September 2025

A Year in Review

How National News Media Covers K-12 Education

By Nora Weber, Marisa Mission, Katrina Boone, Juliet Squire, and Andy Jacob

Executive Summary

Across the country, news media plays a critical role in shaping public discourse about K-12 education. Which K-12 issues media outlets cover, and how they cover those issues, influences how the public understands one of the most important functions in society: providing young people with a high-quality public education. Media coverage also influences decision-makers within the sector — from parents’ schooling choices to leaders’ policy decisions to teachers’ classroom practices.

Yet there is little information about how exactly the media uses its influence. This analysis begins to fill that gap by taking stock of trends in how news media outlets cover K-12 education. Guided by human design and oversight, and aided by artificial intelligence (AI) tools, more than 1,500 online articles published by 12 major print, broadcast, and radio outlets between June 1, 2024 and May 31, 2025 were categorized by primary theme (Appendix: Methodology, Limitations, and Sources).

The findings show that breaking political news drove a significant amount of education coverage over the last year — one that included a national election cycle and a new presidential administration with ambitions to remake the federal role in education. The most common themes were:

The U.S. Department of Education — the primary theme in 18% of articles
Gun violence in schools — the primary theme in 16% of articles
Sex, gender, LGBTQ+ issues, and Title IX — the primary theme in 10% of articles
Religion in schools — the primary theme in 9% of articles
Technology in schools (mostly related to cellphone bans) — the primary theme in 6% of articles

Political and cultural news appeared to crowd out coverage of academic outcomes and efforts to improve them. Some of the least common themes included:

Curriculum, instruction, and extracurriculars — the primary theme in 4% of articles
Teachers — the primary theme in 3% of articles
Academic outcomes — the primary theme in 2% of articles

Overall, fewer than one in 10 articles focused on teachers; curriculum, instruction, and extracurriculars; and academic outcomes. Given this analysis’ sample of national news media outlets, a limited focus on issues that are often debated and decided at the state and local level may be unsurprising. But in a media environment in which local outlets compete with national outlets — and national outlets compete with one another for clicks and page views — the trend is notable nonetheless.

Although the sample for this analysis includes a limited set of online articles over just a one-year time frame, patterns in coverage can provide policymakers, advocates, funders, and other K-12 sector leaders with an evidence-based assessment of how education topics are addressed in national news media. As the 2025-26 school year begins, this analysis can also serve as a useful tool for members of the media to spot trends, identify gaps, and shape coverage moving forward.

Across More Than 1,500 Online Articles About K-12 Education, 20 Themes Emerged

Theme	Coverage Includes
U.S. Department of Education	The U.S. Department of Education, including the debate over dismantling the agency
Gun Violence	School shootings and/or the topic of gun violence more broadly
Sex, Gender, LGBTQ+ Issues, and Title IX	The intersection of education and sex, gender, and LGBTQ+ issues, including transgender-related issues and Title IX changes or enforcement
Religion in Schools	Religious schools, religion in schools more broadly, and policies where religion and education intersect
Technology in Schools	Technology in schools, including cellphone bans, social media, AI, and other educational technology
Student Health and Disabilities	Student well-being, including both physical and mental health; students with disabilities; and disability accommodations related to both physical and learning disabilities.
Education Funding	Education funding at any level of K-12 education, including federal Title I funding, state finance reforms, and historical funding trends
Race, Racism, and Cultural Sensitivity	Race, racism, or race-related incidents, including coverage that disaggregates policy impacts by race
Immigration	Immigration policy and its effect on K-12 education and students
Diversity, Equity, and Inclusion (DEI) Executive Order	Any of President Trump’s 2025 executive orders related to DEI, equity in education, and civil rights
Curriculum, Instruction, and Extracurriculars	Teaching practices, coursework, and materials, including art and music education and extracurricular programs
School Choice	Parents’ rights discussions and school choice policies, including education savings accounts, private school vouchers, and charter schools
Teachers	Educators’ interests, employment, and instructional conditions, including teachers unions
Freedom of Speech and Protests	Student protests and political actions, as well as reactions to them; including students’ First Amendment rights and academic freedom of speech
Political Policy Platforms	Political parties and their education policy platforms, including Project 2025 and overall trends of education policies
Academic Outcomes	Student achievement scores or trends, whether at the national level (e.g., the National Assessment of Educational Progress [NAEP]) or state level
Postsecondary Pathways	Career pathways after high school, including work-based learning, workforce development, and career and technical education
Natural Disasters	The impact of natural disasters and emergencies on schooling
Socioeconomic Status	Household socioeconomic status and income in relation to students’ education
Enrollment	K-12 enrollment trends, including absenteeism, school closures, and/or consolidations

