FEC disclosure regime shapes observed partisan patterns in campaign AI spending

Summary

• An analysis of FEC itemized disclosures by Nathan Sanders and Bruce Schneier documents at least $17 million in reported AI vendor payments across 523 federal candidates and committees since 2020, and the order-of-magnitude gap between those direct payments and Wesleyan Media Project’s $80 million in AI-supported ad spending places the disclosure regime — not partisan preference — at the top of the explanatory ranking for the observed vendor clustering.
• Anthropic’s 2026 jump from essentially zero reporting campaigns to at least 65, with a nearly two-to-one Democratic-to-Republican split, partially fits a perceived-ideological-alignment reading, but the bipartisan composition of major vendor customer lists and Republican Sen. Tom Cotton’s status as the largest individual Anthropic user substantially undercut that explanation.
• Campaign Nucleus’s six-figure Republican disclosed spend and Daisychain’s growth from roughly $50,000 in 2024 to about $300,000 in 2026 — driven by Michigan’s Abdul El-Sayed Senate campaign — fit a reading in which vendor choice inherits from consultant and party-ecosystem relationships, with the analysis’s reported 12 percent dual-vendor rate consistent with that pattern.
• Disclosure-categorization conventions, in which most OpenAI expenses are filed as “office expenses” or staff subscriptions rather than as political services, and the substrate’s silence on the channels — LLC pass-throughs, joint fundraising committees, consultant arrangements — by which AI spending might bypass FEC itemization together keep the verdict provisional and subject to revision if the larger undisclosed universe is materially different in its partisan composition.

A September 29 Guardian report on the Sanders-Schneier analysis of FEC itemized disclosures documents at least $17 million in reported spending on AI technology vendors across 523 federal candidates and committees since 2020 — a figure that, on its face, suggests a small but real AI footprint in American campaigns. The disclosed total captures only direct payments by candidates and committees to AI vendors, and the authors emphasize that spending by political consultants, media firms, outside groups and PACs “are very likely where the bulk of campaign AI activity occurs.” Wesleyan Media Project data referenced by Sanders and Schneier puts AI-supported political ad spending in the current cycle at at least $80 million, an order of magnitude above the direct vendor payments the FEC captures. The disclosed partisan clustering therefore requires a layered explanation: vendor choice reflects what the disclosure regime captures — and how — and only secondarily the product fit, inherited consulting relationships and partial partisan preference that the substrate also documents.

What the FEC filings document

The Sanders-Schneier analysis covers FEC itemized disclosures from 2020 through the current 2026 cycle and reports at least $17 million in spending on AI technology vendors across 523 federal candidates and committees. At least 80 federal campaigns and committees have reported spending with OpenAI since 2024, totaling approximately $50,000 and skewing slightly more Republican than Democratic; the Republican National Committee is the largest single OpenAI buyer, with nearly $10,000 in reported expenses. Top individual OpenAI users include Republican campaigns for Mike Lawler, John Kennedy and Bill Cassidy, alongside California Democrats Ro Khanna and Ted Lieu. Most OpenAI expenses are categorized in FEC filings as office expenses or staff subscriptions to ChatGPT rather than as political services, and OpenAI’s policies prohibit some political uses of the ChatGPT tool.

Anthropic has reached a similar scale of adoption but with a sharply different partisan split: at least 65 candidates or committees report paying Anthropic in 2026, up from essentially zero in previous years, with a nearly two-to-one Democrat-to-Republican ratio. The largest individual Anthropic user is the campaign of Republican Sen. Tom Cotton of Arkansas, who reported more than $4,000 in Anthropic software spending in his June filing. Other major Anthropic users include Montana Independent Senate candidate Seth Bodnar, Jason Knapp — who lost a Democratic House primary in Virginia — and Democratic Alaska Senate candidate Mary Peltola. The analysis reports that approximately 12 percent of campaigns and committees using either ChatGPT or Claude report expenditures to both vendors; campaigns tend to use one or the other.

xAI has been reported as a vendor by seven federal and two state-level candidates or committees, totaling about $5,000, with the majority spent by Robert F. Kennedy Jr.’s 2024 presidential campaign. Other reported xAI users include Republicans Dave McCormick and Thomas Massie.

Specialized political-AI vendors dominate total dollars. The Republican-linked Campaign Nucleus — an AI-powered voter engagement tool associated with former Trump campaign manager Brad Parscale — has attracted six-figure spending from the RNC, multiple Trump-aligned PACs, and five-figure investments from Speaker Mike Johnson and Arizona Senate candidate Kari Lake. The Democratic-leaning Daisychain has garnered about $300,000 in reported candidate spending in the 2026 cycle, up from roughly $50,000 in 2024, with more than half of this year’s spending from the Senate campaign of Democrat Abdul El-Sayed in Michigan.

