Abstract

Using artificial intelligence (AI) in local journalism is more likely to yield civic benefits if it begins with the culture and needs of the community. AI should be used not simply to reproduce traditional journalism more efficiently but to improve residents’ lives by providing useful information and enhancing local representation. Those hoping to use technology to “save” local journalism should shift the discussion about AI in local news away from efficiency and replacement, and toward highlighting the concerns and observations of community members. Drawing on theories of service journalism, civic AI, and political accountability, I argue that adopting AI tools in local journalism should go hand in hand with implementing more equitable, civically-focused, participatory forms of journalism. This approach is exemplified by the “Documenters” program from City Bureau in Chicago, which pays regular residents to take notes at government meetings and uses those notes as the basis for local news. Using several recent examples and a chatbot trained on Documenters’ notes, I show how these participatory approaches offer several concrete civic benefits that AI and traditional journalism alone cannot provide. The AI-assisted journalism of tomorrow should not look like a propped-up version of the stories, narratives, institutions, and inequities of the past. AI should shape and accelerate journalism’s transition, which is already underway, into a force for promoting civic empowerment and political accountability.

I. Introduction

Amid growing concerns about misinformation and democratic decline, few actions could do more to bolster American democracy and restore trust in information than strengthening local news. Greater exposure to higher-quality local news leads to more political participation (Shaker, 2014), reduced polarization (Darr, Hitt and Dunaway, 2021, 2018; Moskowitz, 2021), less corporate malfeasance (Heese, Pérez-Cavazos and Peter, 2022), healthier municipal finances (Gao, Lee and Murphy, 2020), and increased political competition (Rubado and Jennings, 2020), among other positive effects.

Unfortunately, local news finds itself in an unprecedented and existential crisis. Local television news viewership fell nearly 25 percent from 2016 to 2022; weekday newspaper circulation fell 32 percent from 2017 to 2022, dropping from 30 million to 21 million; and an estimated 1,800 newspapers closed nationwide from 2004 to 2018 (Abernathy, 2018; Lipka and Shearer, 2023). Reporters are being laid off at record rates, with newsroom staffing declining by over 25 percent from 2008 to 2020 (Walker, 2021). The financial model that once sustained high-quality local news relied on local advertising, but technological change has largely dismantled this system: Craigslist killed classified ads (Djourelova, Durante and Martin, 2024), and advertising revenues have plummeted from a peak of $49 billion in 2006 to under $10 billion in 2022—a decline of more than 80 percent (Pew Research Center, 2023a). As the old model collapses, it is unclear what the future of local news holds, what its next incarnation should look like, and whether it will have the capacity to continue to produce civically valuable coverage.

As in so many industries, the emergence of AI poses both opportunities and risks for the local news industry and civic information landscape. The journalism industry is working toward a positive vision of that future, where generative AI could reduce reporting costs, quickly synthesize large amounts of government and industry data into accessible resources, and save journalists time and effort by automating some reporting and non-reporting tasks. However, using AI solely to make traditional local journalism practices more efficient would be a missed opportunity, both to grow future audiences and to break destructive historical patterns. Local news coverage has consistently marginalized disadvantaged groups, prioritizing affluent areas with more potential paying customers while relegating other areas to crime coverage and other negative portrayals (Usher, 2021). Journalists also tend to focus on policies that might pass or have already passed, rather than on how residents might actually use the policies already in place (Thorson, 2024). Uncritically adopting “time-saving” AI tools could exacerbate these tendencies, leading to even more inequitable coverage and conflict-driven, partisan, horse-race-style political analysis.

The risks AI poses to democracy mirror the risks it presents for American governance: “autocratic centralization,” where power is concentrated in the hands of too few, such as an overly powerful federal government or a small number of tech companies, or “ungovernable decentralization,” where access to powerful AI tools becomes too diffuse, inconsistent, and unencumbered across localities (Ovadya, 2023). Despite these risks, the potential gains from increased efficiency during a period of devastating layoffs and shrinking revenues are encouraging journalism organizations to adopt AI to reinforce traditional forms of local journalism (Associated Press, 2023b).

In this essay, I argue that the discussion about using AI in local news should be shifted away from journalistic efficiency and replacement, and toward using AI to establish the concerns and observations of community members as the foundation of community-focused news. Drawing on theories of service journalism, civic AI, and political accountability, I outline why implementing more equitable, civically-focused, participatory forms of journalism (Green, Holliday and Rispoli, 2023) would make the adoption of AI tools in local journalism much more civically valuable. As exemplified by City Bureau’s “Documenters” program in Chicago, which pays regular residents to take notes at government meetings and then uses those notes as the basis for local news (Documenters.org, 2025), these participatory approaches provide several concrete civic benefits that neither AI nor traditional journalism alone can achieve.

I argue that the future of AI in local journalism is more likely to yield civic benefits if it begins with the culture and needs of the community, aiming to improve locals’ lives by providing useful information and improving representation rather than merely reproducing the “grunt work” of traditional journalism more efficiently. The AI-assisted journalism of tomorrow should not look like a propped-up version of the stories, narratives, institutions, and inequities of the past. The frankly inevitable incorporation of AI into local journalistic processes can take a different path, and build upon a foundation of participatory journalism aimed at government accountability to amplify that impact and improve civic outcomes.

II. A Theory of Local Journalism, AI, and Politics

The decline of American local news continues to raise concerns among journalists, politicians, and others invested in the nation’s civic health. Despite its recent decline, local news is viewed more favorably than other forms of information. For example, public trust is approximately 10 points higher for local news than for national news (Peterson et al., 2024). While technological innovations such as Facebook and NextDoor have made community-level discussion groups more accessible than ever, the public continues to view local journalism as more interesting, relevant, and trustworthy (Le Quere, Naaman and Fields, 2024). Those who read local news are more informed and more likely to be involved in their community than those who do not (Moy et al., 2004; Shaker, 2014). Without local news, people turn to national news about partisan conflict that can polarize and demobilize audiences (Darr, Hitt and Dunaway, 2018; Ellger et al., 2021; Hersh, 2020).

