surrounding areas, highlighting and making notes about various response resources or other important aspects of the farm, such as locations of utility shutoffs and water sources. This template served as a foundational starting point for developing a customized FEP, generating a concise, easily navigated form to collect necessary emergency information.
The University of Minnesota Extension provided a digital article to highlight major components to include in FEPs (Krekelberg, 2021). The article stressed the importance of having maps with lists of access routes to the farm, a comprehensive emergency contact list for those involved in managing the farm, and external actors and responders. It was recommended that communicating with the insurance agencies providing coverage for the operation, as well as individuals involved in the ownership and management of the farm be done. Krekelbreg (2021) recommended having a farm inventory for livestock, stored crops, farmed acres, equipment, fuel and fertilizer storage capacity, and animal medications. Insurance was a significant focus of this article, stressing the importance of maintaining sufficient coverage to mitigate risk and maintaining regular communication with the agency. Also, it was recommended that emergency supplies be inventoried and that potentially more secure areas of the farm be designated as emergency meeting areas and communicated to staff and family. Several types of plans for specific emergencies that should be considered were addressed, including having an evacuation plan ready for both humans and livestock. Farm emergencies addressed ranged greatly from natural disasters, man-made (both malicious and not), and medical emergencies, such as heart attacks, allergic reactions, or on-the-job incidents, such as machinery overturns, entanglements, and/or back over/runovers. It was noted that it is important to be aware of specific risks associated with each farm/ranch operation and customize the FEP accordingly. Lastly, the article stressed the need to be prepared for toxic gas and entrapment emergencies, namely within confined spaces such as grain bins and manure pits, or exposure to the operation of internal combustion engine-powered devices in these locations.
Farm Bureau Financial Services (2023) developed an article on recommendations to be included in the generation of an FEP. This article recognized that different farms may have varying needs as to what needs to be included in the plan, and it generally recommended that “farm [emergency] plans should include emergency escape procedures and routes, a system for accounting for people, a designated meeting place and information for emergency contacts.” As prominently found in other documents reviewed, this resource also recommended a map of the farm facilities and surrounding area, highlighting areas for sheltering in place, as well as noting evacuation routes and identifying utility shutoffs. Guidance on generating certain areas of the emergency plan was included; and concluded by stressing the importance of maintaining up-to-date insurance policies and communications with impacted parties throughout the design of the emergency plan.
The National Dairy Farm Program (2024) provided a brief guidance document for generating an FEP for a modern dairy operation. This document recommended various types of emergency actions to consider when developing the plan, specifically citing employee injuries, biosecurity, natural disasters, temperature extremes, contagious disease outbreaks, power failure, and manure spills. The document also stressed that the priorities in both developing these plans and in the actual times of disaster should be the safety of people involved first, and then the protection of livestock and property in jeopardy. The guide stated that the emergency plan should be able to identify potential issues that may arise on the farm and walk through the response steps to address those issues. It was specifically noted the importance of including essential contact information for the farm manager, emergency responders, neighbors or family, and contacts for any external organizations the farm interacts with (i.e., nutritionist, veterinarian, etc.). Eleven best practice items for the farm/ranch were identified: (1) training for employees, (2) emergency contacts, (3) backup fuel and power, (4) fail-safe alarms, (5) water supply backup, (6) feed supply back up, (7) emergency releases on gates, pens, and barns, (8) humane euthanasia protocol, (9) animal identification, (10) evacuation equipment, and (11) insurance on buildings and animals.
American Family Insurance (2024) stressed similar aspects that can be seen in other articles and templates found in the literature review. One of the unique focus areas of this article was the importance of learning from previous emergencies and experiences to periodically update the emergency plan to incorporate new information. Additionally, this resource insisted upon the importance of maintaining inventory of all assets and to regularly tour all acreage involved with the farm, taking note of any potential risks or hazards that could arise. The article suggested that mapping fields and structures is not enough and recommended that the farm ensure that the facility is organized in the safest manner; for instance, keeping volatile or hazardous items stored in secured locations away from vulnerable assets.
Summary of Review of Literature for FEP Development
Given the diversity of agriculture in the United States, no “one size fits all” approach to farm emergency planning was found. Each of the reviewed resources fit different agricultural enterprises such as dairy, swine, etc., but none were truly universal. Despite the differences of these sources, all contained common themes that were present throughout all reviewed, nationally available, and generally accepted “quality sourced” material. These themes were assembled into nine categories: (1) primary contact information, (2) employee information, (3) emergency/first responders’ information, (4) farm inventory, (5) direct responses to scenarios, (6) map of the farm, (7) employee training on emergency response, (8) insurance information, and (9) evacuation procedures. Topics are listed with descriptions in table 1.
