Evaluating on-line media and survey data to understand beliefs and sentiment associated with pastured and integrated poultry production among farmers and stakeholders

Commercial pastured and integrated poultry production (PIPP) represents an emerging area of poultry and crop production. As a “niche” area of commercial production, there is little overall knowledge about public opinion associated with PIPP. Here we utilized two complementary approaches to better understand current sentiment and associated challenges with PIPP-based production. Specifically, to understand farmer PIPP sentiment and challenges we surveyed commercial PIPP farmers. In parallel we captured stakeholder sentiment toward integrative farming (a slightly broader classification that includes PIPP) via the extraction of selected data (i.e., web-scraping) on the internet. The methodology combines web-scraping of various media and Natural Language Processing (NLP) to understand sentiment in a longitudinal fashion. Here we compared a farmer survey of 53 producers and 2,399 English language web-scraped online posts (e.g., from online forums, news sites, and blogs) from 1,089 unique authors between 2019 and 2021 to understand opinions of and experiences with PIPP. The survey data and online posts both highlighted soil health, animal welfare, and weed control as perceived benefits of PIPP. In addition, survey respondents identified food safety regulations or incidents, processing, and certifications as barriers to PIPP whereas these topics rarely appeared in online conversations (i.e., on-line discourse capturing news and social media). Furthermore, topics mentioned in online conversations but not included in the survey include challenges related to soil, yield, land efficiency, autonomy, and mental health benefits. Comparing online posts with surveys has the potential to offer a more robust approach toward understanding opportunities and challenges among PIPP and other niche agricultural spaces. The parallel approach of analyzing both farmer survey and on-line stakeholder sentiment provides broader context for PIPP farmers and stakeholders with respect to farmer and stakeholder perspective. This approach provides insights that can be useful for PIPP farmers and other agricultural entities with respect to linking surveys and web-based analysis.

The U. S. poultry industry is characterized by its highly efficient, vertically integrated, and consolidated system of production. (Poultry and Eggs - Sector at a Glance | Economic Research Service, 2026). While these conventional production systems provide the vast majority of poultry products, pastured and integrated poultry production (PIPP) based farming systems offer a niche alternative that produces a diverse array of animal and plant based products by including plants and animals on the same plot of land, either simultaneously or in rotation (Nair and Bilenky, 2019; Rothrock et al., 2019; Ofuoku and Ekorhi̇-Robi̇Nson, 2020; Campbell, 2022; Soares et al., 2022). Multiple iterations of this type of farming system exist with respect to food animals (e.g., layers, broilers, and livestock for milk and meat) and plants (e.g., permanent pasture, orchard floor, vineyard alleys, or annual cash crops). Figure 1 shows one such system used in California that cycles between fall broiler production, winter cover crops, spring broiler production, and summer vegetable production. With respect to PIPP, regardless of the type of poultry or crop, management involves moving poultry by harvesting them or rotating them to other portions of the farm before they fully denude the vegetation in any given area. The area where the birds have been foraging is then given time to regrow before animals are allowed to return to forage. Alternatively, the land is prepared for annual crop production. Mobile chicken coops are moved to evenly forage on plant material and spread raw manure (Soares et al., 2022) with the goal of partially displacing the need for supplementary crop nutrients and increasing overall soil quality (Miao et al., 2005; Hilimire et al., 2013; Zhang et al., 2020). This, in turn, allows the land to be used for crop production.

There is a general lack of knowledge about the size of this industry in part because the USDA’s National Agricultural Statistics Service (NASS) does not track pastured poultry or PIPP as a category in its primary production reports. Furthermore, from an academic perspective, only two previous studies have surveyed PIPP farmers with the main goal of understanding farmer-centric challenges, including economic, regulatory, and labor-related issues (Hilimire, 2012; Elkhoraibi et al., 2017). Likewise, there are similar survey-based studies that capture consumer sentiment toward various PIPP-based production systems with perceived benefits to the environment, poultry, and human health alike (Spencer, 2013; Sossidou et al., 2015; Stampa et al., 2020). However, surveys intended for farmers only, or consumers only, may miss important insights about desirable practices that make PIPP preferable. While traditional surveys (defined for the purposes of this paper as data collection associated with structured questions) can provide useful information from various stakeholders, they are typically cross-sectional in design, often with limited sample sizes, and self-reported response biases (Bethlehem, 2010; Wetzel et al., 2016). Specifically, since many survey designs commonly occur once, they do not capture the changing temporal dynamics of a topic, which could be very important given the fast pace of information transfer in the digital age. Typical surveys occur once and have small sample sizes for a few reasons like the large effort in conducting them, declining response rates, and limited contact lists (e.g., university extension circles). There are also biases such as selection bias (Bethlehem, 2010) and response and non-response bias (Sedgwick, 2013, 2014). While cross-sectional surveys have value, they can be supplemented with new longitudinally based data which capture perception from larger participant pools (Reveilhac et al., 2022).