Political and Cultural Topics Were the Most Common Themes Covered in Online Articles

Nearly one in five online articles in this sample focused on the Trump administration’s efforts to dismantle the U.S. Department of Education and the layoffs of nearly half of the agency’s employees. Gun violence also topped the charts in coverage, including coverage of the legal proceedings following the 2022 school shooting in Uvalde, Texas and school shootings in Georgia, Wisconsin, and Tennessee. Given that both themes often followed breaking news, it is unsurprising that they top the list of most covered.

Coverage of education news related to sex, gender, LGBTQ+ issues, and Title IX was also prominent, largely driven by President Trump’s executive orders and subsequent Title IX guidance constraining how schools handle bathroom access, sports participation, and communication with parents about students’ gender identity.

Religion in schools was also a common topic, heightened by the U.S. Supreme Court case Oklahoma Statewide Charter School Board v. Drummond, which considered the constitutionality of religious charter schools.

School cellphone bans and, to a lesser extent, the use of AI-powered tools in education were also covered extensively across the country.

CNN, The Washington Post, NBC, Fox, USA Today, and The New York Times each published 10% or more of the online articles covering K-12 education for a total of 65% of all online articles covering K-12 education.
MSNBC and The Wall Street Journal each made up 5% or fewer of the online articles in this analysis’ sample.

The high overall coverage of the U.S. Department of Education holds true across most outlets, led by USA Today.
The high overall coverage of gun violence was less consistent and more likely to be covered by outlets that also have television news (e.g., network news channels ABC, NBC, CBS, as well as Fox and CNN).
Fox included the most online articles about sex, gender, LGBTQ+ issues, and Title IX.

Note: Omission of local news resulted in underreporting of K-12 education coverage for three metro areas: Los Angeles, New York, and the District of Columbia. As a result, this sample does not include articles in the Los Angeles Times about the impact of the January 2025 Los Angeles fires on local K-12 education (Appendix: Methodology, Limitations, and Sources).

Cadence of K-12 Education Coverage in the News Media Varied Widely by Topic

Spikes in online articles’ coverage of K-12 education align to major national events — after the September 2024 school shooting at Apalachee High School in Georgia; after the 2024 presidential election; and after President Trump’s 2025 inauguration and subsequent efforts to dismantle the U.S. Department of Education.

Figure 5: Timeline of Online Articles Covering K-12 Education
n = 1,543

Among the five most common themes covered in online articles — the U.S. Department of Education; gun violence; religion in schools; sex, gender, LGBTQ+ issues, and Title IX; technology in schools — the cadence of coverage varied significantly. Major breaking news events drove coverage of gun violence, the U.S. Department of Education, and (to a lesser degree) religion in schools. In contrast, coverage of sex, gender, LGBTQ+ issues, and Title IX as well as technology in schools was characterized more by a steady drumbeat of articles throughout the year.

The Percent of Articles That Focused on Competing Perspectives Varied Widely by Theme

While not “opinion” coverage, 71% of the online articles in this analysis covered education news while presenting competing viewpoints, particularly between political entities. For example, an article about gun violence would not be coded as a debate if the article was simply conveying information about the sequence of events in a school shooting but would be coded as a debate if the article included perspectives on gun control policies. Similarly, an article about political policy platforms would be considered a debate if the article acknowledged that there are multiple perspectives on what the policy platform should be but would not be considered a debate if the article was simply reporting on the content of the platform itself.