At the same time, local journalism is not always a pure civic good. It has historically excluded, demonized, and marginalized communities of color, poorer neighborhoods, and immigrant populations (Bedingfield and Forde, 2021). Local media often chases subscribers in wealthier, whiter neighborhoods while applying “if it bleeds it leads” logic in its coverage of poorer, higher-crime neighborhoods—patterns that exaggerate differences between communities. Disadvantaged areas are frequently portrayed as dangerous places, with no additional context about the lives and needs of their residents (Heider, 2014). As a result, Black Americans are more likely than white audiences to perceive local media as unfair to them and their communities (Pew Research Center, 2023b). White readers, on the other hand, increasingly find local news catering to their communities and interests as newspapers shift from advertising-based to subscriber-based revenue models and target those willing and able to pay (Usher, 2021).

The question of how AI can assist with local journalism should therefore take into account the journalistic approach of each news outlet; the economic, social, and political viability of that approach to news; and whether AI tools are better suited to streamline existing tasks or to reimagine what is possible in “local news.” The news values that journalists traditionally use to select and frame stories—such as novelty, conflict, and personalization—do not always help people navigate their government or strengthen democracy (Shoemaker and Vos, 2009). These values also shape how policy is covered, pushing journalists to emphasize new proposals, focus on personalities and prominent politicians, and evaluate politics and policies through a binary “game frame” of winners and losers (Lawrence, 2000; Zaller and Chiu, 1996). The following sections assess whether alternatives to these problematic traditional news values are sustainable, how these alternatives might help citizens navigate their government more effectively, and why serving local communities remains a challenge for both the local news industry and AI developers.

The economics of public journalism

Journalists have long faced a fundamental tension between objectively reporting on community debates and actively trying to improve their communities through “civic” or “public” journalism practices (Rosen, 1999). Public journalism defines democracy as an action, carried out by people, centered on public problem-solving and public deliberation around common issues. Many of these principles can be found in today’s efforts to retrain journalists, including those from Solutions Journalism Network, Good Conflict, and Trusting News, among others (Robinson, Orozco and Darr, 2025).

Participatory journalism aims to involve the communities traditionally ignored by local news, empowering them to observe government directly to improve both representation and engagement (Green, Holliday and Rispoli, 2023). The reorientation of journalism from gatekeepers of information dissemination to facilitators of information gathering is still the exception within the universe of local news outlets. Many traditional journalists remain unfamiliar with this transition, though the collapse of traditional economic models has made more news organizations willing to reconsider their approaches (Singer et al., 2011). This slowly shifting mindset led many in the journalistic community to think in terms of how AI could benefit journalism, instead of how journalism norms and routines can make AI better.

Philanthropic dollars are currently supporting experiments like these, enabling them to remain nonprofit (Green, Holliday and Rispoli, 2023). The American Journalism Project, a “venture philanthropy” fund for local news, has raised nearly $240 million and funded or founded 50 local news sites nationally. The PressForward initiative, backed by a coalition of funders, has pledged $500 million to support local news startups across the country (Anglin, 2025). These philanthropic efforts are helping to build a rich new ecosystem of predominantly nonprofit news sources, structured around principles and economic imperatives that differ from those shaping traditional, commercially driven news values. Overlooking this emerging perspective when designing AI tools that work with local journalism would fail to capture the breadth of journalistic innovation, and the support behind it, at this crucial moment.

Media, policy, and civic empowerment

Today’s participatory journalism focuses on optimizing both the inputs and outputs of citizen-government interactions. American policymaking occurs at multiple levels (state, local, national), often simultaneously, which can be confusing and intimidating. Information that lowers administrative burden on residents and illuminates how policies actually work can change minds, build resistance to misinformation (Thorson, 2024), and increase the chances of policy uptake. Traditional political journalism does not always provide information that reduces administrative burden (Rosen, 1999). According to recent work by political scientist Emily Thorson (2024), there is far more policy potential (i.e., proposals and ideas) and policy outcome coverage in most media than exploration of current policy. Journalists’ focus on what might happen and their interpretations of what has happened leave people vulnerable to misinformation and partisanship, whereas providing “policy-current” information helps reduce misperceptions across partisan lines and improve the accuracy of policy knowledge (Thorson, 2024). People want to know how their government works and what policies are available to them, but are often underinformed or misinformed because they have incorrectly ‘filled in the blanks,’ not necessarily because they were deliberately misled. When presented with accurate information, they update their beliefs accordingly.

AI reporting and writing tools based on traditional journalism are therefore likely to focus on the potential and outcomes of policy rather than providing practical context and guidance for navigating government in its current form. While traditional frames may serve the politically interested, they continue to portray politics as a game and policy proposals as promises rather than actions. Such coverage could alienate those who most need access to policy options: financially disadvantaged residents in historically marginalized communities. AI tools trained on past journalism could inadvertently prioritize this sort of unhelpful coverage in ways that journalists may not notice—because they are steeped in the values that created it.

Inequality, community, and AI

Creating a community-focused product poses similar challenges for both journalists in those communities and AI chatbot designers. Both professions share a learned disposition and the goal of taking a “view from nowhere,” acting as dispassionate observers of communities and information rather than as participants who assist and shape them (Rosen, 1999). Scientific and journalistic objectivity, though valuable in many ways, encourages generalizable methods and orientations but can make it particularly difficult to understand (hyper)local communities. In political journalism, this often manifests as overly generalized partisan framing and “game framing” of political situations, even when the factors shaping those situations are intensely local. AI researchers seeking to pool large training data sets across contexts may similarly miss the nuances essential for understanding why something matters, how things actually get done, and the local dynamics of data collection (Hsu et al., 2022), while also risking the inadvertent reinforcement of existing inequalities (Noble, 2018).

Regional differences in culture, resources, and context make it difficult to design AI systems that address locally specific concerns (Hsu et al., 2022). The hyperlocal knowledge and values that define a community are hard for nonlocals, such as Report for America-style outsider journalists, to understand (Wenzel et al., 2020). Even when trying to engage community members to understand local culture, both journalists and AI designers can inadvertently create elitism and power dynamics that make locals feel like subjects rather than collaborators (Hsu et al., 2022). Although it can be challenging for both journalists and AI scientists to fully immerse themselves in the culture and concerns of a community, collaborating with community members can improve the quality of data while increasing locals’ civic confidence and sense of ownership in the information produced (Hsu et al., 2022).