| Table 1. List of categories scored for each response. | ||
| Category | Description | |
| Primary contact information | Contact information for key personnel on the farm including most viable access points | |
| Employee information | The names and contact information of all employees and residents of the farm | |
| Emergency/first responder’s information | The contact information for 911, local police, local fire, poison control, or other local emergency responders | |
| Farm inventory | Location, amount, and identification of hazardous materials, as well as items useful in response to an event, or other assets that may be seen as vulnerable or practically contributing to a disaster | |
| Direct response to scenarios | Includes having direct response steps for possible emergencies such as grain bin entrapment, toxic manure pit gas exposure, or chemical spills | |
| Map of farm | The location of critical assets of the farm and should identify any remote locations | |
| Employee trainings | Safety trainings and promotion of employee best practice for everyday work | |
| Insurance information | Any information the farm operator deems necessary to record about the insurance provider of the farm and types of coverage. This information should be coordinated with an insurance professional and may also be useful in establishing and identifying assets and perceived risks | |
| Evacuation procedure | Protocol for evacuating humans or livestock, and/selection of a secure meeting place on the farm. | |
State of Artificial Intelligence in Agriculture
The use and capabilities of AI have improved over the last several years and would be more accurately described as evolving daily in capabilities (Giattino et al., 2023). The surge of AI capabilities, as well as an increase in large-scale models has demonstrated this continued development (Giattino et al., 2023). Progress, however, has not come without challenges; Artificial Intelligence in Agriculture, published in the Journal of Physics: Conference Series, discussed some of the current uses, or perceived possible uses for AI in agriculture (Zha, 2020). Zhanoted challenges with the integration of AI, namely looking at the substantial discrepancy in the adoption of mechanization by agricultural producers across the globe; issues in transferring experimental inputs into real-world applications; and concerns with user information and privacy (Zha, 2020). Yang et al. (2025) expanded on the current distrust that individuals in the agriculture industry have with AI and its impact, especially in relationship to knowledge gaps and ethical data handling. Challenges for AI were then bridged to how AI is being used currently and how it changed the task for the user.
Regardless of current or forthcoming challenges, there is clearly a presence of AI in agriculture, and it will not go away. The increased mechanization and technological advancement of the industry has been seen to be compatible with AI in various facets. Talaviya et al. (2020) have highlighted how AI is used in the optimization of chemical applications and irrigation by way of in-field sensors and decision-making tool. Additionally, Computer Vision in Self-Steering Tractors specifically discussed the implementation of visual-response AI in the development of autonomous and automated machinery (Vrochidou et al., 2022). Adding to these two specific examples, other uses of AI were found in use in livestock management, data processing, crop health monitoring, and unmanned aerial vehicle (UAV) usage, among other areas (Ozguven, 2023; Pandey et al., 2024). However, the literature indicates that the use of AI in agriculture emergency planning has not been extensively pursued. Furthermore, AI has shown an increasingly larger role in agriculture, and the avenue of safety in the industry serves as another, less-explored pathway for implementation.
Comparing Artificial Intelligence Sources
As aforementioned, the determining factor for the three AI systems chosen was their availability to the public and their usage data. No specific tests or observations were identified that have specifically analyzed AI chatbot applications for agricultural safety or emergency preparedness. However, other sources were found where the same three AI systems were used in comparing their accuracy and reliability in healthcare (Eneva and Dogan, 2025; Rossettini et al., 2024). DynaTech Systems also compared each of the three AI chatbots that were used in this study: ChatGPT, Microsoft Copilot, and Google Gemini from a basic usage application. ChatGPT was developed by OpenAI, using the GPT-4 architecture, and Microsoft Copilot leverages OpenAI’s models as well in its architecture. Google Gemini uses Google’s knowledge base and natural language processing capabilities (DynaTech Systems, 2024). It was noted that ChatGPT was further developed, with trillions of tokens compared to the other two chatbots’ billions of tokens, as well as faster response time and customization (DynaTech Systems, 2024). Tokens refer to the pieces of word data that AI utilizes to analyze and respond to text, thus having more tokens would create a more effective AI system (OpenAI, 2024). DynaTech Systems also compared each of the three AI chatbots that were used in this study: ChatGPT, Microsoft Copilot, and Google Gemini from a basic usage application. ChatGPT was developed by OpenAI, using the GPT-4 architecture, and Microsoft Copilot leverages OpenAI’s models as well in its architecture; meanwhile, Google Gemini uses Google’s knowledge base and natural language processing capabilities (DynaTech Systems, 2024).
Proper Prompt Development
Effective Prompts for AI: The Essentials, from the MIT Management STS Teaching and Learning Technologies website, provided important context for the development of the prompts used in this study (MIT, 2024). Identifying detailed prompts for AI to generate detailed and thorough results was identified as essential. Detailed prompts can and should be used to reap more direct and reliable responses, because “the way [the writer] frames prompts shapes the AI’s output.” Three pieces of advice were provided by the MIT authors to create effective prompts: (1) provide context, (2) be specific, and (3) build on the conversation. Prompts 1 and 2 were primarily focused on in the development of the prompts used, giving context and being specific. Therefore, the prompts were used because they were determined to fit the framework for what makes an effective prompt for an AI chatbot.
Methodology
Design
The design of this experiment staged three popular AI chatbots to four different prompts (see table 2) of increasing specificity. The three AI engines selected were ChatGPT 4.0, Microsoft Copilot, and Google Gemini (models as of September 2024, February 2025, and March 2025). The prompts presented sought to generate a farm emergency plan for “typical” mid-sized Midwest row-crop farms that grow primarily corn and soybeans. Following guidance in prompt development, as discussed in the literature review, four prompts were generated, with increased specificity. These prompts were entered identically for each of the three AI engines and the responses were recorded. Additionally, through the literature review, several FEP guidance documents and templates were generated to compare the AI-generated FEPs utilizing a rubric that was developed from common themes of emergency plans from evidence-based sources such as university/academic, governmental entities, not-for-profit disaster response groups, and insurance providers.
