Californians heading to the polls this fall are being asked to weigh in on a slate of ballot measures backed by hundreds of millions of dollars in campaign contributions — money that often arrives through a tangle of political committees with little obvious connection to the interests actually funding them. To help voters see through that fog, CalMatters undertook a sweeping analysis of campaign finance records tied to this year’s governor’s race, the insurance commissioner contest and statewide propositions, tracing nearly $900 million in contributions back to the industries and interest groups that provided them.
The findings, now published as part of CalMatters’ Voter Guide, represent one of the more ambitious efforts to decode California’s famously opaque campaign finance system, in which donations frequently flow through vaguely named political action committees before reaching the campaigns they’re meant to support or oppose.
Reporters and data analysts examined tens of thousands of individual contribution records filed with the California Secretary of State, using the state’s Power Search database as their primary source. The goal was to identify which industries — technology, finance, agriculture, real estate and dozens of others — were actually bankrolling the state’s major political campaigns this cycle, even when the money passed through committees with generic or misleading names.
To do that, the team built a customized classification system loosely based on the federal government’s North American Industry Classification System, but tailored to California’s political landscape. They added several categories not found in standard federal coding, including one for technology firms — covering companies like Uber, Airbnb, Google and Meta, along with newer players in artificial intelligence and cryptocurrency — since no official classification currently captures that industry on its own.
Other custom categories included “Major Donor,” a catch-all for individuals who listed themselves as retired, self-employed, a student or otherwise didn’t provide information that pointed clearly to an industry; “Self-Funding,” for candidates who bankroll their own campaigns; and “Small Donors,” for contributions too modest to trace definitively.
Only contributors who gave at least $5,000 to a single campaign or proposition committee during the 2025-2026 election cycle were individually classified — a threshold the team determined was high enough to suggest a genuine business or industry connection, rather than coincidental giving.
Identifying those connections required cross-referencing multiple databases, including the federal H-1B Employer Data Hub, California’s Employment Development Department business registry, and industry classifications maintained by the nonprofit watchdog OpenSecrets. When those sources didn’t yield a clear answer, the team turned to a custom-built machine learning model trained to infer industry based on the wording of a contributor’s employer and occupation, and ultimately to two large language models — Claude and Gemini — which performed web searches to help identify unfamiliar contributors.
Even then, humans stayed firmly in the loop. Any time automated methods disagreed with one another, or when an AI model was the only source able to produce a classification, the case was flagged for manual review by a CalMatters journalist. Reviewers checked business registries, nonprofit databases and news coverage to confirm or correct each classification, and the team also ran spot checks on results that weren’t initially flagged, to catch errors the system might have otherwise missed.
Political committees posed a particular challenge, since many receive funding from a wide mix of sources rather than a single industry. In those cases, CalMatters calculated a proportional breakdown — so a committee that drew half its money from tech donors and half from insurance interests would have its contributions to a campaign split accordingly. Because committees frequently donate to one another, researchers had to trace money through as many as five layers of interconnected committees to get an accurate picture of where funding ultimately originated.
The work wasn’t without its limits. Because the analysis focused on major donors, the industry patterns identified may not perfectly reflect the universe of smaller contributions. Some individuals listed different employers across multiple donations, requiring judgment calls — tech mogul Reed Hastings, for instance, was classified under “Technology” despite sometimes listing himself as “not employed,” given his role as Netflix’s co-founder. And federal political committees, which aren’t tracked by the California Secretary of State, could only be categorized when their names or public information made their industry affiliation clear.
In all, the project sorted through nearly 60,000 individual contribution records and roughly 1,000 political committees, with CalMatters staff manually reviewing more than 2,500 classifications and nearly 4,000 contributor names to make sure donations from the same person or organization were properly linked together, even when names were recorded differently across filings.
The payoff, according to CalMatters, has already shown up in its election coverage. The analysis revealed that the Building a Better California PAC — which spent close to $400 million this cycle — funneled more than $200 million into supporting Propositions 41 and 42, measures that would make it harder to impose new special taxes and would ban new taxes on personal property. Those two campaigns drew virtually all of their funding from the PAC itself, with outside contributions amounting to barely $200,000. CalMatters’ industry breakdown showed the PAC’s money came predominantly from finance and technology donors, offering voters insight into who stands to benefit most if the measures pass.
The same methodology also informed CalMatters’ reporting on technology industry spending this election cycle, which found that tech money played a significant role in funding half of the top 10 political action committees seeking to influence state legislative races.
CalMatters has published its full dataset and classification methodology publicly, allowing researchers, journalists and the public to review how each contributor was categorized. The project drew on guidance from Foaad Khosmood, a computer science professor at Cal Poly San Luis Obispo and co-founder of the Digital Democracy project, along with CalMatters data reporter Jeremia Kimelman and Digital Democracy engineering manager Thomas Gerrity.
Original source: CalMatters