Both traditional journalism and AI tools, therefore, risk failing to capture the culture of the community they serve and perpetuating local inequality without careful consideration of their inputs (Harrington, Erete and Piper, 2019; Noble, 2018). As traditional local journalism declines, the tendency to serve the “rich, white, and blue” in their communities—i.e., potential subscribers with high political interest—may be amplified by AI tools (Usher, 2021).

For instance, early experiments using AI-generated news in real estate trend reporting can produce articles using data that generate substantial engagement but are difficult for reporters to produce at scale (Press Gazette, 2021). This sort of reporting can be automated using publicly available real estate data sets, and is most valuable for consumers who own homes or can afford them—precisely those who are already overrepresented in government meetings (Einstein, Glick and Palmer, 2019). This unequal participation is not merely due to the accessibility of in-person meetings: even after many meetings moved to Zoom—a technological advancement (much like AI) that could have lowered participation barriers—homeowners continued to dominate local meetings and largely opposed new housing developments (Einstein et al., 2023). If AI automation only serves those who are already overrepresented, and whose interests block policies like building homes that could reduce inequality, it will only deepen societal divisions and reinforce existing hierarchies—even as it makes money from the relatively wealthier subscriber base it is prioritizing.

AI and local journalism: More of the same?

The use of AI tools in traditional local journalism is well underway, including a prominent partnership between the AP and the Knight Foundation in 2023 (Associated Press, 2025). These projects were explicitly designed to “support AP and local newsrooms by automating repetitive tasks, which allows journalists to focus on more impactful aspects of their work and reporting” (Associated Press, 2025). Two of these projects illustrate the how AI is being used by journalists to make their current work more efficient.

The first, from the Brainerd Dispatch in Minnesota, developed AI tools to automate the writing of stories from items in police blotters. Although this approach aimed to free reporters to add more detail to their crime coverage, which could be valuable, the net result would undoubtedly be more coverage of crime, potentially perpetuating stereotypes and deepening divisions, even if such coverage appeals to some audiences (Associated Press, 2023a).

The second, at Michigan’s NPR affiliate WUOM-FM, most closely resembles the Documenters program described in more detail below: using publicly available video of city council meetings statewide to generate a searchable database of transcripts to “decrease the amount of time the newsroom spent on attending meetings in person” (Associated Press, 2023b). The promise of this tool is clear: reporters can skim transcripts for more meetings than they could attend in person, without needing to watch all the videos. There are also risks, however: By making meeting attendance optional and setting up alerts and keyword searches, this AI system risks limiting accountability to only meetings where reporters anticipate news to be made. Unexpected or buried items that may have been seen in their proper context by someone in the room may therefore go unnoticed, particularly since this method is limited to meetings with publicly posted videos. The accountability function also deserves some additional thought, since it does not merely matter what information reporters can catch: it also matters whether officials feel like they are being observed, and we do not know if having someone in the room is interpreted differently by those officials is interpreted as more or less surveillance of their activities than the passive combination of video and AI-generated transcripts.

These are just two examples of how AI might perpetuate the biases and perspectives of traditional reporting without putting additional pressure on local governments, but they are powerful and well-funded ones. The conversation around AI in journalism should not stop at approaches like these. In the following sections, I will introduce a current and successful model of participatory journalism—the Documenters program—and present examples of how this system might be more civically beneficial and highly valued by a wide range of audiences.

One Participatory Approach: The Documenters Model

In this project, I focus on a rapidly growing new model of philanthropy-supported participatory journalism: newsrooms based on a collaboration with a participatory journalism program known as Documenters. Documenters was started by City Bureau in Chicago in 2018 and has spread to over 30 cities (Documenters.org, 2025). It is based on a simple idea: pay local residents to attend local government meetings, take notes, and pose questions that journalists or their fellow citizens might be able to help them answer (Documenters.org, 2025). As such, Documenters provides a rare form of news: information about what government is actually doing and when, framed around the observations and questions of community members who are neither journalists nor affiliated with particular interest groups.

By sending regular people into meeting rooms and financially incentivizing them to pay attention and ask questions, Documenters empowers new perspectives in the newsgathering process and focuses on the policy-current information described above (Thorson, 2024). The Documenters program’s focus on policy-current information at the hyperlocal level, where much of American policy is formed and implemented, has the potential to increase civic knowledge and fight misinformation, provided it reaches readers.

Documenters are an essential part of newly opened nonprofit newsrooms supported by philanthropy organizations, such as Mirror Indy in Indianapolis and the Signal Ohio newsrooms in Cleveland and Akron. The willingness of philanthropists to support Documenters shows their openness to new approaches to newsgathering and story creation, as well as the freedom that nonprofit status provides to news organizations looking to break free of the economically determined inequalities and biases of the industry’s past.

III. Why Participatory Journalism Can Improve AI Journalism

The promise of basing local journalism AI projects on participatory models such as Documenters can be shown in three ways. First, participatory journalism reinforces open meetings laws and ensures human presence at government meetings. Second, these practices offer value to communities by examining local policies through the eyes of the people affected by them, rather than from the perspective of professionals seeking a journalistic angle. Third, by training locals in note-taking and standardizing formatting and procedures across cities, participatory programs like Documenters create conditions for developing more effective and civically valuable AI tools.

Preserving journalistic infrastructure through observation

It cannot be assumed that government will continue to make itself transparent and available if there is no obvious demand for it to do so. Open meetings laws are a common feature of state and local governments whose constituents expect transparency, but supposedly ‘open’ meetings can be difficult to locate online or in person. Documenters provides the valuable service of scraping city government websites to create a standardized list of past and upcoming meetings for city councils, municipal boards, ward committees, school boards, county boards, and other local government fora. For the most part, these meeting schedules are not well-centralized: instead, they need to be collected from across multiple websites that often have conflicting data formats (cityscrapers.org). A central list of public meetings is helpful for citizens, journalists, and others interested in good government in and of itself, even without the notes posted by Documenters. While public meetings laws require cities and counties to publicly disclose meeting schedules, in practice, this is often fragmented across websites, social media, and various formats, including hard-to-find and difficult-to-use PDFs.