Analysis of on-line media (i.e., news, social media-based sharing of text, pictures, and videos) offers a novel and more open-ended approach toward capturing popular sentiment. Since approximately 72% of US adults use social media (Social Media Fact Sheet, 2021), conversations gathered from these online platforms can represent a larger proportion of individuals in the population than would be feasible with traditional survey tools (Hanna et al., 2011). To that point, web-scraping (defined as the automated extraction of selected data from websites) and Natural Language Processing (NLP) (defined as the general methodology of automated analysis of text-based data) (Nadkarni et al., 2011; Singrodia et al., 2019), offers a novel and robust approach for the analysis of selected web-based material in a highly automated fashion. In addition, associated metadata linked to the web-scraped data can also be used to link NLP based analysis to other variables including geography (Peters, 2013; Vaughan et al., 2017) to understand regional differences in opinion and temporal trends (Houghton et al., 2019) to understand how online discourse is evolving over time.

The combination or comparison of both traditional surveys and web-scraping based analysis has been used and the merits of each compared in various fields including nutrition (Brown et al., 2025), hospitality (Han and Anderson, 2021), and software testing (Raulamo-Jurvanen et al., 2016). In general, this approach broadens access to responses and allows for comparisons in order to better determine behavior, knowledge, and attitude. In order to better capture farmer and stakeholder perspective on PIPP, we used both surveys of PIPP farmers, and a commercial web-scraping tool to gather online posts to compare the differences in content and sentiment associated with PIPP. The over-arching objective was to explore the similarities and differences between data gathered through direct, traditional surveys of PIPP farmers and web-scraped data from farmers, consumers and other on-line stakeholders via the analysis of on-line posts/forums, news sites, and blogs about PIPP. By analyzing the topics and sentiments emerging from the surveys and the online post content, the present study aims to understand areas of alignment, differences in topic areas discussed, and gaps in information, in order to identify research and extension needs that could help support producers and stakeholders interested in PIPP.

Survey data collection

A single online survey (Supplementary Appendix A) distributed during four distinct data collection efforts between March 2021 and February 2023 in the United States was part of a larger research study about integrating organic poultry and vegetable production. The distribution of the survey was aligned with extension efforts associated with the above-mentioned USDA-based grant:

Dissemination 1: The survey was distributed by the Practical Farmers of Iowa and the Iowa Fruit and Vegetable Growers Association. Respondents had between March 11, 2021 and April 9, 2021 to complete the survey. Twenty-six responses were received.

Dissemination 2: The survey was distributed by the American Pastured Poultry Association newsletter on February 16, 2022. The survey was closed on June 1, 2022. Seventeen responses were received.

Dissemination 3: The survey was distributed as a QR code by the Organic Association of Kentucky at a meeting on January 27, 2023, and was closed the following day. Eight responses were received.

Dissemination 4: The survey was distributed as a QR code at the Marbleseed Organic Farming Conference meeting on February 23, 2023. Two responses were received.

The survey was funded by the United States Department of Agriculture (USDA) National Institute for Food and Agriculture (NIFA) Organic Research and Education Initiative (OREI). The purpose of this survey was to understand farmers perceptions of and experiences with PIPP so researchers and educators could better address issues of interest to farmers. The survey included questions about production practices used, farm size, number of poultry raised, perceived benefits and challenges of PIPP, preferred methods for learning about PIPP, and attitudes and beliefs regarding PIPP. All questions were multi-select questions or Likert scale questions, except for those asking for the number of poultry produced, acres/square feet of vegetables produced, and years farming, for which respondents entered a number, and one open-ended question asking for a description of their PIPP system. To increase participation, respondents were given the opportunity to enter a drawing to win one of 12 $25 gift cards (four gift cards in each year). On average the survey took 12.9 min to complete, with a median of 8.9 min. A total of 53 responses were received.