The legacy AmplifAI tool — automated text messaging characterized in the substrate as “an evolution of robocalling technology” — absorbed $4.7 million in campaign spending in the 2022 cycle alone, used heavily by Democratic candidates including Mark Kelly, Joe Biden, Bernie Sanders and Adam Schiff; AmplifAI is now owned by media conglomerate Triller, and its spending appears to have tapered in subsequent cycles. Prompt.io, described as the legacy Republican-serving texting vendor, retains about $375,000 in 2026 spending to date, down from more than $500,000 in the 2022 cycle, and continues to draw cross-coalition users including the A More Affordable California PAC sponsored by Uber (more than $1 million in 2026), Republican Massachusetts gubernatorial nominee Michael Minogue, failed Republican California gubernatorial candidate Ché Ahn, and the Republican-aligned Super PAC Neighbors for a Better Colorado.

At the synthetic-media end, just six federal candidates or committees reported spending with the audio generator ElevenLabs (totaling about $1,400, led by independent candidate Samir Witta in Colorado’s sixth congressional district), and five reported about $1,600 with the image generator Midjourney (led by Sholdon Daniels, the Republican primary runner-up in Texas’s 30th district). At the state level, Sanders and Schneier found at least $92,000 in spending attributable to modern generative AI vendors since 2022 across California, Colorado, Massachusetts and Washington, spread across at least 108 candidates and committees, with state-level AI spending running about ten times as high in 2026 as in all of 2024. Daisychain again leads state-level spend, with OpenAI and Claude dominant among general-purpose vendors and DonorAtlas — an AI-powered prospect research tool — sitting behind only Daisychain and OpenAI in state-level use, buoyed by nearly $4,000 in spending from the California Democratic Party.

Who benefits from the disclosed patterns

The Republican National Committee, Trump-aligned PACs, Speaker Mike Johnson and Kari Lake have made six-figure or five-figure disclosed expenditures to Campaign Nucleus, concentrating disclosed Republican AI-vendor spend on a single firm associated with former Trump campaign manager Brad Parscale. On the Democratic side, the Senate campaign of Abdul El-Sayed accounts for more than half of Daisychain’s 2026 spending, making that single campaign the largest disclosed beneficiary of the vendor’s roughly six-fold growth from 2024 to 2026.

The Sanders-Schneier analysis attributes Anthropic’s Democratic adoption tilt to “the company’s perceived political orientation and clashes with the Trump administration” — language that, taken at face value, identifies Anthropic itself (rather than its Democratic customers) as the primary beneficiary of the disclosed partisan split, via brand positioning aligned with the Democratic coalition.

Cross-coalition spending complicates any clean partisan-beneficiary framing: the A More Affordable California PAC, sponsored by Uber, is the single largest disclosed Prompt.io customer at more than $1 million in 2026, and other recurring Prompt.io customers span both parties. The specialized political-AI vendor category as a whole captures the bulk of disclosed spending — AmplifAI’s $4.7 million in the 2022 cycle and the continued six-figure Campaign Nucleus spend in 2026 represent orders-of-magnitude larger disclosed revenue flows than general-purpose vendors like OpenAI ($50,000 total since 2024) or ElevenLabs ($1,400 total).

Competing explanations for the partisan clustering

The substrate supports several overlapping accounts of the disclosed patterns, each with its own supporting and counter-evidence. Perceived ideological alignment fits the Anthropic Democratic tilt, the Republican concentration in Campaign Nucleus and the Democratic concentration in Daisychain. It is undercut by the bipartisan composition of OpenAI’s customer list (Lawler, Kennedy and Cassidy alongside Khanna and Lieu), by Republican Sen. Tom Cotton’s status as the largest individual Anthropic user, and by xAI’s meager Republican uptake despite Musk’s Trump-aligned positioning.

Vendor-side targeting fits the product orientations of Campaign Nucleus and Daisychain, each built for its respective clientele. It is weakened by xAI’s meager Republican uptake despite an explicit Republican positioning.

Vendor-marketing reach is supported by the Wesleyan order-of-magnitude gap between FEC-disclosed vendor payments and AI-supported ad spending; the substrate does not document which vendors market more aggressively to which party, so the channel is inferred rather than measured.

Disclosure-regime artifact is supported by Sanders and Schneier’s note that FEC filings do not capture spending by political consultants, media firms, outside groups or PACs; the substrate also does not document the specific channels (LLC pass-throughs, joint fundraising committees, vendor bundling) by which AI spending might bypass itemization.

Adoption timing and market-entry effects fit Anthropic’s 2026 jump from essentially zero to at least 65 reporting users; the account is weakened by the two-to-one Democratic tilt of those 2026 adopters.

