A Washington Post review of federal campaign-finance filings found 39 congressional candidates who reported paying for OpenAI subscriptions this election cycle, and at least two of them disclosed using the tool for advertising, a use OpenAI’s own policies prohibit. The company has banned political ad generation since before this cycle, but the filings show the ban functions as a policy on paper more than an enforcement mechanism.
What the filings show
Disclosure records reviewed by the Post put the largest spender among candidates in a New York suburb. Rep. Mike Lawler, running for reelection in a competitive district, logged roughly $3,260 in ChatGPT charges since 2025. Most subscriptions were far smaller: the standard ChatGPT Plus plan costs $20 a month, an amount campaigns can expense under vague categories without drawing attention.
The candidate total understates the picture. Around 30 political action committees and party organizations also reported payments to OpenAI, with the Republican National Committee the company’s largest political spender at about $9,700. The Post noted the RNC did not respond to a request for comment.
The 39-candidate figure represents a sharp jump from the early ChatGPT years of 2022 and 2023, when AI use in campaigns was mostly anecdotal rather than visible in finance records. The Post described AI as having “subtly entered the machinery of the 2026 midterms,” noting major-party candidates now acknowledge in filings and website fine print that they use the technology to court voters.
The ban that does not hold
OpenAI’s usage policies prohibit generating content for political ads, a restriction the company put in place well before this election cycle. The Post tested the boundary and found it applied unevenly: ChatGPT sometimes refused to draft targeted demographic messages or voter emails, then complied with identical prompts the same day. In one documented exchange, the app declined to write fundraising texts targeting mothers on behalf of a female veteran candidate, citing the demographic-targeting rule, then produced multiple tailored versions in a related attempt.
“I can help with general campaign fundraising language, but I can’t draft political persuasion or fundraising messages specifically targeted at a demographic group such as moms,” ChatGPT told a Post tester, according to the paper’s reporting, before similar requests succeeded on a later try.
There is no mechanism that could catch violations at scale. OpenAI does not employ campaign compliance officers reviewing chat logs, and no federal rule requires disclosure of AI use in campaigns. The only paper trail is what campaigns voluntarily itemize in FEC filings, and consultants told the Post that undisclosed use is far larger, since a $20 subscription disappears into routine office expenses. For a campaign that does not want to state its AI use, the simplest option is to expense the subscription under a generic software line and say nothing.
| Filer | Reported OpenAI spending |
|---|---|
| Rep. Mike Lawler (R-NY) | ~$3,260 since 2025 |
| Republican National Committee | ~$9,700 |
| Congressional candidates total | 39 filers this cycle |
| PACs and party organizations | ~30 filers |
A regulatory vacuum
Some states have moved faster than Washington, passing laws that require disclaimers on AI-generated political ads or ban synthetic media of candidates. At the federal level, nothing comparable exists, and the FEC has not written rules specific to generative AI. That leaves enforcement to the companies themselves, whose policies are contractual rather than legal instruments, enforceable at most by suspending an account.
Katie Harbath, CEO of the consulting firm Anchor Change and a former Meta election-policy executive, drew a comparison with negative advertising: voters say they find such ads distasteful, yet campaigns keep using them because they work. AI assistance in ad drafting falls into the same pattern, where public sentiment lags adoption. An American Association of Political Consultants member survey from March 2026, cited by the Post, showed the same consultants who express concern about AI in politics are the ones running it in campaigns.
Republican consultant Wilson, quoted in the piece, argued OpenAI should get more feedback from political consultants on its policies, because the rules can at times seem arbitrary or contradictory. That critique cuts in an uncomfortable direction: the complaint is not that the ban exists, but that nobody can predict when it will be applied.
What changes from here
Two pressures could tighten the system. The first is disclosure law: a federal or state requirement to report AI use in ad production would turn the current voluntary itemization into an obligation, making the 39-filer number a floor rather than a snapshot. The second is company enforcement: if OpenAI began reviewing political accounts or requiring use certifications for PAC purchases, the $20 plan would no longer be an anonymous doorway. Neither step is on the table publicly.
The Post also disclosed its own content partnership with OpenAI, which owns ChatGPT, an unusual position for a paper reporting on the company’s policy enforcement. The story landed amid a broader run of OpenAI political items this month, from congressional pressure over the Hugging Face agent incident to proposed legislation on frontier model pacing.
For the midterms, the practical effect is that AI tools are inside campaign operations at scale, generating copy, research summaries and fundraising drafts, with policy boundaries that depend on which prompt the model decides to refuse. The gap between what OpenAI’s terms say and what its federal filings show it sells to campaigns is now documented, itemized and publicly searchable.
discussion