The survey protocol was deemed exempt by Iowa State’s Institutional Review Board (study number 21–081-00).

On-line media data collection

Online posts (e.g., from online forums, news sites, and blogs) related to PIPP were collected using the consumer intelligence platform Brandwatch (Brandwatch Consumer Research, 2026). The platform employs web crawlers—internet bots that aggregate data from a variety of top web-based Application Program Interfaces (APIs). These provided access to data from 100 million online sources and over 1.4 trillion posts on a variety of forums (e.g., Blogs, Reddit, Twitter or X, (henceforth referred to as Twitter), and YouTube discussions, among others) that can be searched via a targeted Boolean query as described below and previously (Sohi et al., 2023). The query was limited to posts that originate from the three major English-speaking poultry producer countries: USA, Great Britain, and Canada, three countries that returned the highest volume of relevant posts.

Because PIPP is not strictly defined by its own unique term but rather it is a set of various practices that different producers may choose to combine in distinct ways, the Boolean query was developed through an iterative process. First, we identified core terms and associated pseudonyms that describe the concept of rotational grazing (foraging), combined with the terms chicken or poultry, and also the term “chicken tractor” which specifically describes a type of mobile housing for poultry. Then, a group of terms describing crop production was added with a series of “AND” statements as described below.

Initial results were reviewed by two researchers and terms were subtracted to reduce noise (i.e., false positive results) from the extracted results. The domain tumblr.com was excluded based on prior experience that its re-blogging feature causes repeated counting of the same content by the web-scraping tool. Based on experience, terms related to the co-occurring avian influenza outbreak (e.g., “flu,” “avian influenza,” “euthanize,” “cull”), which was heavily trending in the news at the time and generated a large volume of outbreak-related posts, were also excluded. Additionally, terms were subtracted from the Boolean query if they generated mainly results that made no reference to poultry husbandry, pastured poultry, or poultry production systems. No formal criteria were applied beyond this categorical judgment, which eliminated the most obviously irrelevant posts but did not necessarily exclude posts that were tangentially related to the topic. Specifically, recipe websites (i.e., food.com, allrecipies.com) were identified for removal by reviewing the domains with the highest ranking post volume. “Ancient Nutrition,” a vitamin brand, was identified for removal by reviewing key topics detected by Brandwatch’s proprietary machine learning-based classifier. While posts were being reviewed throughout this process, additional relevant terms that had not been anticipated in the initial query were added until the researchers judged that further additions were unlikely to result in additional relevant posts. The number of posts reviewed at each stage was not systematically tracked. Through this iterative process, an English-language Boolean string was developed (Supplementary Appendix B).

The web crawler searched for relevant data from between February 28, 2019 through May 2, 2021.

Measurement

Measurements made using survey data included: Percent of respondents raising vegetables, poultry, or a combination of the two; average and median acres operated; average and median number of poultry birds raised; number and percent of respondents using each production practice listed in the survey; and average agreement with statements measuring attitudes and beliefs toward PIPP.

Measurements calculated from the online posts included: number and percent of online posts obtained from each online platform type; number and percent of posts coded into each topic; and number and percent of posts coded as positive or negative sentiment (What is sentiment analysis? | Brandwatch Social Media Glossary, 2024).

Statistical analyses

Survey data was analyzed using SPSS™ (Version 26) using descriptive statistics. Multi select questions were analyzed by calculating the percent of respondents who selected each choice. Likert scale questions were analyzed by calculating the average response to each statement. Numerical answers were analyzed using means and medians. The one open-ended question was coded for themes guided by grounded theory (Heath and Cowley, 2004; Glaser and Strauss, 2017; Thornberg and Dunne, 2019).

Brandwatch’s proprietary natural language processing (NLP)-based tool was used for analyses of the online posts. While the algorithm is proprietary, per Brandwatch, the NLP tool utilizes a supervised machine learning classifier to identify prevalent topics in the text body of the posts. For the full 26-month span of the study, the classifiers performed automatic word clustering, and measured the prevalence of key phrases and terms emerging from the text corpuses of all posts. Brandwatch analysis metrics include measurements of “impact” and “reach” which measure potential impact of an on-line author and the number of unique users who have seen posted content.

