In a single US midterm cycle, spending on political ads made at least partly with AI went from a rounding error to more than $137 million. The Wesleyan Media Project has identified 220 such ads across 34 states, and only about a quarter of them disclose the AI. That is the number that matters: not that campaigns are using generative tools — everyone expected that — but that most of the output runs with no label at all, and a meaningful slice of it puts words in real people's mouths they never said. This guide is not a political take. It is a practitioner's read of what the wave actually proves, because the forces driving it — collapsing production cost, a state-by-state patchwork of disclosure law with no federal floor, platforms racing to label synthetic media, and an audience learning to distrust what it sees — now land on every creator and brand using AI, not just campaigns. It maps the verified 2026 data and its caveats, separates the two very different things people mean by 'AI political ad,' walks the regulation map honestly, and ends on the part that generalizes: the trust line that separates defensible AI content from the slop, and how to build ownership and disclosure into a content operation instead of bolting a disclaimer on after the fact.
In the 2024 US cycle, researchers at the Wesleyan Media Project could identify under $50,000 in political ad spending tied to AI. Two years later, by their October 16, 2026 tracking, the figure for the 2026 midterms is past $137 million — 220 ads across 34 states, and they say the real count is higher because AI use is often impossible to prove. That is not a gradual ramp; it is a category that went from negligible to normal in one election.
The headline the data actually supports is not 'AI ads work.' It is that most of them carry no label, and some of them impersonate real people. Only about 27% of the ads the project identified disclosed the AI. The money skews one way this cycle, the examples range from obvious parody to fabricated clips of events that never happened, and the law governing all of it is a state-by-state patchwork with no federal floor. This guide reads that wave as a working creator would — what is verified, what is a moving target, where the real line sits, and why the lesson lands on everyone making content with AI, not just campaigns.
The core source is the Wesleyan Media Project, a long-running academic tracker of political advertising. Its 2026 reporting is the most-cited count of AI in this cycle, and the figures have moved as the tally grew. A September release put the running totals at roughly $80 million across about 164–170 identified ads; the October 16, 2026 update reported more than $137 million across 220 ads in 34 states. Because the methodology depends on detecting AI — which is genuinely hard — the researchers are explicit that these are floors, not ceilings. Treat any single dollar figure as the count as of a specific date, and check the project's own releases before citing an exact number.
Two findings are more stable than the headline dollar amount. First, the disclosure gap: across the project's data, only around a quarter to a third of AI ads carried a disclaimer, and a separate CNN analysis of TV advertising reached a similar place from the broadcast side, finding most of the money behind ads using AI-generated or manipulated imagery went to spots that did not disclose it. Second, the partisan skew this cycle: by the September breakdown, roughly 80% of the identified ads were sponsored by Republican candidates or aligned groups, with a much smaller share from Democratic and issue sponsors. The skew is a fact about who adopted the tools fastest this cycle, not a permanent feature of the technology.
One more caveat worth stating plainly, because it is where secondary coverage drifts: a tracker counting 'ads that used AI' is not the same as a tracker counting 'ads that broke a law' or 'ads that deceived a voter.' The 7-in-10 undisclosed figure includes benign and parody uses alongside deceptive ones. The number tells you disclosure is rare; it does not tell you how many ads were unlawful. Keep those questions separate and the picture stays honest.
The phrase collapses two cases that deserve to be pulled apart, because only one of them is the problem everyone is actually worried about.
A campaign generates b-roll it is entitled to use, drafts and designs its own graphics, produces a spokesperson video from a script the speaker consented to, or translates its own ad into another language with a synthetic voice. This is AI doing what a production crew, a design shop, and an editing suite used to do, faster and cheaper. It can be entirely legitimate, and in most states it is legal as long as realistic synthetic media is labeled where required. The cost collapse here is the real story for small campaigns and advocacy groups: ad-grade video and imagery at a fraction of a traditional shoot.
A spot fabricates a realistic depiction of a real person — a candidate, an official, a celebrity — saying or doing something they never did, built to be believed and run without disclosure. This is the case the state laws target and the one driving the backlash. Coverage this cycle cited fabricated videos placing real public figures into invented ads using their voices and likenesses with no AI disclaimer, alongside spots depicting events that never happened. Researchers took to calling the broader flood 'slopaganda' — cheap synthetic content aimed at an emotional impression rather than a factual one. Some sponsors defended their ads as obvious satire, which is exactly why the laws below hinge on whether a reasonable person would be deceived.
Hold that split through the rest of this guide. 'Did it use AI' is the wrong question. 'Did it impersonate a real person, and did it disclose' is the one that separates a defensible ad from a deceptive one — and, as the last section argues, it is the same question that separates trustworthy creator content from the slop pile.
As of the 2026 cycle, there is no federal rule in force governing AI in political ads. The Federal Election Commission has not issued binding guidance, and Congress has not passed a baseline — a disclosure bill, the AI Transparency in Elections Act, was introduced in July 2026 but remains pending in committee, so it does not govern this election. In the vacuum, states legislated, and the result is a genuine patchwork.
