73% of AI Political Ads Shipped Unlabelled: Meta, Google and TikTok Now Pay to Verify Ads They Didn't Make

Two tallies, three weeks apart. As of Oct 15, the Wesleyan Media Project counted 220 AI-generated political ads carrying more than $137 million of spend; by the final count published on Nov 3, the book was nearly 300 ads and more than $187 million. Roughly 36% more money in three weeks, and the disclosure rate moved the wrong way: the campaigns bought more AI creative and labelled less of it.

Scale this against the platforms that carry it. The entire $11.9 billion political market is 4.9% of Meta Platforms's $241.6 billion of trailing revenue and 2.5% of Alphabet's $471.1 billion. The AI slice alone is 0.08% of Meta's trailing revenue and 0.04% of Alphabet's. One quarter of YouTube advertising — $11.21 billion — is roughly the size of the entire US political ad market. And the measurement itself is a floor: the disclosed book is a lower bound because unlabelled ads are invisible to it.

The three models are not different policy, they are different accounting. Meta put the classifier on its own P&L and inherited the error cases with them. Google kept the check in front of the buyer, so the attestation and the disclaimer sit with the advertiser who signs them. TikTok removed the transaction and paid for a feed-wide labelling system instead. In every case, each platform has quietly chosen who pays for a wrong label.

The missing labels are not negligence, they are arithmetic. The Interactive Advertising Bureau's August 2026 AI Transparency and Disclosure Framework cites an NYU Stern finding that telling consumers an ad was made with generative AI cuts its click-through rate by 31.5%, and reporting on fully AI-generated display ads puts their unlabelled click-through about 19% above human-made creative. Label the ad and you hand back roughly a third of the performance the AI bought you. Disclosure is priced at 31.5% of clicks, so the incentive to skip it is arithmetic, not ideology.

Meanwhile the federal route is jammed. The Senate's 60-vote threshold, the filibuster and the presidential veto make a single national disclosure standard unlikely before 2029, and the administration has continued to try to preempt state AI laws in the courts. So the practical regime through the 2028 cycle is a patchwork:compliance stays with the advertiser until a state attorney general or an agency forces it back, and the platform's own policy is the de facto federal rule.

Follow the money and the AI political ad business resolves into three layers. Upstream are the generators whose fingerprints Meta now reads, and the provenance standard that makes a creative's history machine-checkable: the Interactive Advertising Bureau's framework makes C2PA metadata the machine layer, with human reviewers applying the materiality threshold on top. In the middle sit the verification vendors — DoubleVerify earned $205.6 million of revenue in Q3 2026, and Nielsen agreed to buy it for $13.60 a share. Downstream are the payers: outside groups spending $3.9 billion, and the inventory owners who take the dollars. The labs that supply the creative are now buying political ads with it, which is the reflexivity that turns a compliance story into a political one.

Democrats won the House and the Senate on Nov 3, 2026 — the first unified control since 2023 — with at least 245 House seats projected and Senate flips in Iowa, Kansas, North Carolina, Ohio, South Carolina and Texas. The new Congress inherits two agencies with authority over ad claims, the FTC and the FCC, plus the possibility of a House select committee on AI with subpoena power in the first half of 2027. What it does not inherit is the ability to pass a disclosure statute: 60 votes, the filibuster and the veto. Oversight moves to hearings and enforcement, not statutes, which means the cost lands on process and records rather than on new legal duties.

Start with the arithmetic that decides this. The AI political book of roughly $187 million is 0.08% of Meta's trailing revenue and 0.04% of Alphabet's. No earnings line moves. What does move is the discount rate: Meta Platforms carries platform-run detection and the mislabelling liability that comes with it, Alphabet keeps attestation with the advertiser, and a Democratic committee in 2027 will ask each of them for the records. Verification will be priced in the multiple, not the margin.

Over one to three years, the structural implication is a standardisation of the cost. Provenance metadata and disclosure thresholds become table stakes for any platform that wants political advertising money, and the platforms that already built the tooling — Meta's detector, Google's verification gate, TikTok's labelling at feed scale — hold the advantage in a world where federal preemption fails and state rules multiply. The main risk to that thesis is not regulatory but commercial: a campaign that can generate a hundred ad variants for the price of one makes a per-ad compliance cost irrelevant to the buyer, so the platforms that treat verification as a feature can price it, while those that treat it as a cost centre absorb it.

Facts first. More than $187 million of AI-made political ads ran on nearly 300 spots in 2026, up from under $50,000 of identified AI political advertising in 2024, and 73% of them carried no AI disclaimer because labelling costs 31.5% of clicks. Meta detects and labels third-party AI creatives, Google gates political purchases and pushes attestation onto advertisers, TikTok refuses the political ad market and labels feed-wide instead, and X leaves the duty with the advertiser. Democrats now run both chambers, there is no federal disclosure law, and roughly 30 state regimes do the work. Inference: the cost of that gap is charged to trust and legal exposure rather than to revenue, so it shows up in multiples and in the platform that chooses to detect rather than attest. Speculation: if labels ever become a paid quality signal, the verification layer gains a revenue line it does not have today, and the platform with the best detector wins the political book. The investable conclusion is narrow and testable: the cost line is trust and legal exposure, not ad revenue.