In order to organize the online posts into appropriate topics about PIPP, two complementary approaches were used with the goal of categorizing and hence comparing the surveys and the on-line media results against each other. These initial categories were further augmented by a manual content analysis methodology (Bengtsson, 2016) where a subset of the online posts were categorized manually. Specifically, text from these online posts were manually coded by defining variables of interest as defined by a codebook (Supplementary Appendix C). The categories selected included specific features of PIPP, such as impacts on vegetation, animal welfare, or soil fertility.

Selection of the categories was guided via the Brandwatch machine learning classifier and further augmented by the researchers’ preliminary review of the posts. Some categories were added that were not necessarily direct impacts of any production system per se, but which reflected broader values expressed by the users, such as self-sufficiency, or sustainability. Lastly, some categories were added based on the interest of extension practitioners that authored this paper. The directionality of user sentiment attributed to these features were defined by a positive or a negative value of 1, which was part of the Brandwatch analysis. Posts indicating both negative and positive sentiments to an attribute were counted twice, otherwise, volumes do not reflect multiple mentions or a particular attribute in one post. The criteria for each category was defined in a codebook (Supplementary Appendix C) that was used during the coding and coder training process to avoid subjectivity and ensure consistency in coding. Before coding the dataset independently, three coders collectively coded a randomly-extracted training dataset which contained no less than 20% of each content source type, until they reached over 90% agreement on the coding of the categories. Descriptive statistics analysis was conducted via Microsoft Excel (v16).

Findings from the survey data and online post data were then analyzed to understand how the web-scraped data compares against survey data. In order to match the criteria assessed in the web-scraped data to those in the survey, certain survey response options were combined. In cases when multiple survey response options were combined to match a web-scraped category, the survey data was coded to count the number of unique respondents who selected any of the response options aligning with the category defined in the codebook (Supplementary Appendix C), as opposed to counting the number of times any of the survey response options was chosen.

Survey results

We received a total of 53 survey responses (Figure 2). We did not calculate a response rate, because the survey was sent out through third parties and we do not know how many people were on the contact lists at that time.

Using the survey, we were able to characterize farms represented on the survey. Forty-seven of 53 (89%) respondents had raised poultry and vegetables. Four of 53 (7%) respondents had grown vegetables only and two of 53 (4%) respondents had raised poultry only. Overall, 34 of 53, 64% of all respondents, had used PIPP. The average size of respondents’ outdoor production area was 13 acres, with a median size of 1.5 acres. The average number of poultry birds raised was 845, with a median of 125 birds. The survey also gave insight into production practices used on these farms.

Vegetable production practices reported by respondents who indicated they had implemented PIPP included organic but not certified (20 of 34 respondents, 59%), certified organic (10 of 34 respondents, 29%), integrated pest management (IPM, 10 of 34 respondents, 29%), and non-GMO (10 of 34 respondents, 29%). Poultry production practices described by respondents who used PIPP included “day-range” (letting birds forage/scavenge during the day and enclosing them at night) (20 of 34 respondents, 59%) and chicken tractors (19 of 34 respondents, 56%), followed by practicing organic but not certified (16 of 34 respondents, 47%), pasture raised (16 of 34 respondents, 47%), and free range (14 of 34 respondents, 41%), shown in Table 1. Note that these descriptions are not mutually exclusive. A producer using organic practices with pasture could use either day-range or chicken tractor housing, and could truthfully represent their products as pasture-raised or free-range (Negowetti, 2018; Thibault et al., 2022).

With respect to whether responding farms were commercially producing poultry and/or vegetables, the majority of respondents (22 of 34, 65%) indicated they produced poultry primarily for sale, with some family consumption, or exclusively for sale. The remainder produced about 50/50 for sale and family consumption (3 of 34 respondents, 9%), mostly for family consumption with some sales (3 of 34 respondents, 9%), or only for family consumption (6 of 34 respondents, 18%). Starting in the second survey dissemination, respondents were asked to identify whether the vegetables they grew were for family consumption, sales, or a combination. All but three respondents sold at least part of the vegetables they produced. Similar to poultry, respondents most commonly produced vegetables mostly for sale, with some family consumption (11 of 19 respondents, 58%).

