Measuring Generative AI Content in Political Campaigns: Evidence of Rising Adoption and Non-Disclosure

Anna-Maria Gueorguieva, Nic WeberarXiv 2026 · arXiv

This study by Anna-Maria Gueorguieva and Nic Weber from the University of Washington provides one of the first large-scale empirical looks at how generative AI is actually being used "in the wild" by U.S. political campaigns. By analyzing over 12,000 candidates, the researchers move beyond the hype of deepfakes to measure the routine adoption of AI-generated content (AIGC) and the (lack of) effectiveness of emerging state regulations.

TL;DR

The researchers analyzed a novel dataset of nearly 13,000 U.S. political candidates from 2023 to 2025, finding that detectable AI-generated content is rising significantly across websites, social media, and emails. Despite a wave of new state laws, direct disclosure of AI use is virtually non-existent, and challengers are adopting the technology at significantly higher rates than incumbents.

What Problem Does This Paper Address?

While public and policy attention has focused on "materially deceptive" deepfakes, there has been little data on the broader, more routine use of generative AI in political messaging. Nearly half of U.S. states have passed laws requiring disclosure or prohibiting deceptive AIGC, but it remains unclear if these laws actually change candidate behavior or if candidates are disclosing their AI use at all.

The authors tackle three main questions:

  1. How prevalent is AIGC in campaigns, and how has it changed over time?
  2. Which campaign features (e.g., incumbency, funding, party) are associated with higher AI use?
  3. Do state-level regulations actually reduce the amount of AI content produced?

Methods

The authors developed two primary metrics: Magnitude (the estimated total volume of AIGC produced) and Modality (how that content is distributed across different communication channels).

To detect AIGC at scale, they employed a multi-modal detection pipeline:

  • Text Detection: They used Pangram (Version 3), a commercial detector chosen for its low false-positive rates on long-form text.
  • Image Detection: They utilized GPT-4o-mini (referred to in the paper as GPT-5.4-mini) to classify images, a method recently validated in similar political contexts.
  • Statistical Modeling: To understand the drivers of AI use, they used zero-inflated beta regressions (to handle the many "zero-use" cases) and Two-Way Fixed Effects (TWFE) models to isolate the impact of state regulations by comparing states before and after laws were enacted.
Fig. 1. We collect data on political candidates across political races from 2023 to 2025 and their campaign data from three modalities (campaign websites, X posts, and emails). We apply AI detection methods to text and image content and introduce two metrics to estimate prevalence of AIGC across races and modalities. We use these estimates to characterize detectable AIGC use over time, which types of candidates use it, and potential effects of state regulations on its use.

Datasets

The study introduces a massive, novel dataset covering 12,977 candidates for U.S. office (federal, state, and local) between 2023 and 2025. The content analyzed includes:

  • 12,206 text samples from campaign websites.
  • 15,714 text samples and 104,098 images from X (formerly Twitter).
  • 2,076 emails from the Archive of Political Emails.

This was supplemented with campaign finance data from OpenSecrets and legislative data from the National Conference of State Legislatures (NCSL).

Key Insights and Ideas

  • The "Challenger's Tool": One of the most striking findings is that challengers use significantly more AIGC than incumbents (incumbents produced about 20% less). This suggests that generative AI is acting as a "force multiplier" for under-resourced campaigns, allowing them to generate content that would otherwise require a larger staff.
  • The Disclosure Ghost Town: Despite the legal push for transparency, the researchers found only one campaign website in their entire sample that explicitly disclosed the use of AI. Manual reviews of detected AI images on social media also found zero disclosures.
  • Regulatory Mismatch: State laws currently focus on "deceptive" content. However, the study found that these laws have no measurable effect on the overall volume of AIGC in state and municipal races. There was a modest reduction in federal races, but the broad adoption of AI for routine tasks seems to fall outside the shadow of current narrow regulations.
  • Platform Variation: AI use isn't uniform. Candidates are more likely to use AI-generated or assisted content on X (Twitter) than in their formal emails or on their primary campaign websites.

Key Results

  • Rising Adoption: Detectable AIGC increased significantly across all modalities from 2023 to 2025.
  • Current Scale: While rising, the overall percentage of content identified as AI-generated remains relatively low: 7.6% of text samples and 6.1% of image samples.
  • Funding Correlation: There is a slight but statistically significant positive relationship between total campaign funding and AI magnitude, though the incumbency effect remains much stronger.
  • Indirect Disclosure Gaps: When cross-referencing with campaign finance reports, the researchers found that 48% of candidates who reported spending on AI tools were not identified as using AI by the detection tools, suggesting AI is often used for "back-office" tasks (like logistics or outreach) rather than just content generation.

Limitations and Future Work

The authors acknowledge several key limitations:

  1. Detection Accuracy: AI detectors are an "arms race" and are not 100% accurate; these results should be viewed as a lower bound of actual use.
  2. Missing Modalities: The study did not analyze video or audio, which are the primary targets of "materially deceptive" legislation.
  3. Substance vs. Scale: The metrics measure how much AI is used, not how it is used (e.g., for a routine policy summary vs. a deceptive attack).

Future work is encouraged to apply these metrics to video content and to investigate the "downstream" chain of AI use, such as when third-party consultants use AI without the candidate's direct knowledge.

Summary updated

Abstract

The use of generative AI in political campaigns has drawn public concern and policy attention. Public concerns range from distrust of candidates to skepticism about the authenticity and accuracy of political advertising, including AI-generated images and video. Nearly half of U.S. states have enacted laws that prohibit, or require disclosure of, "materially deceptive" AI-generated content (AIGC) in elections to guard against these concerns. This study measures how candidates and their campaigns produce and circulate AIGC during an election cycle. We develop two measures that estimate not only wether campaigns use AIGC, but how much they produce (magnitude) and in what forms (modality). We apply these measures to a novel dataset of text and images from the campaign websites, X accounts, and emails of 12,977 candidates for U.S. political office from 2023 to 2025. We report three findings. First, detectable AIGC from campaigns increased significantly across modalities over the observation window, and campaigns that use AIGC frequently do not disclose it. Second, incumbents' campaigns produced significantly less AIGC than challengers' campaigns. Third, candidates in states with AIGC disclosure or prohibition laws do not produce measurably less AIGC than candidates in other states. Our method and findings have implications for transparency and trust in elections, and on state laws that regulate AI use in campaigns. Because current state laws largely target mainly "materially deceptive" content, we identify additional policy mechanisms that could increase electoral transparency and improve trust as AIGC becomes more prevalent.
BibTeX
@misc{gueorguieva2026measuring,
  author = {Anna-Maria Gueorguieva and Nic Weber},
  title = {{Measuring Generative AI Content in Political Campaigns: Evidence of Rising Adoption and Non-Disclosure}},
  year = {2026},
  eprint = {2611.11658},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.11658},
}
Figure