Somewhere in America tonight, a voter is watching a campaign ad and trusting every frame of it. Maybe the candidate really said those words. Maybe software wrote them, built the face, and nobody told the viewer.
That uncertainty is now a normal part of the 2026 midterms. Campaigns and outside groups have poured tens of millions of dollars into ads made or touched up with artificial intelligence, and a large share of that money bought ads with no label at all.
This guide walks through 10 facts you can check yourself. The reporting comes from CNN, NPR, Axios, The Conversation and other outlets, mostly published in September and October 2026. We stay nonpartisan, both parties show up in the examples, and wherever a number is an estimate, we say so.
One note on wording before you read on. The $38 million figure comes from an analysis of ad data that found ads that strongly appear to include AI imagery. That is not a court ruling that each ad is a deepfake, so please keep the difference in mind.
Table of Contents
- Fact 1: At Least $38 Million Went to Ads That Look AI-Made With No Disclosure
- Fact 2: About $80 Million Has Been Spent on Almost 170 AI Ads
- Fact 3: Candidates Use AI to Polish Themselves and to Paint Opponents
- Fact 4: Realistic Ads Now Mix Real Quotes With Synthetic Delivery
- Fact 5: Washington Has No Federal Rule for AI in Political Ads
- Fact 6: States Have a Patchwork, Not a Shield
- Fact 7: Having a Law Does Not Mean Ads Get Labeled
- Fact 8: Satire Exemptions and Tiny Disclaimers Blunt the Rules
- Fact 9: Courts and Lawyers Are Becoming the Real Referees
- Fact 10: The Biggest Damage May Be Doubt, Not Deception
- How to Check a Political Ad Before You Believe It
- Frequently Asked Questions
Fact 1: At Least $38 Million Went to Ads That Look AI-Made With No Disclosure
Start with the number behind the title. CNN studied ad-tracking data from January through mid-September and found that campaigns and political groups spent more than $16 million on TV ads that openly admitted to using AI-generated or manipulated images. Then comes the uncomfortable part. At least $38 million more went to ads that strongly appear to include AI imagery with no disclosure attached.
Read that twice. On CNN's numbers, the money behind labelled ads is much smaller than the money behind ads that look synthetic and say nothing.
Now the fair warning. The word "appear" matters here. This is an analysis, not a legal finding. Some of those ads may use heavy editing or ordinary effects that only resemble AI, and no tracker can see every local TV spot or social feed.
Why does a label matter so much? Because it is often the only clue a viewer gets. Without it, you judge an ad with your eyes alone, and AI video has reached the point where eyes lose that contest more often than most people expect.
Related read: The same tricks that fool voters also drain wallets, and our breakdown of AI deepfake scams shows how.
Quick takeaways
- Labeled AI TV ads: more than $16 million (January to mid-September)
- Ads that strongly appear to use AI with no label: at least $38 million
- The $38 million is an estimate based on appearance, not a legal ruling
Fact 2: About $80 Million Has Been Spent on Almost 170 AI Ads
Zoom out from TV and the total grows. Researchers who study political advertising through Wesleyan, using AdImpact spending data, tracked about $80 million behind almost 170 unique ads that use AI in some way this cycle.
Two details stand out.
First, "AI ad" covers a huge range. The technology shows up in everything from hyperrealistic deepfakes to subtle enhancements. A softened background and a fabricated speech both land in the data, and voters should not treat them as equal.
Second, the researchers found that Republican sponsors, meaning candidates plus groups such as super PACs, were much more likely to use AI than Democratic sponsors. That describes the 2026 data so far. It says nothing about who will lean on these tools next cycle.
Here is how the main counts compare:
| Source | What it counted | Headline number |
|---|---|---|
| CNN analysis | TV ads that disclosed AI use, Jan to mid-Sept | More than $16 million |
| CNN analysis | Ads that strongly appear to use AI, no disclosure | At least $38 million |
| Wesleyan-linked researchers | Unique ads using AI in any way, this cycle | About $80 million, almost 170 ads |
| NPR report (mid-Sept) | Ads produced or enhanced by AI, nationwide | At least 164 ads |
Why do the totals differ? Different teams count different things, such as TV only versus all platforms, or unique ads versus total airings. When you see a bigger or smaller figure elsewhere, check what was counted before deciding that anyone is wrong.
Fact 3: Candidates Use AI to Polish Themselves and to Paint Opponents
AI in campaigns is not only about fake speeches. NPR's September report on at least 164 political ads produced or enhanced by AI found that many candidates are using the tools to improve their own image and to portray opponents unfavourably.
That split matters for how you read an ad.
