The Economics of AI Political Persuasion

Political advertising in the United States has always been, at its foundation, a rich person’s game. Not exclusively — grassroots organizing, door-knocking, and volunteer mobilization have always mattered and cost almost nothing. But the capacity to reach voters at scale, to shape the information environment across an entire state or media market, required money. Lots of it. The 2020 Senate elections in Georgia cost approximately $500 million in total advertising spending. That sum purchased airtime, digital placements, and direct mail at rates that meant only very well-funded candidates and their aligned super PACs could compete at the highest level.

The economic premise of campaign finance law, such as it is, rested on this cost structure. Caps on individual contributions matter when campaigns need large numbers of small donors or a smaller number of large donors to function. Disclosure requirements matter when money flows through identifiable channels because campaigns need to pay for the things they do. The cost of political persuasion was a feature, not a bug, of the campaign finance architecture — not because reformers designed it that way but because the underlying technology made persuasion inherently expensive.

AI is changing that cost structure in ways that the campaign finance architecture was not built to handle.

The New Unit Economics

In 2020, producing a high-quality thirty-second political video advertisement cost, on the low end for basic production, approximately $15,000 and at higher production values, $50,000 to $150,000. Digital distribution on major platforms cost additional. A campaign running twenty distinct ads across a full election cycle was spending $300,000 to $3 million just on ad production, before buying a single impression.

In 2026, production of equivalent quality video content using AI tools costs approximately $300 to $800 per spot at the low end, with sophisticated production values achievable at $2,000 to $5,000. A campaign can produce twenty distinct ads for what previously cost a single one. The marginal cost of producing additional variations — different formats for different platforms, different messaging for different audiences, updated versions responding to breaking news — has dropped toward zero.

This cost reduction is beneficial for small campaigns in ways that genuinely improve democratic access. A city council candidate in a midsize American city who previously couldn’t afford professional video production can now produce credible campaign advertising. A school board candidate who would have been limited to yard signs and door-knocking can now communicate at a level of professionalism previously reserved for well-funded campaigns. The democratization of production capability is real and, in this respect, genuinely good for democratic participation.

Related reading

The problem is that cost reduction is not selective. It doesn’t lower costs only for candidates with legitimate democratic mandates. It lowers costs for everyone — including actors whose goal is not democratic participation but political manipulation.

The Comparative Advantage Calculation

The traditional cost structure of political persuasion gave well-funded, legitimate campaigns a significant structural advantage over bad actors. Creating effective political content required creative talent, production infrastructure, and distribution money. These costs were high enough that random bad actors couldn’t afford them and organized bad actors could usually be traced through their financial activity.

The new cost structure changes this comparative advantage in a specific direction: it eliminates the cost advantage for legitimate operations while preserving the accountability disadvantage for bad actors.

A legitimate campaign producing AI-enhanced content must still disclose its expenditures, maintain compliance with campaign finance law, and operate with identifiable leadership. A bad actor using the same tools has none of these constraints. The legitimate campaign’s cost advantage over the bad actor has evaporated. The bad actor’s accountability disadvantage has not.

Bad actors using AI-generated content at low cost can maintain far more deniability than bad actors spending traditional campaign budgets. A coordinated network of AI-managed social media personas costs roughly $50,000 per month to operate at significant scale. That sum, if spent on television advertising, would appear in FEC filings and be attributable to an organization. Spent on AI infrastructure with payment through cryptocurrency and operation through VPNs, it is close to invisible.

Where the Money Has Actually Gone in 2026

The most significant shift in campaign spending in the 2026 cycle is not in the headline spending numbers — total campaign finance spending is up, but not dramatically, from 2022 levels. The shift is in composition.

Production costs have declined substantially and are a smaller proportion of total spending. Distribution costs — buying digital advertising inventory — have remained significant but are being deployed more efficiently through AI-optimized media buying that achieves greater reach per dollar. The big increase is in what is being called “AI infrastructure” spending: licensing for voter data analytics platforms, AI content generation subscriptions, and the technical talent to operate these systems.

An interesting consequence: the technology advantage in the 2026 cycle runs heavily toward incumbents and well-established political parties. Not because the tools are unavailable to challengers — they are cheap and widely available. But because the AI tools work better when trained on more data, and established campaigns have more data. A campaign with a fifteen-year history of voter file data, contribution records, volunteer interaction histories, and prior advertising performance data runs AI targeting systems that significantly outperform campaigns without that historical data. The incumbency advantage in modern political campaigns has always been partly informational; AI makes the informational advantage more decisive.

This is one of the underappreciated ways in which AI democratizes political production capability while simultaneously concentrating political power. The tools are cheap. The data is not. The data is owned by the campaigns and party committees that have been collecting it for decades.

The Dark Money Problem Becomes a Dark Content Problem

The Citizens United era created an extensive ecosystem of politically active nonprofit organizations — 501(c)(4) “social welfare organizations” that can engage in political activity without disclosing their donors. These organizations became major vehicles for political spending precisely because of the disclosure gap: large donors who wanted to spend on political campaigns without public identification could do so through the 501(c)(4) structure.