Stories from across the Documenters Network illustrate the power of having people in the room to uphold open meetings laws, ensuring that the raw materials for citizen engagement remain available.

San Diego. On July 23, 2024, the Documenters-affiliated inewsource reported that City Council President Sean Elo-Rivera was proposing to end virtual public comment through phone or Zoom, a practice that was introduced and adopted during the COVID-19 pandemic (Niebla, 2024). Other cities, including Los Angeles, had dropped these options for public participation because the Brown Act, California’s open meetings law, does not strictly require them (Dale, 2024). The San Diego Union-Tribune picked up the story (Garrick, 2024), and Elo-Rivera withdrew his proposal after hundreds of public comments opposed the plan.

Newark. City Bureau summarized the situation perfectly: “At an Essex County Board of Commissioners meeting in New Jersey, Newark Documenters Program Manager Hayat Abdelal arrived to find locked doors and no signs directing her to the meeting location, even though it was listed online. ‘Many people could have made it up the long steps of the building, just to turn home when finding the doors locked,’ she shared. This kind of story is all too common in Newark, where public meetings are canceled at the last minute, lack agendas, or are physically inaccessible. That’s when New Jersey Documenters, powered by the New Jersey Alliance for Immigrant Justice, stepped in. Thanks to months of consistent coverage and some well-placed public pressure, they helped turn Newark’s empty ‘no upcoming events’ page on the city’s website into a full meeting calendar.” (Dikcis, 2025)

The open public meetings system requires public participation and regular maintenance, including occasional pressure when government officials do not live up to the intent of the law. It is very difficult for a politician to ignore a request to participate and observe from a constituent. If journalists rely only on meetings that actually occur and for which video is posted, as in the Michigan example (Associated Press, 2023b), they may become victims of open meetings laws violations, as they may not find out about the meetings that do not occur or video was not posted. Using AI to maintain a reliable public schedule, alongside a robust participatory journalism program to monitor compliance, can improve civic representation and accountability far more effectively than reporters relying on a limited set of meetings and predetermined keywords.

Political influence through observation

Local media can also serve an important purpose that depends less on readers and more on the perceptions of elites. Politicians behave differently if they are being watched. In Congress, representatives stand witness before fewer congressional committees, deliver less funding to their districts, and vote against their party less often when local news coverage in their district is weaker (Snyder and Stromberg, 2010). State legislators are more engaged when their local news is stronger: they miss fewer roll-call votes, sponsor more bills, and serve on more powerful committees (Myers, 2025). These incentives should also hold within local government. Even if local news consumption continues to decline, an AI-powered local information-gathering infrastructure could help preserve some of the benefits of ‘watchdog’ journalism. Politicians may still behave more accountably if they know their actions will be noticed and amplified, even if not to the same extent as when constituents are watching in person.

In several recent examples, Documenters-based newsrooms strengthened representation by being in the room for meetings and following up on their observations.

San Diego. The accountability came a little too late in 2024, after Measure E (a sales tax increase) failed. A Documenter at the Library Board of Commissioners meeting the day after the election noted that the library system was facing 10 percent cuts if the measure failed (Bowman, 2024). This would have been valuable information for voters before the election, when the measure was described as supporting “infrastructure, maintenance and construction projects,” not library support (Bowman, 2024). This context may have changed how voters saw the measure, which narrowly failed (50.4% to 49.6%), had they seen it before the election (Keatts, 2025).

Detroit. Acting on a tip from a local, Outlier Media and Detroit Documenters took a proactive approach to a pressing policy issue: “While more than 11,000 Detroit homeowners lost their homes to tax auctions during that period, per Alsup’s analysis, 2,400 is the number of homes sold for more than the tax debt owed—and because of a state Supreme Court ruling, those former owners could be eligible to claim those profits back” (Culpepper, 2025). Since initial claims needed to be filed by a certain date, Detroit Documenters called those who were eligible and coached them through the policy process to recoup what they were owed (Levin, 2025). This active, community-focused approach to delivering policy information started with a local tip but relied on the infrastructure of participatory journalism to make a significant and tangible impact in the lives of Detroiters who needed the help.

In both cases, Documenters helped clarify a potentially overlooked policy implication. While this occurred too late to make a difference in the San Diego example, the Detroit Documenters were in place and ready to help people find the money they were rightfully owed. Mainstream journalists were either unprepared to take this action or did not find the story sufficiently newsworthy. In this case, the context and the solution depended on the engagement and initiative of human non-journalists, making participatory journalism an excellent fit.

Practical advice through observation

The Documenters project opens the possibility of training a chatbot on firsthand accounts of government meetings by non-journalists interested in how government operates, potentially producing more actionable information and reducing administrative burden. The development of AI chatbots focused on local government is important, given recent research showing that biased AI models can influence political decision making (Fisher et al., 2024). Researchers developed biased AI models targeting both liberal and conservative participants and found that, using either opinion-based or behavioral measures (a budget allocation game), individuals from both parties were swayed in the direction of the AI’s bias. These clear political consequences raise the stakes of using AI to explain government, particularly when combined with findings about the persuasive potential of policy-current information (Fisher et al., 2024; Thorson, 2024). Given the volume of consequential policy passed at the city level and the significant variation among local governments nationwide, it is particularly important to scrutinize the data used to train these chatbots and how the data were produced.

To assess these ideas, I created two chatbots using OpenAI’s ChatGPT+ feature, as described below. Their answers to a common set of questions can help assess the civic value of these potential data sources.

What Documenters observe (DocGPT). This data set includes three months of Documenters meeting notes from 2024, comprising 100 local government meetings. These committees include City Council, Building Standards, Board of Control, Zoning Appeals, and more.1 1. To create a Documenters-based civic chatbot, I uploaded 100 documents containing meeting notes—the most recent three months, from January to December 2024—into ChatGPT+, a subscription service from OpenAI. These notes covered a variety of meetings in Cleveland and Cuyahoga County government, as notated by Documenters. It was not a complete set, but rather a variety of government agencies that Documenters were “assigned” by the system and their editor, based in Signal Cleveland’s newsroom. Documenters notes are consistently formatted into several sections: “Summary,” “Follow-up,” and “Notes.” These respectively provide an overview of the meeting, address any follow-up questions the note-taker (who, again, may be unfamiliar with how the meeting typically works) may have, and list their more detailed notes from the meeting. Previous versions used examples from three chatbots, including one based on Signal Clevelandarticles, which were substantially similar to DocGPT and omitted for clarity and brevity. These outputs are available upon request. See Figure 1 for an example.