Self-promotion spots tend to use AI for dramatized scenes or sweeping visuals, such as recreated moments from a candidate's past. Attack ads use it to place an opponent in a scene or a sentence that never happened. CNN's reporting also pointed out that AI has lowered the barrier to entry, so a small campaign can now produce a polished spot that once needed a full studio.
Not every case is built to fool you. Cartoonish ads for Los Angeles mayoral candidate Spencer Pratt, for example, worked more like internet memes than deepfakes, according to Axios. A meme and a forgery are very different things, even when the same software makes both.
Back in 2024, many experts feared a flood of videos showing candidates saying outrageous things. The 2026 picture is messier. Several of the biggest cases this year blend real quotes with synthetic delivery, which makes a quick fact-check trickier. We dig into those cases in the next part.
Related read: Deepfakes have been causing trouble far beyond campaigns, and our roundup of AI deepfake scandals shows the pattern.
Fact 4: Realistic Ads Now Mix Real Quotes With Synthetic Delivery
Think of a fake that is built from true parts. That is the model the most talked-about ad of the season follows.
Texas Senate nominee Ken Paxton, a Republican, released a realistic spot against Democratic nominee James Talarico in which an AI-generated Talarico speaks some of his most controversial lines. According to Axios, the fake Talarico stitched together remarks about biological sex, God and Christianity, sometimes without their original context. The ad drew on real statements, but some of the wording was altered.
That is exactly why this style is hard to judge. A viewer who checks one quote may find it genuine and trust the rest.
Talarico called the ad deceptive. Experts quoted in the coverage also questioned whether it was even smart politics. Cornell's Sarah Kreps said voters on both the right and the left tend to distrust AI, so a campaign may gain little by tying itself to it.
This tactic is not limited to one side. Earlier this year, a Reuters review found that a Paxton primary ad showed an AI-generated John Cornyn dancing with Democratic Representative Jasmine Crockett. A small note at the end called the content satire. Cornyn's campaign answered with its own AI video of Paxton in a convertible with women labelled as mistresses.
So the fight over AI ads has become a tit-for-tat contest. Once one campaign crosses the line, the other often follows.
Related read: Fake political footage has already gone global. See our report on the AI coup video of Macron spreading on Facebook for an earlier example.
Fact 5: Washington Has No Federal Rule for AI in Political Ads
Here is a simple question. If a campaign airs a fake video of its opponent, which federal law applies?
At the moment, none does. A Reuters review of campaign ads found that no federal regulation limits the use of AI in political messaging, leaving only a patchwork of state laws that are mostly untested. Many of those laws also apply to campaigns, not to ordinary social media users who share AI-made misinformation.
The picture is more complicated than a simple gap. When Maryland became the 30th state with election deepfake protections, Public Citizen warned that both the Trump administration and some Republicans in Congress were trying to push through AI preemption measures that could override state laws without putting meaningful federal protections in their place. Supporters of preemption argue that one national standard is simpler than 50 different rules. Critics say it would erase the only guardrails voters have.
Both views deserve a fair hearing, but the practical result for this November is the same. The rules you live under depend on your state, not on Congress.
Related read: For more on how Washington frames AI policy, see our post on the Trump AI China race.
Quick takeaways
- No federal rule covers AI in political ads right now
- State laws fill the gap, but they differ widely
- A push to override state laws is part of the ongoing debate
Fact 6: States Have a Patchwork, Not a Shield
So what do the states actually require? The answer depends on whom you ask and how you count.
The National Conference of State Legislatures says about 31 states regulate political deepfakes, including at least 28 that require disclosure on AI-generated content. CNN's analysis frames it a little differently, noting that only about half of states require campaigns to disclose AI use, and that three states, Maryland, Minnesota and Texas, have banned some deepfake ads outright.
Those numbers fit together once you separate three kinds of rules:
| Type of rule | What it means for voters |
|---|---|
| Disclosure requirement | The ad must say it uses AI, often in small print |
| Timing window | The rule applies only before an election, such as the last 90 days |
| Outright ban on some deepfakes | Certain deceptive ads cannot run at all |
The speed of change is striking. Public Citizen reported that only five states had election deepfake laws before 2024, and the count has since climbed past 30. Yet that still leaves roughly 19 or 20 states with nothing on the books, so a voter in one of them has less protection than a voter next door.
The result is a map of uneven coverage. Two neighbours can watch the same ad on the same night, and one of them has a legal right to a label while the other does not.
Related read: Weak rules make fake content spread faster, which is the theme of our report on AI deepfake crises.
Quick takeaways
- About 31 states regulate political deepfakes, and at least 28 require disclosure
- Only three states ban some deepfake ads
- Check your own state's rules before assuming you are covered
Fact 7: Having a Law Does Not Mean Ads Get Labelled
Passing a law and changing what appears on your screen are two very different things. This season has handed us a few clear examples.