By mid-2026 tracking, roughly 30 states had passed laws addressing AI or deepfakes in election content; counts differ by source (the NCSL tracker and law-firm analyses land a few apart depending on effective dates), so treat 'around 30' as the safe figure and check the current tracker for specifics. Most of these are disclosure laws — label realistic synthetic media and you comply. A few go further and prohibit certain deceptive election deepfakes outright; Maryland, Minnesota, and Texas are the commonly cited examples. The content of a required disclosure also varies: some states want only an AI label, while others ask for more detail about how and when the content was altered.
Three practical wrinkles matter more than the exact count. First, most of these laws deliberately exempt obvious parody and satire, usually by requiring that the synthetic media be realistic enough to deceive a reasonable person before the rule bites — which is why 'it's satire' is the standard defense. Second, a disclosure requirement is not a takedown requirement: in many states, labeling a manipulated ad as AI-generated satisfies the law, and the ad keeps running with that label attached. Third, the laws are being tested in court and the early results do not point one direction — federal courts have blocked election-deepfake statutes in California and Hawaii. Hawaii's law, which paired a ban on deceptive deepfakes with a disclaimer safe harbor, was permanently enjoined in January 2026 on First Amendment grounds; a California companion law was separately struck down in 2025 on federal preemption grounds (it conflicted with Section 230), while another California statute remains only partially enjoined. In other words, courts have not settled on disclosure rules being safer than outright bans — a law built around a disclaimer requirement lost in Hawaii too. The map is real but unsettled.
The operational takeaway for anyone publishing across state lines or across platforms: the requirement is not uniform, so a per-post, per-state scramble does not scale. The durable approach is to make disclosure a default of your workflow rather than a decision you relitigate every time you post. We cover the ad-specific labeling mechanics in the guides on AI-generated ad disclosure and UGC-style creatives and Google's AI-generated ad disclosure, and the step-by-step in how to disclose AI-generated ads.
Platforms are not waiting for regulators. The policy and tooling moves of 2026 draw the lines everyone making content will operate inside. TikTok auto-labels content generated through its own Symphony ad tools, attaches invisible watermarking, and embeds C2PA Content Credentials, so provenance travels with the file. Both TikTok and YouTube offer, and in cases require, a creator-facing 'AI-generated / altered content' toggle on realistic synthetic media. Google began requiring advertisers to declare AI-generated or materially altered ad creative across several ad products in 2026. Adobe's Firefly embeds Content Credentials in its output by default, turning provenance into a feature rather than a chore.
Political advertising specifically has hit a different wall: Meta restricts political, electoral, and social-issue advertisers from using its generative-AI ad creation features, even as its broader AI stack can produce political-adjacent content — a tension we break down in the piece on Meta AI and political campaign content and the honest Meta AI alternative for political and issue creators. The through-line for every creator: AI content is now assumed, labeling infrastructure is being built into the pipes, and provenance is becoming a default. If you make content with AI, plan for disclosure and provenance as standing features of your workflow, not edge cases.
The political cycle is where these norms get stress-tested in public, which makes it the clearest preview of the environment every creator and brand now works in. Three shifts transfer directly.
AI content is the baseline, and it is under scrutiny. When nine figures of political money flows through generative tools in one cycle, platforms, regulators, and audiences stop treating AI-made content as a novelty and start treating it as the norm to be labeled. Any creator using AI should assume disclosure expectations and platform labels apply to them too — not because a campaign-finance law reaches a cooking channel, but because the platforms' AI-labeling rules do.
Cost is no longer the moat, so consistency and distribution are. The same collapse that let small campaigns produce ad-grade video at volume applies to everyone. When anyone can generate a passable clip, the advantage stops being production budget and becomes whether your output is recognizably, consistently yours — and whether it actually reaches the surfaces your audience uses. Authenticity gets scarce exactly as synthetic lookalikes fill the feed, and a clearly-your-own identity becomes the thing that keeps an audience trusting what they see. This is the same dynamic behind platform crackdowns on undifferentiated AI slop.
Disclosure is turning into a credibility advantage rather than a cost. The ads drawing enforcement and backlash are the undisclosed, impersonating ones. The reflex to hide the AI is backwards: in an environment this primed to distrust synthetic media, the creators and brands that label clearly and build on an identity they own are the ones who keep trust while producing at the same speed. Transparency is becoming the moat, not the tax.
Strip the politics out and the 2026 wave resolves into a single line between content that earns trust and content that erodes it. Four properties put you on the right side of it, and they apply to a campaign, a coach, an agency, or a DTC brand equally.
Own the identity. Build on a persona, voice, and likeness you have the right to use — ideally your own or a consented, branded one — rather than a scraped or fabricated depiction of a real person. This is the exact property that separates case one from case two above, and it is a design choice, not a disclaimer you add later.
Add original value and vary the output. The thing audiences and platforms punish is undifferentiated, templated synthetic filler. A human point of view, a real angle, and genuine variation across pieces are what keep AI-assisted content out of the slop category — which is as much a monetization and reach requirement now as an ethical one.