Product fit fits the AmplifAI-to-Daisychain displacement on the Democratic side. On the Republican side the displacement is less clean: Campaign Nucleus is described as an AI-powered voter engagement tool rather than a pure texting application, and Prompt.io — not Campaign Nucleus — retains the bulk of Republican texting spend.

Consulting-inheritance and network effects fit the 12 percent dual-vendor rate and the cross-cycle persistence of vendors like AmplifAI within the Democratic coalition; Prompt.io’s cross-coalition adoption (the Uber-sponsored PAC at $1 million, Minogue, Neighbors for a Better Colorado) indicates the pattern is a tendency with notable exceptions, not a perfect partisan sorting.

Disclosure categorization fits the convention under which most OpenAI expenses are filed as office expenses or staff subscriptions; the substrate does not establish whether the convention is systematic across campaigns.

The null hypothesis — that the observed patterns reflect filing noise rather than a real signal — is weakened by the state-level growth figure, which spans at least 108 candidates and committees across four states and shows state-level AI spending running roughly ten times as high in 2026 as in all of 2024.

Why the disclosure regime ranks first

The Wesleyan order-of-magnitude gap — at least $80 million in AI-supported ad spending versus the disclosed $17 million in direct vendor payments since 2020 — sits at the center of the weighting. That gap, combined with the authors’ explicit observation that FEC filings miss consultant, media-firm, outside-group and PAC spending, makes any vendor-choice reading drawn from the disclosure data inherently provisional. The two disclosure-regime accounts (vendor-marketing reach and disclosure-regime artifact) are the only explanations in the substrate that directly address why the captured sample is non-representative of the larger universe; every other account can be true in some measure and still leave the bulk of campaign AI expenditure unmeasured.

Network effects and adoption timing sit at intermediate weight. The 12 percent dual-vendor rate is the strongest evidence for the network reading; Anthropic’s 2026 jump from essentially zero is the strongest evidence for the timing reading. Vendor-side targeting and product fit fit partially — product fit fits the AmplifAI-to-Daisychain displacement on the Democratic side but not on the Republican side, where Prompt.io retains most spend; vendor-side targeting is weakened by xAI’s meager Republican uptake despite its explicit Republican positioning.

Perceived ideological alignment is the weakest of the surviving hypotheses. The bipartisan composition of major vendor customer lists and the largest-Anthropic-user-is-a-Republican-senator counter-example substantially weaken it, and the dollar figures are too modest to support sustained ideological vendor loyalty at the federal level. The xAI-meager-Republican-uptake observation is the one piece of evidence consistent with the alignment account’s strongest form and is preserved as such.

The synthesis is provisional because FEC filings capture only direct vendor payments and not the larger universe of consultant and PAC spending, leaving the documented patterns subject to revision if the larger undisclosed universe is materially different in its partisan composition. Disclosure-categorization effects provide an additional reason for caution: systematic filing of AI subscriptions as “office expenses” rather than as political services may distort vendor-category comparisons.

What remains unseen and what comes next

The substrate does not document the specific channels by which AI spending might bypass FEC itemization — LLC pass-throughs, joint fundraising committees, vendor bundling, consultant arrangements — so the strength of the disclosure-regime reading rests on the Wesleyan order-of-magnitude gap rather than on channel-level evidence. Wesleyan measures ad spending on AI-generated media rather than vendor choice, so it cannot adjudicate partisan skews in vendor choice.

Sanders and Schneier note that a large share of campaign spending occurs in the final weeks before November, leaving 2026 totals incomplete at the time of analysis. The Anthropic jump, the ten-times state-level growth, and the 2026-cycle Daisychain, Campaign Nucleus and Prompt.io totals are all mid-cycle snapshots. Wesleyan Media Project reports at least 164 political ads in this cycle have included AI-generated media, supported by at least $80 million in ad spending — a figure that, if extrapolated across the consultant and PAC channels the FEC does not capture, would imply that the disclosed $17 million since 2020 may represent a fraction of total campaign AI expenditure for the 2026 cycle alone.

A September 2025 Pew survey cited in the substrate found that more than 70 percent of Americans would think less of a candidate who used AI to help write a speech — a backdrop against which campaigns may have reasons to file AI subscriptions as office expenses rather than as political services, regardless of how the tools are actually used. The disclosed pattern converges on three overlapping causal accounts — network effects, vendor political visibility and the disclosure regime — such that any single-mechanism account is likely to be incomplete. The patterns are real and partially correlate with partisan affiliation, but they reflect a market in which vendor choice is shaped primarily by what the FEC disclosure regime captures — and how — and only secondarily by product fit, inherited consulting relationships and partial partisan preference.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Differential Diagnosis

Lists the candidate explanations for a symptom and rules them out one by one.