The prevalence of online articles framed around competing perspectives suggests that many issues in education are contested, especially when they intersect with political or cultural issues.

Among the themes most often covered with competing perspectives are the U.S. Department of Education; sex, gender, LGBTQ+ issues, and Title IX; religion in schools; the diversity, equity, and inclusion executive order; freedom of speech and protests; school choice; and political policy platforms.

On the other end of the spectrum, only six online articles focused primarily on the theme of socioeconomic status in relation to education and none of them was presented with competing perspectives. Online articles about enrollment, natural disasters, and gun violence were also less likely to be framed with competing perspectives.

Teaching- and Learning-Adjacent Themes Often Earned the Least Coverage in Online Articles

Across 1,543 online articles covering K-12 education in this analysis’ sample, just 132 (9%) addressed themes that many would associate closely with teaching and learning: academic outcomes; curriculum, instruction, and extracurriculars; and teachers. This may be due in part to this analysis’ focus on national news coverage.

Academic Outcomes: These articles largely focused on lasting impacts of the COVID-19 pandemic and related school closures on student academic achievement, NAEP results, and the results of the Trends in International Mathematics and Science Study.

Curriculum, Instruction, and Extracurriculars: These articles covered a wide range of topics, from the research and debates on reading instruction, to how educators were choosing to include or omit election news from classroom discussion. Some articles also discussed implications of specific policies, such as instructional content related to religion in schools or sex, gender, LGBTQ+ issues, and Title IX.

Teachers: These articles largely covered special interest stories, but also included several articles about how teacher unions have pushed back on Trump administration policies.

Appendix: Methodology, Limitations, and Sources

This news media analysis was designed by Bellwether and conducted using OpenAI’s API with significant human oversight and testing. The table below explains how Bellwether authors conducted the analysis.

Phase of Research	Phase 1: Extraction of Deductive Metadata	Phase 2: Thematic Coding and Extraction of Inductive Attributes	Phase 3: Trend Analysis
Human Design, Validation, and Analysis	Cleaned dataset of online articles. Identified key article themes by manually coding 50 articles and validating results with two separate analysts, then compiled those themes into a codebook for consistency. Created a prompt for OpenAI API to extract key article details (e.g., author, date, outlet). Iterated on AI prompt until OpenAI results aligned with human results at a minimum of 80% accuracy, based on a comparison of results in Microsoft Excel: Tested prompt on the manually-coded sample. Validated results against manual coding. Refined prompt, as needed.	Manually coded 100 additional articles for a total of 150 manually-coded articles. Created a thematic analysis prompt to use with the OpenAI API. Iterated on the prompt until results reached a minimum of 80% alignment with human results, based on a comparison of results in Microsoft Excel: Tested prompt on the 150 manually-coded articles. Validated results against manual coding. Refined prompt, as needed.	Spot-checked the final deductive and inductive results from the AI-powered coding. Removed articles that were duplicates or not related to education (e.g., special interest stories). Analyzed the final dataset according to project’s research questions.
AI Tool	Pulled key details (e.g., author, date, outlet) from every article in the sample to create a dataset of article attributes.	Converted the articles and theme codebook into numeric representations and stored them in a specialized search index to be used for thematic analysis. Conducted the thematic analysis by identifying each article’s main topic and coding it according to the theme codebook.	N/A

Sample Notes

The online articles in the sample were collected by Mercury LLC according to the following criteria:

Time Frame: June 1, 2024 to May 31, 2025
Outlets: Online articles from 12 major U.S. print, broadcast, and radio outlets, including: ABC, CBS, CNN, Fox, Los Angeles Times, MSNBC, NBC, NPR, The New York Times, The Wall Street Journal, The Washington Post, and USA Today
Search Logic: Headlines/subheads containing “education,” “school(s),” “teachers,” or “students”; K-12 only (excludes early ed and higher ed). Three outlets (New York Times, Washington Post, and Los Angeles Times) included considerable coverage from their regional or metro area. This coverage was omitted except when it directly addressed national news. This yielded 1,579 articles initially. After de-duplicating and excluding non-education-related (e.g., special interest) articles, the final sample was 1,543 articles.