Disclose by default. Where a platform or a law asks for an AI label, apply it as a standing rule, not a case-by-case judgment call. Realistic synthetic media of people is where this matters most. A default is auditable and scales across platforms; a per-post decision drifts.
Distribute on brand, everywhere. Reach is half the equation, and consistency across every surface is what makes an owned identity legible enough to trust. Publishing the same recognizable identity across platforms is what turns 'we use AI responsibly' from a claim into something an audience can actually see.
The hard part is not agreeing with those four properties; it is running them at volume without turning every post into a compliance review. That is a workflow problem, and it is the one Kompozy is built to solve for creators and brands on the defensible side of the trust line. The design starting point is ownership: content is generated from your own AI Influencer persona — a face and voice you control, governed by a written Persona Brief with banned-word filters — so consistency and provenance are properties of the system rather than something you police afterward. The engine produces across its full format range, from a Persona Short to carousels, quote graphics, blog posts, and newsletters (the full output set), and because they all flow from one brief, a week of content reads as one identity instead of a scatter of synthetic one-offs — the original-value-and-variation property, built in.
Disclosure and distribution are handled at the same place. Every piece clears a per-post review gate before it ships — the natural checkpoint to confirm an AI label belongs on a given post and that nothing impersonates anyone — and for the platforms that carry native AI tags, connected accounts can apply the platform's own disclosure automatically rather than you retrofitting a disclaimer onto a dozen copies of the same clip. Autopilot then schedules and publishes the approved set across the eight social platforms plus blog and email from one queue, so the same owned, labeled identity lands everywhere at once. That is the difference between treating disclosure as a feature of your pipeline and treating it as a chore you repeat per post across a patchwork of rules.
The honest scope: Kompozy is a content generation and publishing engine, not an ad-buying platform and not a tool for fabricating a real person — it will not place a paid political campaign or clone someone without consent, which is the point. For the in-auction paid creative itself, see the comparison of native AI ad creation tools; for the strategy behind the shift, the guide on AI-powered ad creation.
The $137M number is the attention-grabber, but the disclosure gap is the lesson. A single cycle proved that generative tools are now the default way political content gets made, that cost is no longer what separates a serious operation from a hobbyist, and that the output nobody trusts is the undisclosed, impersonating kind. None of that is unique to politics. Every creator and brand now works in the same environment — AI assumed, labels expected, audiences wary — and the ones who win the trust war are not the ones using the least AI. They are the ones building on an identity they own, adding a real point of view, disclosing by default, and shipping that consistent, labeled identity across every platform at the same speed the slop merchants ship theirs. For the news write-up of the midterm data itself, see AI-generated political ads top $137M ahead of the 2026 midterms.
Tracking by the Wesleyan Media Project, reported on October 16, 2026, put spending on political ads created at least partly with AI past $137 million ahead of the November 3, 2026 midterms, across 220 identified ads in 34 states. An earlier September release had lower running totals (around $80 million and roughly 164–170 ads), so the figure is a moving tally that depends on the reporting date, and researchers caution the true number is likely higher because AI use is hard to prove.
It depends on the state the ad runs in. Around 30 US states have passed laws on AI in election ads — most require a disclosure, while a few (such as Maryland, Minnesota, and Texas) prohibit certain deceptive deepfakes outright. There is no federal rule in force for the 2026 cycle, so an identical ad can be legal to run unlabeled in one state and unlawful in another. Most of these laws carve out obvious parody and satire.
A legitimate ad uses AI to produce or enhance content a campaign is entitled to make — generated b-roll, a scripted spokesperson video with the speaker's consent, designed graphics — and labels it where required. A political deepfake fabricates a realistic depiction of a real person saying or doing something they did not, with intent to mislead, typically without disclosure. Most state laws target the second case specifically, using a 'would a reasonable person be deceived' standard that exempts clear satire.
It is a term researchers have used for the 2026 wave of cheap, AI-generated political content — synthetic clips and images made to land an emotional impression or a 'deeper truth' rather than report a factual one. It runs from obvious parody to misleading deepfakes, and the common thread in the criticism is how much of it ships without disclosing that AI was involved.
Because the political wave is where the new norms get set in public. A single cycle just normalized AI as the default production method and put disclosure, platform labels, and audience distrust on the front page. The practical lesson transfers directly: build content on an identity you own rather than impersonating anyone, keep it consistent, and disclose AI where a platform or law asks. In a feed full of synthetic lookalikes, being clearly and verifiably yourself is the scarce, trust-earning position.
AI-generated political ads surged in the 2026 US midterms — the Wesleyan Media Project tracked more than $137 million in spending across 220 identified ads in 34 states, and only about 27% disclosed the AI. Most US states require disclosure but there is no federal rule, enforcement is uneven, and the real dividing line is not whether an ad used AI but whether it impersonated a real person without disclosing it.
Get started → · ← All guides · Compare Kompozy vs other tools