| Table 2. Prompts used to retrieve response from AI systems. | |
| Prompt Number | Prompt |
| 1 | Write me a farm emergency plan. |
| 2 | Write me a comprehensive farm emergency plan for a mid-size grain operation in Indiana. |
| 3 | Write me a farm emergency plan for a mid-sized corn and soybean operation in Indiana, with an on-site grain system. This plan should be at least 5 pages in length, should give at least 5 real and reliable sources, and should give real instruction for employees about different safety practices and responses to emergency scenarios. |
| 4 | Create me a comprehensive farm emergency preparedness plan, that is 7 pages in length, provides 10 reliable and real resources, and is for a mid-size corn and soybean farm in Indiana, with an onsite grain system. This should include general safety practices for everyday activities, as well as for emergency scenarios. This should not just be a template but should provide actual breakdowns and procedure for each emergency scenario. |
Data Analysis Technique
The literature reviewed identified several sources that outline and guide the development of FEPs. From these sources, nine topic areas were chosen (see table 1), based on consistencies seen between farm emergency documents, that would be used to scale the accuracy and extent of each AI response. Each of these elements chosen was mentioned as an important characteristic of an effective FEP by these resources and thus was highlighted as an important element in this study. The topics with their weighting are noted in table 3.
A weighting scheme was developed for each of these areas as part of a scoring system for how effectively each AI response addressed a given topic, based on the findings of the literature review. For each of the categories, the weight was determined based on how frequently it was mentioned in reviewed literature, and the importance that the category has in maintaining a complete FEP. The weights of the nine categories were percents of one hundred, and each topic was given a score from zero to five on the effectiveness that a particular response addressed that topic using a Likert scale. The scorecard used for each response, with the topic weighting, is shown in table 3.
| Table 3. Weighting scale utilized for the nine categories used in the study. | ||
| Category | Weight | Score (0-5) |
| Primary contact information | 20.0% | |
| Employee information | 15.0% | |
| Emergency/first responder's information | 10.0% | |
| Farm inventory | 10.0% | |
| Direct responses to scenarios | 10.0% | |
| Map of farm | 15.0% | |
| Employee trainings | 5.0% | |
| Insurance information | 5.0% | |
| Evacuation procedures | 10.0% | |
| 100% | 0 | |
Table 4 displays the Likert Scale that was used by the authors to score the responses from the AI systems. As aforementioned, if a topic was not present in the response, it was scored as a zero. However, if a topic was present in the response, it was scored on a scale of one to five, with one being that a topic was present but needed significant adjustments, and an increasing score reflecting more complete representation of a given topic. The scoring was based on the quality of the delivery of each topic using the combination of the weighting scale (table 3) and the Likert Scale (table 4) developed from the literature review of government, university, and nonprofit sources. The scores for each response were evaluated by the authors of this study. Further descriptions for each score are seen in table 4.
| Table 4. Likert Scale used to score AI system responses. | ||||||||||||
| Rating | 0 | 1 | 2 | 3 | 4 | 5 | ||||||
| Description | Topic is not present | Topic is present, but delivered very poorly, requires significant modification from user | Topic is present, delivered poorly, and requires modification to be used in final draft | Topic is delivered fairly well, but is a work in progress, needing some attention from the user for final draft | Topic is delivered well and requires a minimal amount of modification to be used in final emergency plan | Topic is very well delivered, and requires little to no modification from user to be adequate in final emergency plan | ||||||
For the two prompts that requested sources, the sources were tested for repeatability for relation to the initial prompt. During verification of the sources, it was evident that the sources provided fell into three distinct categories: (1) resources provided that referenced direct webpages but linked the homepage of the website; (2) sources that were referencing a website in general; and (3) sources that either referenced broken links or were not linked whatsoever.
Additionally, three common farm emergencies, grain bin entrapment, agricultural material spill, and anhydrous ammonia exposure, were inserted into each of the AI systems used. This was done to determine how effectively AI addressed well recognized emergencies and if there was potential for AI to be used to educate farmers on how to respond to these situations. To evaluate, the authors compared the responses for each scenario to peer-reviewed articles to determine the accuracy and safety of the AI responses. The prompts were given to the AI on 12 February 2025, and 7 March 2025, and the prompts are shown in table 5.
Results and Discussion
AI responses were analyzed for three chatbot offerings with four prompts each, yielding a total of twelve analyzed reports. These reports were ranked utilizing the developed rubric, ranking each of the nine criteria on a 0 to 5 Likert scale.
ChatGPT Responses
Table 6 shows the results from the analysis of the four prompt responses from ChatGPT. In the first prompt, four elements were missing, and most others at least provided some guidance for the user to utilize. The higher scoring categories in this prompt, specifically the primary contact information, were set up as a format that allowed the user to just enter their own information, which was seen as very effective; the score out of five for this result was 1.9. The second prompt included seven of the targets, with a score of 2.55. The third and fourth results reaped scores of 2.70 and 2.95, respectively, with the increase in score primarily coming from the addition of missing categories in previous responses or by providing more context. Noteworthy was that although the final prompt was the best result, it did omit the evacuation procedure section, which was present in all other iterations.