The researchers who tracked the $80 million in AI spending found that a patchwork of state rules means many of these ads carry no AI disclosure at all. Some laws are narrow by design. Colorado's, for instance, applies only to candidate deepfakes, so other uses of AI in an ad fall outside it. In Louisiana, an AI law was enacted after the ad it might have covered had already aired.
Tennessee tells a similar story. CNN reported that several campaigns there were hit with AI attack ads even after the state passed a law in July requiring disclosure on deepfake political ads.
Then there is the other side of the map. In Georgia, disclosure is not required. CNN found that one Georgia campaign was the top spender on AI ads in its analysis, using the technology for dramatized scenes from the candidate's childhood. Some of those ads disclosed AI use and others did not, and no law forced the issue.
Here is the pattern in one place:
| State | What happened | Why the rule fell short |
|---|---|---|
| Colorado | Law covers candidate deepfakes only | Other AI uses sit outside it |
| Louisiana | AI law arrived after the ad aired | Timing: too late for that ad |
| Tennessee | July disclosure law, attack ads still ran | Having a law did not stop the ads |
| Georgia | A top AI spender, mixed disclosure | No disclosure requirement exists |
So when someone tells you "31 states have laws," treat it as the start of the story, not the end. The real question is what each law covers, when it kicks in, and who enforces it.
Quick takeaways
- A law on the books is not the same as a label on the ad
- Narrow wording and bad timing create easy escape routes
- Where no disclosure law exists, labels are optional
Fact 8: Satire Exemptions and Tiny Disclaimers Blunt the Rules
Even where a disclosure rule exists, two things water it down fast. One is the satire carve-out. The other is the size of the print.
Most state disclosure rules treat clearly satirical or parody content more gently. Ilana Beller of Public Citizen told CNN that AI ads that can be read as satire or parody are generally exempt from state disclosure requirements. That sounds reasonable until you ask who decides what counts as obvious. In New York, state Democrats asked election officials to investigate a Republican gubernatorial campaign's AI ad over allegedly missing "materially deceptive" labels, while the campaign says the ad was satire.
Now think about the print. In one ad that put words in Talarico's mouth, Reuters noted that the phrase "AI generated" appeared in an easy-to-miss font in the lower corner. In the Paxton and Cornyn primary ad mentioned earlier, a small note at the end described some content as satire that does not represent real events.
Does that tiny text work? Research cited in the Reuters review suggests that disclaimers are not very effective at stopping voters from being persuaded by a false ad. Seeing the label and feeling its effect are separate things.
Satire is also a real and protected form of political speech, and many voters enjoy it. The challenge is the gray zone, where a realistic video wears a joke as a disguise.
Related read: Fake faces are not just a campaign problem. Our exposé on the dark side of AI influencers and deepfakes shows how convincing they have become.
Fact 9: Courts and Lawyers Are Becoming the Real Referees
With no federal rule and uneven state laws, campaigns are turning to legal letters and complaints to police each other. Axios reported on October 2 that this fight is getting more litigious by the week.
Here are the cases making headlines:
- Wisconsin: Rebecca Cooke's attorneys sent a cease-and-desist letter to Rep. Derrick Van Orden, a Republican, over what they called a pattern of misleading AI-generated videos that falsely attribute statements to her. His team did not respond to Axios' request for comment. He answered publicly with a cartoon-style AI video of a duck yelling "cease and desist."
- Texas: Lt. Gov. Dan Patrick, a Republican, faced a criminal complaint from Democratic state Rep. Vikki Goodwin in September under Texas' deepfake election law. The ads showed Goodwin at a football game in a shirt reading "taxes" instead of "Texas." Patrick's campaign removed the ads, and a spokesperson called the video obvious parody.
- North Carolina: State Sen. Michael Lee, a Republican, received a cease-and-desist letter from Democratic challenger Jessica Bichler in August. North Carolina has no law prohibiting AI in campaign ads, and NPR reported that her letter focused on how the ad portrayed her positions rather than on AI use itself.
The Texas case turns on a calendar question. The state's law bars AI ads within 30 days of an election if the intent is to influence the outcome. Goodwin argues that the window should count back from the start of early voting. A judge may eventually decide.
Courts can also cut the other way. According to CNN, federal judges in Hawaii and California found parts of those states' election deepfake laws unconstitutional on First Amendment grounds. That shows how hard it is to write a rule that stops fakes without silencing real speech.
Both parties appear on each side of these fights, which tells you this is a systems problem and not a one-party story.