Ensuring Rigor When Using AI

The use of machine learning and natural language processing is neither new nor uncommon in content analyses.¹ The use of newer generative AI and large language models (LLMs), on the other hand, has undergone less testing — but research suggests promise in its application.² Using an LLM for content analyses can greatly improve speed and efficiency, allowing for deeper analysis. At the same time, it also introduces new risks, such as the potential for fabricated data, inaccurate coding, and a lack of transparency due to the “black box” of how these models function.³ To mitigate these risks, the Bellwether team grounded this analysis in validated design and analytical practices, as well as consulted resources and used the following strategies to test the reliability and validity of results.

Strategy 1: Guardrails for Validity

These validity criteria were determined prior to beginning the analysis and set safeguards to ensure that analysis did not move forward on faulty assumptions:

A minimum of 10% of the sample, chosen at random, must be first coded by humans independent of one another and independent of seeing the AI results.
The thematic analysis codebook must be developed by humans independent of AI.
Prompts must be developed through iterative and progressive testing, where AI outputs are checked against the human coding. The analysis will not move forward unless the results reach a minimum threshold of 80% accuracy.
After the analyses are run on the full sample of articles, humans will conduct spot checks of the AI results.
All prompts, model versions, validation sets, thresholds, and codebook changes must be documented.

Strategy 2: Using Retrieval for Thematic Analysis Accuracy

Because the full sample of articles covered a wide swath of topics, the codebook developed became robust (Across More Than 1,500 Online Articles About K-12 Education, 20 Themes Emerged). Had the entire codebook been sent through the API to ChatGPT alongside the article text, it would have introduced a large amount of irrelevant text that would have led to a) hitting the model’s context limits faster, and/or b) context dilution — irrelevant themes competing with relevant ones — increasing the risk of off-target or inconsistent theme assignments.⁴

Instead, the thematic analysis implemented a retrieval-augmented generation (RAG) pipeline. RAG is a strategy often used for enterprise search, customer-support assistants, and document question answering to ground model outputs in the most relevant context.⁵ In machine learning research, RAG has consistently improved accuracy and factuality on knowledge-intensive tasks and is now a common pattern applied across developers and use cases.

This analysis used a vector search for the retrieval method, which involved embedding both the articles and the codebook of themes into two vector databases and calculating the eight themes closest to each article. Those eight themes, along with the article text, were sent to the larger LLM for final thematic coding. Adding this RAG layer limited the choice of themes fed to ChatGPT, minimizing “noise” (i.e., irrelevant themes or text) in the prompt to improve accuracy of the thematic analysis.

Strategy 3: Model Choice and Technical Specifications

As LLMs grow and are updated and modified, selecting the appropriate model to handle a prompt is critical to ensuring the accuracy of results.

Both the deductive and inductive coding were aided by ChatGPT 4.1, which in July 2025 (when the analysis was conducted) was OpenAI’s largest and most advanced nonreasoning model. Reasoning models were determined to be unnecessary given the general straightforwardness of the deductive and inductive coding, but future versions of this analysis will likely use reasoning models as nonreasoning ones are phased out.

The vector embedding was handled by OpenAI’s text-embedding-3-large, which was OpenAI’s most capable embedding model as of August 2025.⁶