| Table 5. Prompts used for direct emergency scenarios. | |||||
| Emergency Scenario | Prompt | ||||
| Grain bin entrapment | How should I respond to grain bin entrapment? | ||||
| Agriculture material spill | How do I respond to fertilizer chemical spill on a roadside? | ||||
| Anhydrous ammonia exposure | How do I respond to anhydrous ammonia exposure? | ||||
Table 6. ChatGPT Response Scores by Prompt. | |||||
| Unweighted Score | |||||
| Category | Prompt 1 | Prompt 2 | Prompt 3 | Prompt 4 | |
| Primary contact information | 4 | 4 | 4 | 4 | |
| Employee information | 0 | 1 | 0 | 0 | |
| Emergency/first responder's information | 4 | 5 | 5 | 5 | |
| Farm inventory | 0 | 0 | 2 | 3 | |
| Direct responses to scenarios | 3 | 4 | 4 | 5 | |
| Map of farm | 0 | 0 | 0 | 3 | |
| Employee trainings | 2 | 3 | 4 | 4 | |
| Insurance information | 0 | 3 | 4 | 4 | |
| Evacuation procedures | 3 | 4 | 4 | 0 | |
| Unweighted Total Score (out of 45) | 16 | 24 | 27 | 28 | |
| Weighted Total Score (out of 5) | 1.9 | 2.55 | 2.7 | 2.95 | |
The format of the ChatGPT response reflected that of the FEP templates that were found in the literature review, such as the one found in Purdue University Extension IN-PREPared (2018). It should be noted that from the second prompt forward, the response included phone numbers for national or state agencies, namely, Indiana Poison Control, the Indiana Department of Labor (labeled as OSHA in the response), and the Environmental Protection Agency, Region 5. These telephone numbers were confirmed accurate, except that the OSHA number provided was the direct contact for the Indiana Department of Labor. It should also be noted that each response gave directions for at least five different emergency scenarios, adding more information with each new prompt.
It was clear that as the prompts became more complex, more information was provided, maintaining accurate information. For instance, from the second prompt onward, grain bin entrapment was an included section, with the recommended response reflecting similar strategies to expert sources. When analyzing the sources provided by ChatGPT, it was not clear if this information was directly pulled from the sources provided or if random sources were provided. In the third prompt, four of the five sources were actual links, while the other led to a 404 error, meaning the page was not found or that the link does not exist, while the fourth prompt only yielded five of ten sources going to actual websites. The issue was that specific pages were referenced; however, the links provided went to website homepages, rather than the specific webpages.
Microsoft Copilot Responses
The results from Microsoft Copilot are recorded in table 7. Overall, the scores from Copilot were not as high as ChatGPT, achieving scores of 1.50, 2.00, 2.45, and 2.05 for each of the prompts. A noticeable difference between Copilot and ChatGPT was that Copilot did not present its information in a way that was a “fillable” format like ChatGPT. Copilot formatted its information by providing very brief descriptions of the areas that it included, rather than showing a way to format them. This produced a document that was more like a how-to document, rather than an actual FEP. The additional effort required to generate an FEP was what caused Copilot to score lowly. However, it should be noted that the information was an adequate starting point for an FEP, but lacking the maturity seen in other trials. Additionally, unlike the other two chatbots, the fourth prompt response was shorter than the third prompt response, showing a disregard of the prompt directive. The fourth prompt was also not the highest scoring response, which was not the case in the other two chatbot trials.
| Table 7. Microsoft Copilot Response Scores by Prompt. | ||||
| Unweighted Score | ||||
| Category | Prompt 1 | Prompt 2 | Prompt 3 | Prompt 4 |
| Primary contact information | 2 | 3 | 2 | 2 |
| Employee information | 2 | 2 | 3 | 3 |
| Emergency/first responder's information | 2 | 3 | 3 | 3 |
| Farm inventory | 2 | 2 | 3 | 0 |
| Direct responses to scenarios | 0 | 0 | 3 | 3 |
| Map of farm | 0 | 0 | 0 | 0 |
| Employee trainings | 2 | 3 | 3 | 4 |
| Insurance information | 0 | 3 | 3 | 0 |
| Evacuation procedures | 3 | 3 | 4 | 4 |
| Unweighted Total Score (out of 45) | 13 | 19 | 24 | 19 |
| Weighted Total Score (out of 5) | 1.5 | 2.00 | 2.45 | 2.05 |
As previously mentioned, the format of Copilot’s response did not follow a template but rather presented a document that walked through briefly what sections should be included in an FEP. For instance, rather than listing out different first responder contacts for a user to fill in, the responses noted examples of services that should be listed in an emergency contact section. Additionally, the flow of the document did not reflect standard components found in other FEP guiding documents. An example being that rather than having the primary contacts for the farm being a section with specific farm information, it was listed in the third section under emergency contacts. It was not until the third prompt that the response gave a specific example of how to react to a given emergency, under the section “detailed instructions for employees.” Moreover, the responses from Copilot contained a high frequency of the target elements of an FEP, presenting the information as a description of what would be expected of a given section, rather than filling in the information directly, leading to a lower overall score.
Copilot did not provide official citations, but rather just the name of the webpage or article it claimed to utilize. Upon evaluating these sources, the third prompt provided two sources that were real, relevant, and that went directly to the web page being referenced by the document. However, the sources also included two links that worked but were not related to FEPs directly but rather to grain bin management in general. The last source for this prompt was not linked at all. Meanwhile, the final prompt’s sources included one correct, unrepeated source; one source that was linked twice as separate sources; one source that was incorrectly linked; one link that went to the homepage of a website; four unlinked sources; and one file that was marked as unsafe by the Google Chrome browser being used. Microsoft Copilot did not perform well with the references it gave, as it did not provide them with a proper citation format.