What is publicly available (ChatGPT). The City of Cleveland posts resources about its activities and policies, and under the leadership of Mayor Justin Bibb, is updating many of its electronic resources to make them more accessible to residents. Non-paywalled news sites are also included in a general ChatGPT search for guidance on navigating and understanding Cleveland’s municipal policies.

I then asked both chatbots the following question—“What advice would you have for how to obtain an extension request for a building code violation?” I included the following language to direct DocGPT to use only information from the meetings notes:

The purpose of this chatbot is to answer questions residents of Cleveland have about the actions of local government. The only source material used should be these Documenters’ meetings notes that I uploaded: nothing from the web or other. Answers should help users understand how to navigate policies, what the important debates are, who are the government officials that are working on these, the differing responsibilities of Cleveland and Cuyahoga County, and so on. In each answer, please cite the meeting and date that the information was pulled from.

Figure 1: Example of Documenters notes from a Cleveland City Council meeting, October 18, 2024.

Results: Chatbot advice

The full responses from both chatbots are included as an appendix to this paper. ChatGPT produced an answer entirely about the formal process: file appeal, wait, stay in communication. DocGPT offered more useful local knowledge based on the events in the relevant board meetings. It made a strong recommendation to “Show Up to Your Appeal Hearing,” supported by a claim that the board “has denied multiple extension requests from property owners who did not attend their meetings ... if you request an extension but fail to appear, your request is likely to be denied.” It added that the board “frequently denies extensions for property owners who fail to appear at their hearings.” The DocGPT answer highlights the stakes most clearly in its first point: “Key meeting insight: Many appeals were denied automatically if the appellant failed to appear before the board.” Crucially, it suggests an action: “If you cannot attend, notify the board in advance to reschedule.” This makes the process seem manageable as well as mandatory, which might lower psychological barriers to participation and administrative burden.

DocGPT directly addresses another possible obstacle—financial difficulties—left unmentioned by ChatGPT. DocGPT spotlights this concern as a separate point at the end of the answer, adding significantly more detail and making the issue seem more common:

Helpful Tip: If you’re facing financial difficulties, ask whether the city offers any financial assistance programs to help with repair costs. This was a recurring concern during board meetings.

By framing financial difficulty as a frequently discussed concern that can be addressed directly with the city, DocGPT significantly lowers the psychological barriers to accessing the city’s financial hardship waiver policy. If individuals who find it difficult to take time off work to attend their hearing are informed about both the (informally) mandatory nature of attendance and the common use of accessible financial hardship waivers, those with the fewest resources may be more likely to receive an extension to a more convenient date. Relying solely on ChatGPT would have contributed to a vicious cycle that perpetuates inequality, whereas DocGPT provides clearer guidance, helping people understand what they need to do to keep their property.

The results of these chatbot tests, while preliminary and anecdotal, show that DocGPT provides more useful, policy-current information than ChatGPT, despite ChatGPT’s much larger potential information base. DocGPT used the notes and observations of non-journalist residents attending city government meetings to provide the sort of “insider knowledge” about informal norms that is especially useful for residents unfamiliar with local policies. The interactive nature of chatbots can theoretically improve how local government serves its citizens at low cost, but in practice, the inputs appear to be crucial.

The benefits of participatory journalism

This section details several recent examples of participatory journalism that led to improved civic outcomes unlikely to be achieved by either AI tools or traditional journalism. Participation empowers participation by actively enforcing open meetings rules, including essential factors such as public comment and meeting accessibility. Careful attention to detail and a strong focus on assisting disadvantaged residents produced valuable policy context in San Diego and prompted action to address a frequently overlooked policy in Detroit. Finally, the Cleveland example illustrates how a chatbot trained on Documenters notes can provide more precise and useful information, and potentially better outcomes, for locals compared with ChatGPT.

Designing AI tools that incorporate materials and lessons from participatory journalism rather than traditional journalism could solve several important problems. Participation pressures government to comply with its open meetings laws, generating more data that can be used to train AI to better reflect the public voice. Participation also provides the local policy context necessary for acting on community-specific issues and building community trust, both of which are uphill battles for AI-based tools that resist localization. AI could also build upon the materials produced by local participatory journalism, such as Documenters’ notes, which are written from the perspective of locals, consistently formatted across all Documenters cities, and provide more useful and practical advice than government websites alone. AI is efficient at collecting meeting transcripts (when those meetings exist), isolating keyword-specific sections of meetings (assuming journalists know what to search for), and synthesizing procedures for navigating government systems (but lacking the local knowledge to help locals succeed within these systems). Participatory journalism addresses these gaps particularly well, and should be both supported by (and supportive of) AI whenever feasible.

IV. Discussion and Limitations

This paper is a statement of theory and an initial attempt to answer these questions, not a complete data set for causal analysis. A broader set of tests on Documenters-based chatbots, from some or all of the 20-plus cities where Documenters are active, would be necessary to determine whether the civic usefulness observed here is generalizable. Ideally, this would be paired with a large-N survey experiment in which users test ChatGPT alongside their local DocGPT, rating each for usefulness, informativeness, and trustworthiness.

This study’s major limitation is its limited scope and small sample size. Within Cleveland, collecting more requests for advice would enable a more robust comparison of the two GPTs tested. Interviews with Cleveland city officials and residents could provide greater detail and context about the accuracy and usefulness of answers provided. Finally, paywalls protecting legacy local media make it difficult to assess what a GPT trained on their archives might look like. While these organizations may wish to develop their own chatbots, they should be aware of the potential dangers described above and incorporate structured, reliable participatory journalism whenever possible.

