What the filings show
A Washington Post review of Federal Election Commission disclosures found that 39 US congressional candidates reported paying for OpenAI subscriptions during this election cycle. Political action committees and parties reported roughly 30 further payments, with the Republican National Committee the largest single spender at about $9,700. Among individual candidates, Representative Mike Lawler of New York reported around $3,260 since 2025.
At least two candidates stated in their filings that the subscription had been used for advertising. OpenAI’s usage policies prohibit candidates from using its tools to generate political ads. A third disclosed using AI for drafting and personalising messages or creating synthetic media.
The gap between policy and practice
The Post found that OpenAI enforces its restrictions unevenly, with the assistant sometimes declining a request it had granted on another occasion. A policy applied that way functions as friction rather than as a rule: it raises the cost of a prohibited use without preventing it, and it teaches users to rephrase rather than to stop.
The disclosed figures also understate the real number. A monthly ChatGPT subscription is small enough to be expensed under a vague category, and consultants told the Post that far more campaigns use the tools than say so. One Republican digital strategist described the common uses plainly: writing emails, writing ad copy, writing scripts.
Why campaigns hide it
The incentive to stay quiet is not regulatory, it is reputational. Voters distrust political messaging they know to be AI-generated, so disclosure carries a cost with no offsetting benefit. Consultants expect campaigns to push more of this work out to super PACs, which are further from the candidate and further from the disclosure requirements that do exist.
That is the direction of travel worth noting. The response to a norm against AI-written political material has not been less AI-written political material; it has been more distance between the material and the campaign that commissioned it.
What to watch
As of mid-2026 the FEC had not finalised binding disclosure rules for generative AI in political advertising, leaving a patchwork in which some states require disclaimers on AI-generated political ads and federal law does not. Until that changes, the constraint on campaign use of these tools is a vendor’s terms of service, enforced by the vendor, against customers who pay it. The 2026 midterms in November are the first federal election to run entirely under that arrangement.