Limitations

Like all content analyses, this analysis relies on a sample of news media coverage. Bellwether authors selected 12 large, major print, broadcast, and radio outlets based on circulation or broadcast audience.⁷ This analysis represents coverage by those outlets but is not necessarily generalizable to all national coverage or coverage at other levels. Patterns of education coverage at the regional or local level, or in industry-specific publications, may differ significantly from this analysis’ findings.
Due to time and platform constraints, all coverage in this sample was analyzed as online articles from each source’s website. By platform, digital news accounts for the largest share of news consumed in the U.S., which included audiences consuming news on legacy media outlets’ websites.⁸ Data suggests that there are some differences between online, print, and broadcast coverage overall, but nothing indicates major differences in the coverage and tone of a given outlet, comparing their digital and nondigital assets.⁹ Therefore, this analysis is an accurate representation of the outlets sampled.
To capture education coverage as thoroughly as possible, Bellwether authors worked with Mercury LLC to search coverage for headlines with the following education-related keywords (and close variants): “education,” “school,” “teacher,” and “student.” This search consistently returned highly relevant articles and authors did not identify any systematic omissions; however, there were likely articles related to education that were omitted because their headlines did not include any of this analysis’ search terms.
As noted in “Political and Cultural Topics Were the Most Common Themes Covered in Online Articles,” three media outlets include “local” sections: the Los Angeles Times, New York Times, and Washington Post. Coverage from each of these sources’ local or regional sections was omitted, except when it specifically mentioned education broadly or at the national level. Due to this, the sample may slightly under-sample regional coverage for the following three cities: Los Angeles, New York, and the District of Columbia. The authors have no reason to believe this skews the analysis more than including this coverage would and did not include any regional comparison in this report.
During thematic coding, the authors strove to identify the “primary” theme of an article. The coverage of most articles had a single distinct focus; however, some articles had joint primary themes. In these instances, the authors selected the theme around which the article was most closely framed, based on what appeared most clearly in the headline or body text. The authors worked carefully in developing a thematic codebook to avoid overlap and validated the coding manually between reviewers. However, differences in selection between primary and secondary themes could have resulted in small changes to the results.
The use of AI introduced potential errors that may have yielded different results than if all articles had been manually coded. The authors carefully tested and worked to mitigate these risks, as detailed above.

Sources

Bart Bonikowski and Laura K. Nelson, “From Ends to Means: The Promise of Computational Text Analysis for Theoretically Driven Sociological Research,” Sociological Methods & Research 51, no. 4 (2022): 1469–1483, https://doi.org/10.1177/00491241221123088; Jelle W. Boumans and Damian Trilling, “Taking Stock of the Toolkit: An Overview of Relevant Automated Content Analysis Approaches and Techniques for Digital Journalism Scholars,” Digital Journalism 4, no. 1 (2016): 8–23, https://doi.org/10.1080/21670811.2015.1096598; Justin Grimmer and Brandon M. Stewart, “Text as Data: The Promise and Pitfalls of Automatic Content Analysis Methods for Political Texts,” Political Analysis 21, no. 3 (2013): 267–297, https://doi.org/10.1093/pan/mps028.
Suhaib Abdurahman, Alireza Salkhordeh Ziabari, Alexander K. Moore, Daniel M. Bartels, and Morteza Dehghani, “A Primer for Evaluating Large Language Models in Social-Science Research,” Advances in Methods and Practices in Psychological Science 8, no. 2 (2025):1–25, https://doi.org/10.1177/25152459251325174.
Davide Castelvecchi, “Can We Open the Black Box of AI?,” Nature 538, no. 7623 (October 2016): 20–23, https://www.nature.com/news/can-weopen-the-black-box-of-ai-1.20731.
Minglai Yang et al., “How Is LLM Reasoning Distracted by Irrelevant Context? An Analysis Using a Controlled Benchmark,” arXiv at Cornell University, May 24, 2025, https://arxiv.org/abs/2505.18761.
“What Is Retrieval Augmented Generation (RAG) [Examples Included],” SuperAnnotate blog, February 3, 2025, https://www.superannotate.com/blog/rag-explained.
“Azure OpenAI in Azure AI Foundry Models,” Microsoft, August 12, 2025, https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models?tabs=global-standard%2Cstandard-chat-completions.
Christopher St. Aubin and Jacob Liedke, “News Platform Fact Sheet,” Pew Research Center, September 17, 2024, https://www.pewresearch.org/journalism/fact-sheet/news-platform-fact-sheet/.
“News Media Tracker Frequently Asked Questions,” Pew Research Center, June 10, 2025, https://www.pewresearch.org/journalism/2025/06/10/news-media-tracker-frequently-asked-questions/.
Laura K. Nelson, Derek Burk, Marcel Knudsen, and Leslie McCall, “The Future of Coding: A Comparison of Hand-Coding and Three Types of Computer-Assisted Text Analysis Methods,” Sociological Methods & Research 50, no. 1 (2018): 202–237, https://doi.org/10.1177/0049124118769114.