V. Conclusion

The crisis of local news decline, which has the potential to further nationalize politics and deepen partisan divisions (Darr, Hitt and Dunaway, 2018; Hopkins, 2018), also presents an opportunity to remake local news and address problems of the past. There is immense civic potential in providing policy-current information by answering the questions of non-journalists seeking to understand and effectively navigate their municipal and county governments. AI could be a scalable and efficient way to deliver and summarize those answers in an actionable format. This type of information requires breaking free of old patterns of journalistic logic that find policy-current information less “interesting” than crime news, horse-race coverage, or political analysis.

Even as a new generation of civic-minded journalists engages more with their readers, journalistic perspectives cannot fully capture the public’s practical concerns and questions (Robinson, 2023; Wenzel, 2020). The training materials provided by Documenters—a comprehensive database of local government meetings, many of which also include notes from regular residents—could be the basis for a valuable AI-based contribution to local news.

The future of local news need not look like an exaggerated version of its past. Historically disadvantaged communities could serve as partners in creating a new future for local news. Instead of perpetuating a legacy of harmful practices, many new local news models are choosing to engage with these previously alienated populations to expand their audience and impact (Robinson, 2023). If AI-based local journalism is to offer a distinctive and more civically valuable product, it should be comprehensively local and avoid reporting that perpetuates existing inequalities. Historically marginalized communities are traditionally ignored or villainized in the local news because coverage focuses on wealthier areas and sensational or violent stories. Any AI based on journalistic output from the archives of legacy local news is likely to reproduce these information inequalities.

Preserving and improving local news requires adaptation and creative thinking, and new technological tools like AI should be a part of that. While there is much work already being done on AI and local news, Documenters adds a much-needed voice to the local news environment—encouraging popular participation by creating the raw materials to make local journalism and local governments work better for the people. Within this context, AI could help provide scale, accessibility, and accountability to rebuild trust in local media and improve representation for all members of the communities they serve. Researchers, technicians, and journalists should keep that context in mind when designing AI tools.

References

Abernathy, Penelope Muse. 2018. The expanding news desert. Chapel Hill, NC: Center for Innovation and Sustainability in Local Media, School of Media and Journalism, University of North Carolina at Chapel Hill.

American Journalism Project. 2025. “Impact report 2025.” https://www.pewresearch.org/journalism/2023/09/26/black-americans-experiences-with-news/. Accessed: 2025-12-18.

Anglin, Dale. 2025. “Two years in: $400 million mobilized for local news.” https://www.pressforward.news/two-years-in-400m-mobilized-for-local-news/. Accessed: 2025-12-18.

Associated Press. 2023a. “Artificial intelligence in local news case study: Brainerd Dispatch.” https://www.ap.org/wp-content/uploads/2024/02/ap-local-ai-brainerd-dispatch-oct-2023.pdf. Accessed: 2025-10-10.

Associated Press. 2023b. “Artificial intelligence in local news case study: Michigan Radio.” https://web.archive.org/web/20240321200722/https://www.ap.org/wp-content/uploads/2024/02/ap-local-ai-michigan-radio-oct-2023.pdf. Accessed: 2025-10-10.

Associated Press. 2025. “Artificial intelligence.” https://www.ap.org/solutions/artificial-intelligence/. Accessed: 2025-10-10.

Bedingfield, Sid, and Kathy Roberts Forde. 2021. “Journalism and the world it built.” In Journalism and Jim Crow: White supremacy and the Black struggle for a new America, eds. Kathy Roberts Forde and Sid Bedingfield. University of Illinois Press.

Bowman, Jennifer. 2024. “Measure E’s failure could lead to budget cuts at San Diego libraries.” San Diego Documenters. https://sandiegodocumenters.org/2024/11/25/san-diego-library-budget-cuts-measure-e/. Accessed: 2025-10-10. Based on notes by Vincent Outlaw, San Diego Documenter.

Cityscrapers.org. 2025. “City Scrapers: A City Bureau project.” Cityscrapers.org. Accessed: 2025-12-18.

Carroll, John M, Blaine Hoffman, Kyungsik Han and Mary Beth Rosson. 2015. “Reviving community networks: Hyperlocality and suprathresholding in Web 2.0 designs.” Personal and Ubiquitous Computing 19(2):477–491.

Cook, Timothy E. 2005. Governing with the news: The news media as a political institution. University of Chicago Press.

Culpepper, Sophie. 2025. “How Outlier is helping Detroiters get millions of dollars back from Wayne County.” NeimanLab. https://www.niemanlab.org/2025/04/how-outlier-is-helping-detroiters-get-millions-of-dollars-back-from-wayne-county/. Accessed: 2025-10-10.

Dale, Mariana. 2024. “LA strips telephone comment option from council committee meetings.” LAist. https://laist.com/news/politics/los-angeles-city-council-commitee-comments-virtual-in-person-email-phone. Accessed: 2025-10-10.

Darr, Joshua P, Matthew P Hitt and Johanna L Dunaway. 2018. “Newspaper closures polarize voting behavior.” Journal of Communication 68(6):1007–1028.

Darr, Joshua P, Matthew P Hitt and Johanna L Dunaway. 2021. Home style opinion: How local newspapers can slow polarization. Cambridge University Press.

Dikcis, Maria. 2025. “How documenters are transforming community reporting around the country.” City Bureau. https://www.citybureau.org/notebook/how-documenters-are-transforming-community-reporting-around-the-country. Accessed: 2025-10-10.

Documenters.org. 2025. “City Bureau’s Documenters network: People-powered news and information.” Documenters.org. Accessed 2025-12-18.

Djourelova, Milena, Ruben Durante and Gregory J Martin. 2024. “The impact of online competition on local newspapers: Evidence from the introduction of Craigslist.” Review of Economic Studies: 92(3): 1738–1772.

Einstein, Katherine Levine, David Glick, Luisa Godinez Puig and Maxwell Palmer. 2023. “Still muted: The limited participatory democracy of Zoom public meetings.” Urban Affairs Review 59(4):1279–1291.

Einstein, Katherine Levine, David M Glick and Maxwell Palmer. 2019. Neighborhood defenders: Participatory politics and America’s housing crisis. Cambridge University Press.