The survey asked respondents who had implemented PIPP to identify any benefits they had experienced as a result. The benefits most experienced were improved soil fertility (27 of 34 respondents, 79%), high-quality eggs (26 of 34 respondents, 76%), healthy poultry birds (25 of 34 respondents, 74%) and reduced expenditure on off-farm manure and/or compost (23 of 34 respondents, 68%). In addition, respondents who had practiced PIPP identified challenges they had experienced via the survey. The challenges most experienced by who had practiced PIPP were bird predators (14 of 34 respondents, 41%), low availability of organic feed (10 of 34, 29%), and lack of organic processing nearby (8 of 34 respondents, 24%).

Finally, respondents were asked to rate their agreement with five statements to measure their beliefs about PIPP. Thirty-five of 51 respondents (69%) somewhat or strongly agreed that integrating poultry and vegetable production would save money on feed and off-farm compost or manure. There was identical agreement with the statements that demand for organic eggs is high and demand for organic poultry is high, with 35 of 52 respondents, 67%, somewhat or strongly agreeing with both statements. Twenty-five of 52 respondents, 48% somewhat or strongly agreed that consumers will pay enough for organic poultry and vegetables to make it profitable for the farmer.

Online posts descriptive statistics

A total of 2,399 posts from 1,089 unique authors (determined based on the metadata) made between February 28, 2019 and May 2, 2021, were retrieved and downloaded. Relevant metadata included author, date, country, web URL, and excerpts from the post text where it matched the Boolean string criteria described in the methods. The posts retrieved were automatically categorized by Brandwatch into content source types based on the web domains. Specifically, 40% of the posts were news articles, 27% Forums, 12% were blogs, 8% Reddit posts, 6% Tweets, and 5% YouTube comments. Posts made on review sites such as Yelp constituted less than 1% of the total and were discarded as false positive posts. The top 10 sites with the highest volume of posts are indicated in Table 2, which shows that the majority of the top sites were online discussion forums.

Six of the top 10 websites with the highest volume of relevant posts were online forums (Table 2). This included Backyardchickens.com, a popular hobbyist poultry keeper forum, ranking at #1 with 275 posts and an impact score of 30, which was the second highest impact score of any website. Other popular online forum sites were not poultry or agriculture specific. They included Reddit.com, one of the leading content-sharing online community platforms in the digital world (Rogers, 2017); 4channel.com, an online topic-based forum that allows anonymous posts; uberpeople.net, an online community for Uber drivers; and survivalistboards.com, a community of people who practice outdoor survival skills as a hobby. Online forums (e.g., backyardchickens.com) generated a high volume (n = 656, 27%), high impact (661), and high reach (Table 2) posts.

Online news sites (Table 2) returned the highest volume of posts (n = 960, 40%) pertaining to PIPP. However, collectively, the news sites generated a moderate impact score and had the lowest reach relative to other content types such as forums, blogs, and Twitter (Table 2). Twitter or X, despite producing a relatively lower volume of posts in our study (n = 186, 6%), generated the highest reach (2,265). However, a majority of the reach was generated by a small minority of posts, with almost 40% of the posts having zero reach. Blogs also had a relatively low volume of conversations (n = 282, 12%) but displayed the second highest reach (840). They were tied with Forums for the highest impact score (29).

While generating a low post volume (n = 113, 5%), YouTube ranked 4th in terms of sites that produced relevant posts (Table 2). However, due to the high volume of non-functional URLs and links, missing post body text, and/or irrelevant YouTube results generated by the web scraper, it was not included in the manual text analysis. YouTube’s API did not provide reach and impact metrics.

Machine-generated topic analysis and sentiment analysis

Using Brandwatch’s proprietary topic modeling algorithm, the online posts were grouped into the following topics: “mobile coop,” “moving,” “chicken,” “crops,” “management,” “soil,” “animals,” “vegetable garden,” “build,” “chicken tractor,” “vegetables,” and “pasture.” The terms “cover crops,” “garden beds,” “mobile coop,” and “soil” are relevant key terms that we did not explicitly define in our Boolean query described in the Methods section.