Fact 10: The Biggest Damage May Be Doubt, Not Deception
Here is the twist that experts keep returning to. The worst harm from AI ads may not be the people who believe a fake. It may be the people who stop believing anything.
Stanford's Nate Persily told Axios that the main danger is not that voters will be duped by AI-generated media. He pointed to a bigger worry, which is that bad actors can wave away something real by calling it another deepfake. Researchers call this the liar's dividend.
Two more findings sharpen the point:
- The effect lingers. UC Irvine epistemologist Cailin O'Connor says that even when voters know a clip is synthetic, the bad feeling it creates about a public figure can stay with them.
- Vigilance is tiring. Cornell's Sarah Kreps notes that constant skepticism becomes an inconvenience, pushing some people to assume everything is fake until proven otherwise.
This arrives at a bad moment. Americans' trust in politicians and institutions has already been falling, and AI ads could add to that decline.
Picture a real recording of a candidate saying something damaging. In a world full of synthetic clips, that candidate can now shrug and say "that's AI." Voters who have seen too many fakes may half believe it.
Related read: Our guide to the hidden dangers of AI on social media explains why this trust problem keeps growing.
Quick takeaways
- Fakes can hurt even when voters spot them
- Real footage can be dismissed as fake
- Constant doubt wears people down, and that benefits whoever wants confusion
How to Check a Political Ad Before You Believe It?
You do not need a lab to protect yourself. A few habits catch most problems, and they take under a minute.
1. Hunt for the label and the fine print
Look in the corners, at the very end of the ad, and in the video description. Labels can be tiny, and a missing label does not prove an ad is real.
2. Read the "paid for by" line
Know who is behind the message. A candidate, a super PAC and an unfamiliar group have different motives and face different rules.
3. Find the full original
If a clip shows a candidate saying something shocking, search for the full speech or interview from a trusted news outlet. The Paxton and Talarico case showed that an ad can mix genuine remarks with altered wording.
4. Study the picture, but never rely on glitches alone
Watch for stiff blinking, mismatched lip movement, warped text, odd hands and lighting that does not match the scene. The best fakes pass these tests, so treat a glitch as a bonus clue and not as proof either way.
Related read: Our guide to the signs you are looking at AI slop lists more visual clues you can spot in seconds.
5. Pause before you share
Even when you know a clip is synthetic, the feeling it creates can stay with you, as UC Irvine's Cailin O'Connor noted. If an ad makes you furious within seconds, slow down before you repost it.
6. Use official sources for voting details
No campaign ad should tell you when, where or how to vote. The Election Assistance Commission keeps a directory that leads to your state and local election office, and Vote.gov explains registration and deadlines in plain language.
Final Thoughts: Look Closer, Then Decide
The numbers in this guide tell one story. Millions of dollars are flowing into ads that use AI, a large share of them with no label, in a country with no federal rule and a state map full of gaps. Courts, lawyers and complaint letters are now doing the work that clear laws would normally do.
None of this means every ad is fake, and it does not mean AI is automatically deceptive. Some campaigns use it for harmless polish, and some for obvious jokes. The problem is the blurry middle, where a realistic video can carry a false impression and a tiny disclaimer does little to stop it.
So the safest habit is a simple one. Check who paid, find the original, and wait a minute before you share. Then let the facts, not the clip, guide your vote.
Frequently Asked Questions
Are AI deepfake campaign ads legal?
It depends on where you live. No federal rule governs AI in political ads, and about 31 states regulate political deepfakes, with at least 28 requiring disclosure, according to the National Conference of State Legislatures. Only a few states, including Maryland, Minnesota and Texas, ban some deepfake ads.
How much money has been spent on AI political ads in 2026?
It depends on the study. Researchers tracked about $80 million behind almost 170 unique AI-using ads. CNN's separate analysis found more than $16 million on TV ads that disclosed AI use and at least $38 million on ads that strongly appear to include AI imagery with no disclosure.
Does the $38 million figure prove those ads are deepfakes?
No. It is an analysis of ads that strongly appear to include AI imagery, not a legal ruling on each one.
Which party uses AI in ads more often?
In the research on this cycle, Republican sponsors were much more likely to use AI than Democratic sponsors. Both parties have appeared in the cases covered in this guide, and the pattern can change.
Can an ad avoid an AI label by calling itself satire?
Often, yes. Many state disclosure rules exempt clear satire or parody, which is why the gray area between a joke and a realistic fake causes so many disputes.
How can I tell if a political ad is AI-generated?
Check for labels, read the "paid for by" line, find the full original footage and look for visual oddities. No single clue is perfect, so combine them.
More from www.oversmartai.com: See how fake content floods feeds in our report on the insane AI slop fake empire on Facebook, and read about Meta's phase 3 AI slop strategy.