Ellger, Fabio, Hanno Hilbig, Sascha Riaz and Philipp Tillmann. 2024. “Local newspaper decline and political polarization – evidence from a multi-party setting.” British Journal of Political Science: 54(4):1256–1275 .

Fisher, Jillian, Shangbin Feng, Robert Aron, Thomas Richardson, Yejin Choi, Daniel W Fisher, Jennifer Pan, Yulia Tsvetkov and Katharina Reinecke. 2024. “Biased AI can influence political decision-making.” arXiv preprint arXiv:2410.06415.

Gao, Pengjie, Chang Lee and Dermot Murphy. 2020. “Financing dies in darkness? The impact of newspaper closures on public finance.” Journal of Financial Economics 135(2):445–467.

Garrick, David. 2024. “Furor over virtual public comment highlights a quandary for San Diego— participation or efficiency?” San Diego Tribune. https://www.sandiegouniontribune.com/2024/08/18/furor-over-virtual-public-comment-highlights-a-quandary-for-san-diego-participation Accessed: 2025-10-10.

Green, Elizabeth, Darryl Holliday and Mike Rispoli. 2023. “The roadmap for local news: An emergent approach to meeting civic information needs.”https://localnewsroadmap.org/.

Harrington, Christina, Sheena Erete and Anne Marie Piper. 2019. “Deconstructing community-based collaborative design: Towards more equitable participatory design engagements.” Proceedings of the ACM on Human-Computer Interaction 3(CSCW):1–25.

Heese, Jonas, Gerardo Pérez-Cavazos and Caspar David Peter. 2022. “When the local newspaper leaves town: The effects of local newspaper closures on corporate misconduct.” Journal of Financial Economics 145(2):445–463.

Heider, Don. 2014. White news: Why local news programs don’t cover people of color. Routledge.

Hersh, Eitan. 2020. Politics is for power: How to move beyond political hobbyism, take action, and make real change. Scribner.

Hopkins, Daniel J. 2018. The increasingly United States: How and why American political behavior nationalized. University of Chicago Press.

Hsu, Yen-Chia, Ting-Hao ‘Kenneth’ Huang, Himanshu Verma, Andrea Mauri, Illah Nourbakhsh, Alessandro Bozzon. 2022. “Empowering local communities using artificial intelligence.” Patterns 3(3).

Keatts, Andrew. 2024. “How Measure E’s failure affects San Diego’s budget.” Axios San Diego. https://inewsource.org/2024/07/23/san-diego-city-council-virtual-public-comment-meetings/. Accessed: 2025-12-18. Last updated November 22, 2024.

Kim, Eunji, Michael E Shepherd and Joshua D Clinton. 2020. “The effect of big-city news on rural America during the COVID-19 pandemic.” Proceedings of the National Academy of Sciences 117(36):22009–22014.

Lawrence, Regina G. 2000. “Game-framing the issues: Tracking the strategy frame in public policy news.” Political Communication 17(2):93–114.

Le Quéré, Marianne Aubin, Mor Naaman and Jenna Fields. 2024. “Not quite filling the void: comparing the perceptions of local online groups and local media pages on Facebook.” Proceedings of the ACM on Human-Computer Interaction 8:1–22.

Levin, Koby. 2025. “How to reclaim auction profits from Wayne County.” Outlier Media. https://outliermedia.org/txt-outlier/repay/. Accessed: 2025-10-10. Last updated March 7, 2025.

Lipka, Michael and Elisa Shearer. 2023. “Audiences are declining for traditional news media in the US–With some exceptions.” a.

Moskowitz, Daniel J. 2021. “Local news, information, and the nationalization of US Elections.” American Political Science Review 115(1):114–129.

Moy, Patricia, Michael R McCluskey, Kelley McCoy and Margaret A Spratt. 2004. “Political correlates of local news media use.” Journal of Communication 54(3):532–546.

Myers, Andrew CW. 2025. “Press coverage and accountability in state legislatures.” American Political Science Review: 1–19.

Niebla, Crystal, and San Diego Documenters. 2024. “City of San Diego may eliminate virtual public comment at meetings.” iNewsSource. https://inewsource.org/2024/07/23/san-diego-city-council-virtual-public-comment-meetings/. Accessed: 2025-12-18. Last updated July 23, 2024.

Napoli, Philip M, Ian Dunham and Jessica Mahone. 2017. “Assessing news media infrastructure: A state-level analysis.” Dewitt Wallace Center for Media & Democracy.

Noble, Safiya Umoja. 2018. Algorithms of oppression: How search engines reinforce racism. New York University Press.

Ovadya, Aviv. 2023. “Reimagining democracy for AI.” Journal of Democracy 34(4):162– 170.

Peterson, Erik, Joshua P Darr, Maxwell B Allamong and Michael Henderson. 2025. “Can Americans’ trust in local news be trusted? The emergence, sources and implications of the local news trust advantage.” American Journal of Political Science.

Pew Research Center. 2023a. “Newspapers Fact Sheet.” https://www.pewresearch.org/journalism/fact-sheet/newspapers/. Accessed 2025-12-18. Last updated November 10, 2023.

Pew Research Center. 2023b. “Black Americans’ Experiences With News.” https://www.pewresearch.org/journalism/2023/09/26/black-americans-experiences-with-news/. Accessed 2025-12-18. Last updated September 26, 2023.

Press Gazette. 2025. “Hot property: How the Real Estate Robot reporter drove millions of page views.” https://pressgazette.co.uk/news/real-estate-robot-journalism-automated-property/. Accessed 2025-12-18.

Robinson, Sue. 2023. How journalists engage: A theory of trust building, identities, and care. Oxford University Press.

Robinson, Sue, Margarita Orozco and Joshua P. Darr. 2025. “How the engagement journalism movement is changing political news content: an applied-research study.” Journalism & Mass Communication Quarterly 0(0):10776990251318649.

Rosen, Jay. 1999. What are journalists for? Yale University Press.

Rubado, Meghan E and Jay T Jennings. 2020. “Political consequences of the endangered local watchdog: Newspaper decline and mayoral elections in the United States.” Urban Affairs Review 56(5):1327–1356.

Shaker, Lee. 2014. “Dead newspapers and citizens’ civic engagement.” Political Communication 31(1):131–148.

