Analysis
Axios reported on August 5 that the 2026 midterms are shaping up as the first election where AI is core campaign infrastructure rather than an edge-case threat. Two uses stand out: generating persuasion content at a volume no human staff could produce, and running message tests against simulated voter panels instead of live polling.
The simulated-electorate practice is the more consequential of the two and the least regulated. Traditional polling costs tens of thousands of dollars per wave and takes days. Modeling a synthetic electorate and testing a hundred message variants against it costs almost nothing and returns results in minutes. Campaigns adopt whatever is cheaper and faster, and the accuracy question -- whether a language model's guess about a voter resembles the voter -- gets answered after the election rather than before.
“Cheap synthetic panels mostly change how campaigns allocate ad budget, and deepfake scares have consistently underperformed predictions in the elections since 2024.”
The regulatory picture is thin. The FEC has not issued binding rules on AI-generated campaign content. A patchwork of state disclosure laws covers synthetic media in political ads with wildly varying scope and enforcement. Meanwhile the Trump administration's AI framework, reviewed this week with OpenAI, Anthropic and Microsoft, has not been released publicly -- so the federal government's own evaluation standards are not available to the people writing election rules.
The counterweight to the alarm: the durable effect here is probably operational rather than deceptive. Cheap synthetic panels mostly change how campaigns allocate ad budget, and deepfake scares have consistently underperformed predictions in the elections since 2024. The measurable harm so far is degraded polling quality and a flood of low-quality content, not a stolen election.
What to watch: whether the FEC issues guidance before November, whether any campaign discloses that its internal polling ran on synthetic respondents, and whether platform policies on AI-generated political content survive the volume they are about to face.