Google Gemini Responses
Google Gemini did not provide sufficient responses to the prompts provided to develop an FEP. Google Gemini’s response results are given in table 8 below, with scoring totals of 0.60, 0.90, 1.25, and 1.50. The key contributing factor to this was the disregard that Gemini showed for the input message and its relevance to agricultural scenarios. All four of the prompt responses did not include a section for primary farm contact information, nor employee information. None of the responses called for a farm inventory of essential supplies or a map of the farm. In the first three responses, there were few references to agriculture-related topics beyond one bullet point on animal care. No other references were made in the first response, and the singular agriculture example was a section on grain in the second and third responses. The final prompt added a section on equipment failure, natural disasters, and crop and animal diseases. The Gemini responses resembled a template, with little in-depth evidence or justification, and lacked some key features with the absence of farm manager contact or employee information and the lack of farm examples, reflecting in the scoring seen in table 8.
| Table 8. Google Gemini Response Scores by Prompt. | ||||
| Unweighted Score | ||||
| Category | Prompt 1 | Prompt 2 | Prompt 3 | Prompt 4 |
| Primary contact information | 0 | 0 | 0 | 0 |
| Employee information | 0 | 0 | 0 | 0 |
| Emergency/first responder's information | 3 | 3 | 3 | 4 |
| Farm inventory | 0 | 0 | 2 | 2 |
| Direct responses to scenarios | 0 | 2 | 2 | 3 |
| Map of farm | 0 | 0 | 0 | 0 |
| Employee trainings | 0 | 0 | 3 | 3 |
| Insurance information | 2 | 2 | 2 | 3 |
| Evacuation procedures | 2 | 3 | 3 | 3 |
| Unweighted Total Score (out of 45) | 7 | 10 | 15 | 18 |
| Weighted Total Score (out of 5) | 0.6 | 0.9 | 1.25 | 1.5 |
The sources referenced were not given in citation format, only proving the title of the website and a hyperlink to that website. Also, none of the sources referred to specific webpages, but rather the home page of the website mentioned. In the third prompt, all five sources were relevant and repeatable, but the information that would have been used in a plan was not directly referenced. Meanwhile, the final prompt reaped three links that were incorrect, one missing link, and six links that were correct and relevant.
Direct Response to Emergencies
All three AI systems were presented with three emergency situations to see if AI was able to effectively provide guidance for those situations. Like the FEPs, the responses from AI were compared to other relevant and evidence-based sources.
Scenario 1: Grain Bin Entrapment: For grain bin entrapment scenarios, the responses were compared to Caught in the Grain from North Dakota State University Extension (Hellevang, 2023) and the Purdue Agricultural and Biological Engineering Agriculture Confined Spaces website (Ag Confined Spaces, 2024). Responses for this scenario were recorded in table A1. All three of the AI responses were very similar and consistent with the information from these two sources, containing very similar elements to each other and with the sources. The only noticeable difference was that Google Gemini also explained how to remove the grain around a trapped individual, while the other two AI instructed to wait for or work with trained rescuers. Beyond that difference, all the responses emphasized contacting 911, turning off all equipment, and maintaining communication with the victim, among other notes. Additionally, ChatGPT and Microsoft Copilot did provide advice for future prevention, while Google Gemini did not. Yet, all three responses were accurate and complete when compared to the two sources.
Scenario 2: Roadside Fertilizer Spill: The responses were compared to two sources, one from Farm Bureau Financial Services (2022) and one from the Minnesota Department of Agriculture (2025). Responses for this scenario were recorded in table A2. All three AI systems provided adequate responses to this scenario, including key points that were highlighted in both comparative sources, including ensuring personal safety, securing the area around the spill, calling emergency services, reporting the spill, and only cleaning up the spill independently if the individual is trained to do so—otherwise wait for professionals to assist. As with the previous scenario, only ChatGPT and Microsoft Copilot provided information for future prevention; however, only Google Gemini made note that the different chemical compositions of various fertilizers could impact the situation, the environmental impacts that may be caused, and that regulations may vary on the location of the spill.
Scenario 3: Anhydrous Ammonia Exposure: Anhydrous Ammonia Safety from Farm and Ranch eXtension in Safety and Health (Farm and Ranch eXtension in Safety and Health (FReSH) Community of Practice, 2012) and Safe Handling of Anhydrous Ammonia from Ohio State University Extension (Jepsen and McGuire, 2017) were both used in comparing the results from the AI systems. Responses for this scenario were recorded in table A3. Both sources highlighted the importance of flushing the exposed skin and eyes with water immediately, removing contaminated clothing, and seeking immediate medical attention. Like the previous two scenarios, each of the AI systems highlighted many of the key points noted in the two sources. In all three sample emergency scenarios, the AI systems provided step-by-step responses that, when compared to sources mentioned, were accurate in their instructions.
Discussion
Figure 1 shows the comparison of each chatbot’s performance score for each of the form prompts. It is evident that ChatGPT performed best, with an average of 0.525 points better than Microsoft Copilot and 1.46 points better than Google Gemini. Again, it should be noted that on average, the more specific prompts provided more satisfactory outputs, except for with Microsoft Copilot in one situation.
 |
| Figure 1. Comparing chatbots by weighted score by prompt. |
Figure 2 shows the average score from each prompt, showing that even though it was not the case for all three chatbots, there was a general trend of more detailed prompts providing more sufficient results. However, it was observed that at a certain point, increased specificity of a prompt would reap diminishing returns in the quality of a response. Thus, in this trial, the relationship between increased detail of an input is not necessarily linear to the quality of the output.