Shoemaker, PJ and T Vos. 2009. Gatekeeping theory. New York: Routledge.

Singer, Jane B., Alfred Hermida, David Domingo, Ari Heinonen, Steve Paulussen, Thorsten Quandt, Zvi Reich and Marina Vujnovic. 2011. Participatory journalism: Guarding open gates at online newspapers. Wiley-Blackwell.

Snyder Jr., James M and David Strömberg. 2010. “Press coverage and political accountability.” Journal of Political Economy 118(2):355–408.

Thorson, Emily. 2024. The invented state: Policy misperceptions in the American public. Oxford University Press.

Usher, Nikki. 2021. News for the rich, white, and blue: How place and power distort American journalism. Columbia University Press.

Walker, Mason. 2021. “US newsroom employment has fallen 26% since 2008.” Pew Research Center.

Wenzel, Andrea. 2020. Community-centered journalism: Engaging people, exploring solutions, and building trust. University of Illinois Press.

Wenzel, Andrea, Sam Ford, and Efrat Nechushtai. 2020. “Report for America, report about communities: Local news capacity and community trust.” Journalism Studies 21 (3): 287–305.

Zaller, John and Dennis Chiu. 1996. “Government’s little helper: U.S. press coverage of foreign policy crises, 1945–1991.” Political Communication 13(4):385–405.

Appendix: Full ChatGPT Answers

Appendix A: DocGPT prompt and answers

DocGPT prompt: What advice would you have for how to obtain an extension request for a building code violation, according to the meetings notes in your database only? 

DocGPT answer:

Based on Cleveland's Board of Building Standards and Building Appeals meeting notes, here is specific advice on how to improve your chances of receiving an extension request for a building code violation:

1. Always Attend Your Hearing

• Key Meeting Insight: Many appeals were denied automatically if the appellant failed to appear before the board.
• Action: If you cannot attend, notify the board in advance to reschedule.

2. Provide Clear Documentation

• Key Meeting Insight: In cases where extensions were granted, applicants presented detailed repair plans, contractor agreements, or financial constraints.
• Action: Bring evidence such as:
  • Contractor bids or repair estimates.
  • Communication with the city about the violation.
  • Proof of progress on repairs (photos, receipts).

3. Communicate with City Officials Before the Hearing 

• Key Meeting Insight: Property owners who documented their communication with city officials had stronger cases.
• Action: Keep a record of emails, calls, and in-person meetings with the Department of Building and Housing.

4. Be Ready to Explain Any Delays 

• Key Meeting Insight: Some appellants received six-month extensions when they could demonstrate delays beyond their control, such as contractor shortages, supply chain issues, or financial hardship.
• Action: Clearly outline why more time is needed and what steps are in progress.

5. Stay Professional and Engage with the Board 

• Key Meeting Insight: In the DAKM Holdings LLC case, the property owner argued against an aggressive inspector and miscommunication but still received an extension by presenting a clear, structured request.
• Action: Remain calm, respectful, and solutions-focused when addressing the board.

6. Consider Alternative Compliance Options 

• Key Meeting Insight: Some residents proposed temporary compliance solutions (e.g., fencing a property while waiting for structural repairs), which showed good faith efforts.
• Action: If full compliance isn’t immediate, offer a reasonable interim solution to show commitment.

7. Inquire About Financial Assistance 

• Key Meeting Insight: Some property owners questioned whether financial aid or city grants existed for costly repairs.
• Action: Ask if the city offers programs to assist with code compliance or rehabilitation funding.

Final Takeaway 

To maximize your chances of an extension, attend your hearing, document everything, communicate with officials, explain your delays, and show progress toward compliance. By following these strategies, you can present a strong case for additional time.

Appendix B: ChatGPT prompt and answers

ChatGPT prompt: What advice would you have for how to obtain an extension request for a building code violation in Cleveland? 

ChatGPT answer:

If you’ve received a building code violation notice in Cleveland and require additional time to address the cited issues, it’s essential to follow the city’s established procedures to request an extension. Here’s a step-by-step guide to assist you:

1. Review the Violation Notice

• Upon receiving a violation notice, carefully examine it to understand the specific infractions and the timeframe allotted for corrections. Typically, the City of Cleveland provides a 30-day period to rectify violations. If you anticipate needing more time, it’s crucial to act within this initial period. clevelandohio.gov

2. Contact the Issuing Inspector

• The violation notice should include the name, phone number, and email address of the inspector who issued it. Reach out to this individual promptly to discuss the violations and inform them of your intention to request an extension. Maintaining open communication can be beneficial in demonstrating your commitment to compliance. clevelandohio.gov

3. File an Appeal with the Board of Building Standards To formally request an extension:

• Prepare a Written Appeal: Draft a letter addressed to the Board of Building Standards detailing your request for additional time. Clearly state the reasons for the extension, which may include factors such as adverse weather conditions, health issues, financial constraints, or difficulties in securing contractors. Specify the amount of additional time you're seeking and provide any supporting documentation that substantiates your request.
• Submit the Appeal: Within the initial 30-day correction period, submit your written appeal to the Board of Building Standards located at: 
  City Hall
  601 Lakeside Avenue Room 516
  Cleveland, OH 44114

For detailed instructions on filing appeals, refer to the city’s official guidelines. clevelandohio.gov

4. Await the Board’s Decision

• After submitting your appeal, the Board will review your request and notify you of their decision. It’s important to continue making progress on addressing the violations during this period to demonstrate your commitment to compliance.

Important Considerations

• Timeliness: Ensure that your appeal is filed within the initial 30-day correction period specified in your violation notice. Late submissions may not be considered, leading to potential legal actions or fines.
• Documentation: Maintain thorough records of all communications, receipts, and efforts made to address the violations. This documentation can support your appeal and demonstrate your proactive approach to resolving the issues.
• Communication: Stay in regular contact with the assigned inspector or relevant city officials to keep them informed of your progress and any challenges you encounter. Transparent communication can foster goodwill and may influence the outcome of your extension request.

By proactively engaging with the Board and clearly communicating your circumstances, you can increase the likelihood of obtaining an extension to address the building code violations adequately.