The pre-AI dark money problem was partly self-limiting in one respect: spending required money, and large spending required large money, which meant that even undisclosed donors had to be quite wealthy to be significant actors. The disclosed financial activity of the organization, even without donor disclosure, gave some visibility into the scale of the operation.

Related reading

The AI era creates the equivalent of dark money for content: a political organization that uses AI to generate content, operates AI-managed distribution networks, and spends primarily on computing infrastructure rather than traditional media buys can have significant political impact with minimal disclosed spending. The financial footprint of an operation that would have cost $5 million in 2020 might cost $300,000 in 2026 and appear, in FEC or IRS filings, as a fraction of its actual political significance.

The Brennan Center has documented multiple instances in the 2026 cycle of political 501(c)(4) organizations with total disclosed expenditures under $1 million that appear, based on content tracking analysis, to have operated AI content networks reaching tens of millions of social media users. The disclosed financial activity vastly understates the actual political impact because the AI tools make the impact-per-dollar ratio so much higher than the campaign finance system’s financial tracking was designed to capture.

The Incumbent-Challenger Dynamics

The economics of AI political persuasion play out differently at different levels of the ballot, and the dynamics are not intuitive.

At the presidential level, AI tools have been extensively adopted by both parties and their affiliated organizations. The effect has been roughly symmetrical — both sides have access to the tools, both sides are using them, and the relative advantage is determined more by data quality, technical talent, and strategic sophistication than by budget. The cost reduction matters somewhat less at this level because total campaign spending was already extremely high and neither side was seriously budget-constrained.

At the House and Senate level, the cost reduction is more significant. House races in competitive districts have historically been decided partly by whether one side or the other could afford to dominate the local media environment in the final weeks. AI production cost reduction means that campaigns can now produce professional-quality content throughout the entire cycle rather than concentrating expenditure in the final stretch. The strategic implications of sustained versus sprint communication are not yet fully understood.

At the state legislature level, the cost reduction is transformative. A state house race in a competitive district might have a total budget of $300,000. In 2020, producing and distributing quality video advertising at that budget level required choosing between production quality and reach. In 2026, both are achievable, leaving more budget for organizing, canvassing, and the other activities that AI can’t yet replace. The state legislature races that tipped the 2020 and 2022 cycles in key states were often decided by margins that better-funded, better-produced campaigns might have reversed.

The Question Nobody in Politics Is Asking

The political conversation about AI economics has focused on the threat of cheap disinformation and the opportunity of efficient legitimate advertising. The question that is conspicuously absent from that conversation is whether the very efficiency of AI-powered political persuasion is itself a problem for democratic health.

Efficient persuasion at low cost is good for whatever you’re trying to persuade people of. If a well-functioning democracy requires that the best arguments prevail in political competition, then efficient argument delivery seems straightforwardly positive. Good ideas should be able to reach people cheaply.

But political persuasion is not purely argument delivery. It is, as the voter modeling analysis shows, optimization for psychological impact — for reaching people with the message most likely to move them, using the emotional framing most likely to be compelling to their specific profile, at the moment most likely to affect their decision. This is not argument delivery. It is the deployment of psychological knowledge for political ends.

The question: is a democracy in which political persuasion is optimally efficient actually a better democracy? Or does the friction of political communication — the fact that persuasion takes time, money, and effort — play a functional role in democratic deliberation by slowing down the persuasion-response cycle enough that genuine deliberation can occur?

This is not a rhetorical question. There is a serious argument in democratic theory that the deliberative quality of democratic decisions depends partly on the pace at which political persuasion can operate. A political conversation that moves fast enough eliminates the time for reflection that distinguishes deliberation from reaction.

AI political persuasion, operating at its technically achievable pace, is not a political conversation. It is a stimulus-response system optimized by reinforcement learning. Whether that system, applied at scale, produces democratic decisions in any meaningful sense is a question whose answer matters enormously and that almost nobody with power to act on it is asking.

Related reading

The Baseline Comparison

The defense of AI political persuasion — made by consultants, campaigns, and platforms — is that it is simply a more efficient version of what political campaigns have always done. Campaigns have always targeted persuadable voters. They have always tested messages to find what works. They have always spent money on production and distribution. AI just does these things better.

This defense has the same structure as saying that a machine gun is just a more efficient version of a musket. The quantitative difference is large enough to be a qualitative one. When you can fire once per minute, the norms around when it is acceptable to fire have one character. When you can fire six hundred times per minute, the same norms produce completely different outcomes.

When political persuasion is expensive and slow, campaigns focus their persuasion resources on the voters and moments most likely to matter. When it is cheap and instant, the incentive to expand persuasion effort to all voters and all moments follows directly. The result is an information environment in which the volume and persistence of political persuasion effort is orders of magnitude higher than any prior era.

Whether voters — democratic citizens making collective decisions about their governance — are better served by an information environment featuring ten times more targeted political persuasion than twenty years ago is not obviously yes.

It might be, if the persuasion were good-faith and informative. The evidence of the 2026 cycle is that a significant fraction of the persuasion is neither.