 |
| Figure 2. Average score by prompt (out of 5). |
ChatGPT and Microsoft Copilot both included a similar number of the desired topics, even though they were not presented to the same level; and considering both ChatGPT and Copilot utilize the OpenAI structure, it could be hypothesized that this would be the case. However, the content of the responses was notably different. As aforementioned, ChatGPT reaped responses that were very similar to the FEPs found during the literature review. Prominent features of the ChatGPT responses were that the emergencies discussed had directions to respond to the crisis and other areas such as contact information simply needed to be filled in by the user, with some contacts already being included. Conversely, the Microsoft Copilot results are more closely related to guiding documents rather than fillable templates. Many of the categories mentioned stated that they need to be included, why they are important, or what may be some aspects that would make them effective, rather than directly saying what needs to be included. Meanwhile, Google Gemini was significantly less accurate at including the target categories and at displaying them effectively. This included a lack of agricultural examples for the FEP and the exclusion of primary farm contacts. The average scores for each of the AI chatbots are shown in figure 3.
 |
| Figure 3. Comparison of average weighted scores among AI chatbots. |
Recommended Best Practices
Upon the completion of this study, the authors found that certain notes were required regarding this study and future best practices for study replication and/or if this study is to be used in practice on the farm. Firstly, it should be noted that the FEP portion of this study was generated on 30 September 2024, and that the specific scenarios were generated on 12 February 2025, and 7 March 2025, utilizing the most up-to-date AI available at that time; and that AI is constantly growing and learning every day, thus the responses to prompts provided in this study would likely yield different results as AI continues to develop.
Additionally, the intention of this study was to mimic the situation in which a farmer or rancher would approach AI looking for an FEP template for their operation. A zero-shot approach, meaning no FEP material was provided, was taken, with the context that can be seen in the methodology of the prompts, with one response being taken. This was done to mimic the real-world scenario that is most likely to occur: that a farmer may go directly to a public and free AI chatbot and provide simple context and expect a result. This study was done to verify the validity and completeness of such a scenario.
It should be noted that it is the recommendation and opinion of the authors that to achieve the optimum results for generating an FEP or response to a direct scenario, the person inquiring of the AI chatbot with the task should provide as many resources as possible, such as the ones that were discovered in the literature review for this study, to provide better results. Also, it is recommended that a “scaffolded” approach of building AI prompts in small, manageable processes be utilized so that better, more sophisticated responses are yielded. Liu et al. defined “answer mapping” to describe the method of building and developing prompts to provide to AI models (Liu et al., 2023). Liu et al. noted the importance of utilizing properly engineered and developed prompt templates to achieve the best possible response (Liu et al., 2023). Additionally, as stated in the study, the more context for the specific scenario that is provided, the better results that AI will provide in the response. Ouyang et al. stated that they “were able to mitigate most of the performance degradations introduced by [the use of] fine-tuning” when working with AI models (Ouyang et al., 2022). Ouyang et al. noted that strong models are reactive to details and fine-tuned prompts, which would stress the importance of providing such prompts in a replication of the study performed in this paper. Again, as AI continues to develop and become more sophisticated every day, the ability of AI to complete these tasks is likely to increase as well.
Conclusions
Upon review of the ability of AI to create FEPs, there were distinct differences in the manner in which each responded to the given prompts. As previously mentioned, ChatGPT resembled farm emergency templates like that of the Purdue University Extension IN-PREPared document, being stylized as a template for the user to fill in. Meanwhile, the Copilot results reflected more of what should be included in an emergency plan, rather than being a fill-in-the-blank template. This style mirrored articles found in the literature review, such as from the University of Minnesota (Krekelberg, 2021). Google Gemini was minimal in its response, not providing detail beyond the necessities of an emergency plan for any scenario, whether it be for a person’s home or a business.
Regardless of the current results of this study, it should be noted that the FEP results were recorded on 30 September 2024, and the emergency scenario responses were recorded on 12 February 2025, and 7 March 2025. Additionally, as aforementioned in the literature review, AI is rapidly growing in use and competence not only year after year but also day to day (Giattino et al., 2023). Thus, if this study were to be conducted again, it is likely that the AI chatbots will perform differently. However, there are some key conclusions about the role that AI plays in developing effective FEPs and responding to agricultural emergencies that can be drawn. Although none of the results from the prompts were sufficient to create a complete FEP, the results from ChatGPT and Copilot proved to be an appropriate starting point to help in the development of an emergency plan if the prompt provided is detailed and direct. The importance of thorough prompts is shown as the results followed a trend of including more needed information as the prompt became more detailed. However, it should be recognized that at a certain point, the increase in the quality of an output from increased detail will eventually diminish.
It should also be noted that in the test of the AI’s ability to provide responses to direct scenarios, all three provided acceptable and accurate responses. The performance and accuracy of AI in these situations lead the authors to believe that there may be potential applications of AI in providing guidance or instruction to emergency responders or farm personnel on responding to emergencies they may encounter. For instance, the implementation of an AI application on a mobile device that may already be available to emergency first responders could prove useful in emergencies if the individual needed additional guidance in a timely manner.
As AI has an increasing role in the agriculture industry, this knowledge will be extremely beneficial, as there may be an interest in using AI to generate documents for farm safety and emergency preparedness. Although the results from 2 of 3 chatbots studied were good starting points for an FEP, it should still be noted, due to the real risks to human life and property, that additional research and collaboration should be done to make sure that specific details regarding responding to specific emergencies are covered and potential risks to first responders are identified. Also, a key lesson that was seen in this study was that, although it might not be perfect at this point, AI will have a significant place in agricultural emergency management, especially as its knowledge base continues to grow. In conclusion, the ceiling for the use of AI in agriculture is high, and this study served as one example of a use that may not be at the forefront of the industry’s mind.