Online users mostly mentioned attributes of PIPP with a positive sentiment. Only 15% of the attributes were mentioned with a negative sentiment in forum posts. This percentage was 5% for Reddit posts, and 1% for Tweets. The overwhelming majority of conversation related a positive sentiment about the attributes (Figure 3B). Most negative sentiments were related to ease of use, animal welfare, and economic benefits (Figure 3B). One percent or less of online posts mentioned food safety, certifications, processing or associated scientific studies.

Manual content analysis of twitter, forums, and Reddit data

Detailed manual review of the online posts determined that 21% of the posts were not relevant for PIPP based stakeholders, but contained related terms or concepts (for instance, discussion of integrated systems with cattle instead of poultry). Seven percent of the posts mentioned PIPP and were deemed relevant, but did not specifically mention any attributes defined in our codebook (Supplementary Appendix C and Figure 3). Manual review of social networking content indicated that the number of posts discussing layer production was almost twice as numerous as posts mentioning broiler production. The attributes regarding soil health, animal welfare, and autonomy were most mentioned in conversations on Twitter, Reddit, and forums (Figure 3A).

Comparison of online data to survey data

When comparing the results of the online data with those of the survey, both methods identified effects on soil and animal welfare as top benefits of an integrated crop and poultry production system (Figures 2, 3). Both mentioned benefits of weed and pest control at similar percentages (Figures 2, 3). The online data identified several benefits of integrating poultry and vegetable production that were not included in the survey, including the perception that this type of production is more natural, sustainable, climate friendly, land efficient, water efficient, supportive of ecosystem services, is anti-industrial, and provides mental benefits (Figure 2). Conversely, the survey identified a few benefits of integrated production at a higher rate than what was found through web crawling. These include cost-effectiveness of integrated production, economic benefits of production, food or egg quality, and increased yield (Figures 2, 3).

Regarding challenges of the integrated production, animal welfare concerns (including predation, environmental exposure and others) emerged in both the web crawling and survey (Figures 2, 3). Both methods identified complexity of use as a challenge in implementing PIPP systems (Figures 2, 3). Fifty-three percent of survey respondents identified food safety regulations or incidents as the top challenge or barrier to integrated production (Figure 2), whereas this topic appeared only 0.2% of online conversations (Figure 3). Processing and certification were mentioned frequently in the survey data (Figure 2), but very little in the online conversations (Figure 3). Additional topics mentioned in online conversations but not included in the survey include challenges related to soil, yield, land efficiency, and benefits to mental health (Figure 3). Negative sentiments expressed toward these aspects of PIPP comprised of 0.5% or less of the online posts.

While on-line based media is one of the primary methods of communication and information gathering in the 2020s, at the research level, cross-sectional surveys still represent the primary method of extension-based information gathering in agriculture. While commonly used in the context of non-agricultural social sciences, web-scraping represents a relatively novel method of information gathering in agriculture. On-line media analysis can help understand social trends and public perception around production methods, consumer values, and effective market messaging. Consequently, learning how best to leverage web-based data extracted via web-scraping and the relationship between those data which can be further organized by topic, sentiment, geography, and time, to more traditional forms of data gathering (e.g., surveys) represents a new approach to better understand opportunities and challenges in a variety of fields, including niche agricultural spaces such as PIPP.

This study is the first to report beliefs and sentiment on PIPP by combining and comparing insights from a total of 2,399 online posts from 1,089 unique authors over 26 months (Figure 3 and Table 2) and a total of 53 farmer-based responses to surveys over 7 non-continuous months based on available contact lists and extension-based in-person meetings with farmers (Figure 2 and Table 1). These quantitative results highlight the advantages of each research method.

Comparison between survey and online post analysis approaches

The on-line analysis allows topic-specific data to be relatively easily extracted and analyzed in a longitudinal fashion. In contrast, convenience and cross-sectional surveys have low response rates among farmers and often require incentives (e.g., gift cards) to increase participation. Using web-scraped information, topics can be categorized and ranked by abundance and/or sentiment, which can be used to modify traditional surveys. However, surveys have the advantage of being able to verify the target audience was reached. In our study the professions of the online post authors are unknown whereas the surveys were from growers who were the intended beneficiaries of our USDA-supported PIPP research. It is likely useful to get insights from both on-line stakeholders and farmers in that the online post analysis allowed us to identify the breadth of beliefs surrounding PIPP (Figure 3), whereas the survey was limited to what we as researchers could identify as potential topics of interest. For example, the potential for PIPP to contribute to “soil,” “animal health,” and “autonomy” was commonly discussed in online posts (Figure 3). While the surveys included questions related to “soil” and “animal health,” it did not include questions related to “autonomy” (Figure 2).