Additional References Consulted:

Jyoti Belur, Lisa Tompson, Amy Thornton, and Miranda Simon, “Interrater Reliability in Systematic Review Methodology: Exploring Variation in Coder Decision-Making,” Sociological Methods & Research 50, no. 2 (2018): 837–865, https://doi.org/10.1177/0049124118799372.
Andrea J. Bingham, “From Data Management to Actionable Findings: A Five-Phase Process of Qualitative Data Analysis,” International Journal of Qualitative Methods 22 (2023), https://doi.org/10.1177/16094069231183620.
Jennifer Kavanagh, William Marcellino, Jonathan S. Blake, Shawn Smith, Steven Davenport, and Mahlet Gizaw, News in a Digital Age: Comparing the Presentation of News Information over Time and across Media Platforms (RAND, 2019), https://www.rand.org/pubs/research_reports/RR2960.html.
Aaron Smith, “Q&A: How We Used Large Language Models to Identify Guests on Popular Podcasts,” Pew Research Center, February 6, 2024, https://www.pewresearch.org/short-reads/2024/02/06/how-we-used-large-language-models-to-identify-guests-on-popular-podcasts/.
Claudia Deane, “How Pew Research Center Is—and Is Not—Using Large Language Models in Our Work,” Pew Research Center, August 27, 2024, https://www.pewresearch.org/decoded/2024/08/27/how-pew-research-center-is-and-is-not-using-large-language-models-in-our-work/.
Ann Reisner, “Chapter 12: Judgment Rule 4 for Content Analysis,” in Reading Social Science Methods (Urbana, IL: Windsor & Downs Press, 2023), https://iopn.library.illinois.edu/pressbooks/socialsciencemethods/chapter/judgment-rule-4-for-content-analysis/.

Acknowledgments, About the Authors, About Bellwether

Acknowledgments

Thank you to the Walton Family Foundation for its financial support of this project. We would also like to thank Mercury LLC for its help in compiling the sample for this analysis. Thank you also to our Bellwether colleague Alexis Richardson for her support. Thank you to Amy Ribock, Kate Stein, McKenzie Maxson, Zoe Cuddy, Julie Nguyen, Mandy Berman, and Amber Walker for shepherding and disseminating this work, and to Super Copy Editors. The contributions of these individuals and entities significantly enhanced our work; however, any errors in fact or analysis remain the responsibility of the authors.

About the Authors

Nora Weber

Nora Weber is a senior analyst at Bellwether in the Policy and Evaluation practice area. She can be reached at nora.weber@bellwether.org.

Marisa Mission

Marisa Mission is a senior analyst at Bellwether in the Policy and Evaluation practice area. She can be reached at marisa.mission@bellwether.org.

Katrina Boone

Katrina Boone is a senior associate partner at Bellwether in the Policy and Evaluation practice area. She can be reached at katrina.boone@bellwether.org.

Juliet Squire

Juliet Squire is a senior partner at Bellwether in the Policy and Evaluation practice area. She can be reached at juliet.squire@bellwether.org.

Andy Jacob

Andy Jacob is a partner at Bellwether and leads the External Relations team. He can be reached at andy.jacob@bellwether.org.

Bellwether is a national nonprofit that exists to transform education to ensure systemically marginalized young people achieve outcomes that lead to fulfilling lives and flourishing communities. Founded in 2010, we work hand in hand with education leaders and organizations to accelerate their impact, inform and influence policy and program design, and share what we learn along the way. For more, visit bellwether.org.