Limitations of Study
A limitation of this study was that the request to develop an FEP was applicable to only one type of farm in the corn belt. It should be noted that the FEPs are variable by function, size, and numerous other factors that make each farm unique. The intent of this study was to select a basic farm type to test the overall reliability of AI in generating templates or specific breakdowns of an FEP. It should also be noted that these prompts were generated on September 30, 2024, and that the specific scenarios were generated on 12 February 2025, and 7 March 2025, utilizing the most up-to-date AI available at that time. Additionally, as discussed in an article from the MIT Management STS Teaching & Learning Technologies website, the importance of prompt development for AI chatbots is crucial to receiving good results (MIT, 2024). Therefore, due to the continuous and rapid development of AI, it should be noted that the results of this study are limited to the capabilities of AI on the date of the study, and that further adjustments and research would most likely show different results.
Acknowledgment
This work is supported by the National Needs Graduate and Postgraduate Fellowship (NNF), project award no. 2021-38420-34063, from the U.S. Department of Agriculture’s National Institute of Food and Agriculture.
Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and should not be construed to represent any official USDA or U.S. Government determination or policy.
References
Ag Confined Spaces. (2024). First Responders (Emergency Rescue Training) Materials. https://www.purdue.edu/engineering/abe/agconfinespaces/firsts-responder-resources/
American Family Insurance. (2024). Farm emergency management plan. Retrieved from https://www.amfam.com/resources/articles/on-the-farm/farm-emergency-management-plan
DynaTech Systems. (2024). Microsoft Copilot vs. ChatGPT vs. Gemini: Which one is the best AI Chatbot? Retrieved from https://dynatechconsultancy.com/blog/microsoft-copilot-vs-chatgpt-vs-gemini-which-one-is-the-best-ai-chatbot
Eneva, Y., & Dogan, B. (2025). Evaluation of medical diagnosis capabilities of three artificial intelligence models – ChatGPT-3.5, Google Gemini, Microsoft Copilot: Sustainable Development Goals (SDGs). J. Lifestyle SDGs Rev., 5(2). https://doi.org/10.47172/2965-730X.SDGsReview.v5.n02.pe03545
Farm and Ranch eXtension in Safety and Health (FReSH) Community of Practice. (2012). Anhydrous ammonia safety. Retrieved from https://ag-safety.extension.org/anhydrous-ammonia-safety/
Farm Bureau. (2023). How to create an emergency action plan for your farm. Farm Bureau Financial Services. Retrieved from https://www.fbfs.com/learning-center/your-farm-emergency-plan
Farm Bureau Financial Services. (2022). How to handle a chemical spill on your farm. Retrieved from https://www.fbfs.com/learning-center/how-to-handle-a-chemical-spill-on-your-farm
Giattino, C., Mathiew, E., Samborska, V., & Roser, M. (2023). Artificial Intelligence. Our World in Data. Retrieved from https://ourworldindata.org/artificial-intelligence#article-citation
Hellevang, K. (2023). Caught in the grain! AE1102. NDSU Extension. Retrieved from https://www.ndsu.edu/agriculture/ag-hub/publications/caught-in-grain
Jepsen, S. D., & McGuire, K. (2017). Safe handling of anhydrous ammonia. Ohioline - Ohio State University Extension. Retrieved from https://ohioline.osu.edu/factsheet/aex-594#:~:text=First%20Aid,for%20at%20least%2015%20minutes.
Krekelberg, E. (2021). Creating farm emergency action plans. Retrieved from https://extension.umn.edu/farm-safety/creating-farm-emergency-action-plans
Liu, P., Yuan, W., Fu, J., Jiang, Z., Hayashi, H., & Neubig, G. (2023). Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing. ACM Comput. Surv., 55(9). https://doi.org/10.1145/3560815
Minnesota Department of Agriculture. (2025). Responding to a spill. Retrieved from https://www.mda.state.mn.us/pesticide-fertilizer/responding-spill
MIT. (2024). Effective prompts for AI: The essentials. MIT Management STS Learning & Teaching Technologies. Retrieved from https://mitsloanedtech.mit.edu/ai/basics/effective-prompts/
National Dairy Farm Program. (2024). Comprehensive emergency action plan guidance. Retrieved from https://nationaldairyfarm.com/wp-content/uploads/2020/01/Comprehensive-Emergency-Action-Plan-Guidance.pdf
OpenAI. (2024). What are tokens and how to count them? Retrieved from https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them
OSHA. (2002). Emergency Action Plans (Standard No. 1910.38). Occupational Safety and Health Standards. Retrieved from https://www.osha.gov/laws-regs/regulations/standardnumber/1910/1910.38
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P.,... Ray, A. (2022). Training language models to follow instructions with human feedback. Proc. Advances in Neural Information Processing Systems 35 (NeurIPS 2022), 35, pp. 27730-27744.