While large language models (LLMs) can facilitate both topic modeling and sentiment analysis we utilized a manual analysis of the online posts for several reasons. First and foremost, due to the proprietary nature of the Brandwatch™ software, running an LLM on Brandwatch’s model could be somewhat redundant and biased as we are not privy to their topic modeling methodology. Hence, we opted to evaluate the categories and systematically create categories to allow for a comparison of the survey and online post data (Figures 2, 3). Nonetheless, LLM could be highly effective at filtering out false positives. In the future, the detailed manual language analysis data generated from this study could be used to train improved NLP based models.

Web-scraping methods present certain challenges for researchers relative to surveys. While open source web-scrapers and NLP-based tools exist, they can be difficult to develop. Pre-developed commercial tools such as Brandwatch™ are limited by license fee and restricted access to proprietary NLP and LLM methods. Consequently, in our research we cannot know how exactly the reach and impact metrics were calculated, preventing full evaluation of the tool’s accuracy or replicability of the results. One additional limitation of web scrapers in general is that they analyze text rather than pictures and video content commonly found on YouTube, Instagram, Snapchat, and TikTok. This gap could be filled by the development of image classifiers (Alam et al., 2022) to capture videos with PIPP based images. It should also be noted that we selected Brandwatch™ in 2019 when there were not as many commercial choices available and developing customized web-scrapers was more technically challenging. Hence, many of the above limitations may not be as significant in 2026. However, new challenges exist with respect to filtering out bots from real-human responses.

It should also be noted that the effectiveness of web-scraping is limited by the precision of the Boolean search terms. This is similar to challenges in survey design where responses are limited by the questions posed— determining what to ask and how to phrase the questions draw on the researcher’s expertise and iterative refinement rather than a fixed protocol. Unlike surveys, Boolean queries can be made extensive, without the practical constraints of respondent fatigue. However, this often leads to high false positive rates. Detecting false negatives is more challenging as the investigator ultimately does not know what posts were not captured. A critical balance is required when refining Boolean searches: minimizing false positives without inadvertently excluding relevant data.

Perceptions of pasture and poultry integrated production

Demonstrating the complexity of the topic modeling approach, the reality is that certain topics crossed over to other topics. For example, while the topic of “weed control” (Figure 3A) typically contained posts describing how poultry are effective at clearing unwanted vegetation, this would potentially also be seen as something positive for “animal health” by PIPP stakeholders. Hence while the topic modeling is important to get a sense of what topics are most discussed, the overall sentiment is likely a good overall proxy for PIPP (Figure 3B). More specifically, every topic area had a higher percentage of posts that had a positive sentiment than negative sentiment. Interestingly, the topic with the highest negative sentiment was “ease of use” (Figure 3B) which was also reflected in the survey where “ease of use” was second to “food safety regulations/incidents” (Figure 2) with respect to barriers.

While the overall sentiment was positive as captured by web-scraping as reflected by a 9:1 of positive/neutral sentiment versus negative sentiment (Figure 3B), of the negative sentiments expressed, ease of use, animal welfare, and economic viability of PIPP were most frequently mentioned (Figure 3) in the web-scraping analysis. Specifically, negative sentiments related to ease of use predominantly occurred in online forums. This could suggest that people were turning to discussion boards for troubleshooting when they had difficulties implementing PIPP methods. Users commonly discussed challenges with moving their poultry flocks for rotational grazing, and improving coop designs for ease of mobility. Other complaints revolved around undesired extra labor, time or logistics involved in executing this type of production system. While the topic of “animal welfare” was typically seen as a benefit via both the survey and the web-scraping, negative sentiments associated with this topic were typically associated with predation, which has been a well-documented challenge associated with pastured and/or free-range systems (Hilimire, 2012; Elkhoraibi et al., 2017).

Interestingly, while social media (i.e., web-based tools that facilitate communication about various content) is known to skew toward promulgating negative opinions (Tsugawa and Ohsaki, 2015), this dynamic was not observed for the topic of PIPP (Figure 3B). Across all platforms, conversations tended to focus much more on the beneficial attributes rather than barriers related to the key attributes studied. For example, with respect to the topic of “animal health” online users state a belief that chickens are happy and healthy when allowed to forage and scavenge. They report their usage of mobile chicken coops as a way to allow birds to forage on pasture, while protecting flocks from predators. Likewise, looking at individual posts associated with the topic of “autonomy,” (which was not included in the survey), people related their desire to increase self-reliance in the face of supply chain disruptions, high food costs, and mistrust toward the food system as motivators for PIPP. It is not possible to determine whether people see this sentiment as a motivator to raise their own poultry and crops, or to only purchase and consume these products versus other commercial production systems.

Some sense of motivation can likely be gleaned from the platforms utilized. Forums such as “4-channel” are more likely to be used by survivalists, and forums like “backyardchickens” are likely used more by “homesteaders” (Table 2). In addition, it should be noted that the topic “mental benefit” (Figure 3) had terms such as “happiness,” “wellbeing,” and “satisfaction” with respect to PIPP. In summary, as the U. S. struggles to recruit beginning farmers, “autonomy” could be an area that could be further highlighted by various programs such as USDA’s Beginning Farmer and Rancher Development (BFRDP) program which focuses on recruiting and assisting beginning farmers.

Doubts and concerns around the economic viability of PIPP were captured using both the web-scraping and via the survey (Figures 2, 3), with some users arguing that such farming systems cannot be profitable or provide a reliable source of income. A minor share of concerns was expressed around the expenses of implementation. This may indicate that the economic challenges were perceived, by many, to lie in access to market or marketing, rather than cost of entry. It was unclear, in many cases, whether the users were speaking from firsthand experience, or merely for the sake of theoretical debate. However, the economic challenges of PIPP-based farming is well documented both in the literature (Angioloni et al., 2016; Elkhoraibi et al., 2017) and anecdotally.

Similarly, the topic of animal health ranked high as both a benefit and a barrier. We speculate some producers may have experienced chickens being injured by chicken tractors or killed by predators, which are obviously negative experiences in the realm of animal health. On the other hand, the perception that PIPP allows poultry to engage in more natural behaviors and access a more varied diet than another production systems may have contributed to positive sentiments related to animal health.

Discussions around relevant scientific research and findings, food safety, poultry diseases, processing, and certification were minimally present or absent in the web scraped data set. Several of the top sites, such as backyardchickens.com, survivalistboards.com, beginningfarmers.com, and helpfulgardener.com are targeted toward hobby producers and those with a lower level of experience. Therefore, it is possible to speculate that this demographic may be trialing PIPP production at such a small scale that they are less concerned about certification or processing.

Limitations of study and lessons learned

While the survey data and web-based data provide insights as described above, it should be noted that the survey and web-based analysis were collected during two different collection periods (March 2021 to February 2023 for the survey and February 28th 2019 and May 2nd 2021 for the web-scraped data). Ideally, these data would be compared at the same time to avoid the confounding variable of time. That being said, the continuous nature of web-based data compared to the point in time nature of surveys makes comparisons challenging.

In summary, the ability to extract and analyze online posts is growing in importance as a greater percentage of Americans get their news and information from on-line media (Forman-Katz and Matsa, 2022). In addition, because the analysis of on-line media can be continuous, insights can more easily be gained with respect to trends, topics, and sentiment. In contrast, surveys offer a targeted approach with respect to audience and question type. However, surveys in agriculture often have low participation rates and are not longitudinal. This use case demonstrates the potential synergy of surveys and web-scraping to facilitate a more holistic approach toward understanding a specific production system (e.g., PIPP). Conceptually, web-scraping facilitates more of a “hypothesis generating” approach as opposed to surveys which require assumptions about the surveyed audience. This approach is likely highly useful for poorly studied topics such as PIPP.

Further improvements in AI with respect to images and video-based analysis will likely provide a new layer of continuous stakeholder data that could be analyzed in parallel with text-based on-line media. This ability to continuously extract and analyze web-based data is perhaps the most relevant aspect of utilizing this approach as it can facilitate continuous assessments as opposed to a point in time survey.