Ozguven, M. (2023). The digital age in agriculture (1st ed.). CRC Press. https://doi.org/10.1201/b23229
Pandey, K., Kushwaha, N. L., Pande, C. B., & Singh, K. G. (2024). Artificial intelligence and smart agriculture: Technology and applications. Springer. https://doi.org/10.1007/978-981-97-0341-8
Purdue Agricultural and Biological Engineering. (2024). Agriculture confined spaces. Retrieved from https://www.purdue.edu/engineering/abe/agconfinespaces/
Purdue University Extension IN-PREPared. (2018). Farm emergency plans. Retrieved from https://www.purdue.edu/engineering/ABE/INPREPared/farm-emergency-plans/
Reid, L. (2024). Generative AI in search: Let google do the searching for you. Google. Retrieved from https://blog.google/products/search/generative-ai-google-search-may-2024/
Rossettini, G., Rodeghiero, L., Corradi, F., Cook, C., Pillastrini, P., Turolla, A.,... Palese, A. (2024). Comparative accuracy of ChatGPT-4, Microsoft Copilot and Google Gemini in the Italian entrance test for healthcare sciences degrees: A cross-sectional study. BMC Med. Educ., 24(1), 694. https://doi.org/10.1186/s12909-024-05630-9
SentiSight.ai. (2025). Google Gemini: How has it been received by users so far? Retrieved from https://www.sentisight.ai/google-gemini-how-has-it-been-received-by-users-so-far/#:~:text=Recent%20analysis%20reveals%20that%20Gemini,different%20scenarios%20of%20practical%20use
Similarweb (2025, October). chatgpt.com traffic analytics, ranking & audience [Website overview]. Similarweb. https://www.similarweb.com/website/chatgpt.com/#overview
Sood, A., Bhardwaj, A. K., & Sharma, R. K. (2024). Towards sustainable agriculture: Key determinants of adopting artificial intelligence in agriculture. J. Decis. Syst., 33(4), 833-877. https://doi.org/10.1080/12460125.2022.2154419
Spataro, J. (2023). Introducing Microsoft 365 copilot - your copilot for work. The Official Microsoft Blog. Retrieved from https://blogs.microsoft.com/blog/2023/03/16/introducing-microsoft-365-copilot-your-copilot-for-work/
Talaviya, T., Shah, D., Patel, N., Yagnik, H., & Shah, M. (2020). Implementation of artificial intelligence in agriculture for optimisation of irrigation and application of pesticides and herbicides. Artif. Intell. Agric., 4, 58-73. https://doi.org/10.1016/j.aiia.2020.04.002
Vrochidou, E., Oustadakis, D., Kefalas, A., & Papakostas, G. A. (2022). Computer vision in self-steering tractors. Machines, 10(2), 129. https://doi.org/10.3390/machines10020129
Yang, C. X., Baker, L. M., Mattox, A., Diehl, D., & Honeycutt, S. (2025). The forgotten factor: Exploring consumer perceptions of artifical intelligence in the food and agriculture systems. Future Foods, 11. https://doi.org/10.1016/j.fufo.2025.100553
Zha, J. (2020). Artificial intelligence in agriculture. J. Phys. Conf. Ser., 1693(1), 012058. https://doi.org/10.1088/1742-6596/1693/1/012058
Appendix
| Table A1. Grain Bin Entrapment Responses. | ||
| AI System | Response | |
| ChatGPT | Grain bin entrapment is a life-threatening emergency that requires immediate and careful action. Here’s how you should respond:
| |
| Table A1 (continued). Grain Bin Entrapment Responses. | ||
| AI System | Response | |
| Microsoft Copilot | Grain bin entrapment is a serious and potentially life-threatening situation. Here are some key steps to follow if you encounter or witness a grain bin entrapment:
Stay safe and take care! If you have any more questions or need further assistance, feel free to ask. | |
| [1] Grain Bin Entrapment: Don’t Let It Happen To You! [2] Grain Bin Entrapment: What If It Happens to You? A Case Study from an... [3] Grain Engulfment and Entrapment | Great Plains Center for Agricultural... | ||
| Google Gemini | Grain bin entrapment is a serious hazard that can lead to suffocation or death. If someone is entrapped in a grain bin, it is important to act quickly and safely. Here are the steps you should take:
| |
| Table A2. Agriculture Material Spill Responses | ||
| AI System | Response | |
| ChatGPT | A fertilizer chemical spill on a roadside requires a swift and careful response to protect people, the environment, and nearby water sources. Follow these steps:
| |
| Table A2 (continued). Agriculture Material Spill Responses | ||
| AI System | Response | |
| Microsoft Copilot | Responding to a fertilizer chemical spill on a roadside requires careful and immediate action to ensure safety and minimize environmental impact. Here are the steps you should follow:
Stay safe and take care! If you have any more questions or need further assistance, feel free to ask. | |
| [1] Roadside Chemical Spills: What You Need to Know and How to Handle Them [2] Ammonia Release During Transportation | Assessment of Chemical... [3] The First 30 Minutes: Critical Actions for Chemical Spill Response | ||
| Table A2 (continued). Agriculture Material Spill Responses | ||
| AI System | Response | |
| Google Gemini | Responding to a fertilizer chemical spill on a roadside requires immediate action to protect yourself, others, and the environment. Here’s a step-by-step guide: Immediate Actions:
| |
| Table A3. Anhydrous Exposure Responses | ||
| AI System | Response | |
| ChatGPT | Exposure to anhydrous ammonia is extremely dangerous and requires immediate action. If you or someone else is exposed, follow these steps:
| |
| Microsoft Copilot | Exposure to anhydrous ammonia can be very dangerous. Here are the steps you should take if you or someone else is exposed:
Stay safe and take care! If you have any more questions or need further assistance, feel free to ask. | |
| [1] Ammonia Solution, Ammonia, Anhydrous | NIOSH | CDC [2] Ammonia | Chemical Emergencies | CDC [3] Ammonia - Prehospital Management - HHS.gov | ||
| Table A3 (continued). Anhydrous Exposure Responses | ||
| AI System | Response | |
| Google Gemini | Responding to anhydrous ammonia exposure requires immediate and careful action.1 Here’s a breakdown of key steps, based on information from reliable sources like the CDC and OSHA: Immediate Actions:
| |