Why are there two disclosures on a generated political ad instead of one?

Because they answer different questions and come from different bodies of law. The campaign finance disclaimer answers who paid for the communication and whether it was authorised by a candidate; it is a funding-transparency rule that long predates generative tools and attaches to public communications by committees and, in many cases, to independent expenditures and electioneering communications by anyone else. The synthetic-media notice answers how the depiction was produced; it comes from a newer set of state statutes aimed at fabricated depictions of candidates. An ad can satisfy one and violate the other, and the two notices have different wording, placement and duration requirements.

Do these state synthetic-media laws apply to an out-of-state agency or ad platform?

Generally the hook is the election, not the vendor's address. These statutes typically reach communications distributed to voters in the state within a defined window before an election that depict a candidate for office in that state. An agency in another state that builds and places the buy is producing a communication distributed there. Some statutes name the person who publishes or distributes as well as the person who creates, which is how platforms and placement vendors get pulled in, and several include exceptions for a distributor that did not know the depiction was fabricated.

Does a disclaimer cure a fabricated depiction?

Sometimes, and that is the design of most of these statutes: the label is the safe harbour. A conspicuous notice that the media was generated or manipulated is frequently what moves an ad from prohibited to permitted. But the safe harbour is drafted narrowly — it usually specifies the wording, that it be clearly visible or audible for the duration of the relevant portion, and a minimum size or reading speed. A small notice in an end card, or an audio-only ad with a text disclosure, is the common failure. And a label does not cure separate exposure: defamation, right of publicity and platform policy all continue to apply.

What about ordinary AI use — a generated background, a synthesised voiceover, an upscaled photo?

This is where the practical difficulty sits. Most statutes target a depiction of a real person, usually a candidate, that a reasonable viewer would believe to be authentic. A generated landscape, a stock-style crowd or a synthetic narrator reading approved copy is not usually a depiction of a candidate. But the boundary is fuzzy at the edges — a composited photograph of a real official, a b-roll clip with a recognisable face, an audio bed that mimics a known voice. Because the exception is judged on what a reasonable viewer would believe rather than on the production technique, the safe operating rule is to log the provenance of every asset that contains a real person.

Who is actually liable in the chain — the campaign, the agency or the tool vendor?

The committee that pays generally owns the campaign finance disclaimer; that duty is not delegable by contract. Synthetic-media statutes tend to reach further down, naming whoever creates, produces or knowingly distributes the depiction, which puts an agency and sometimes a placement vendor inside the statute rather than merely inside an indemnity. Tool vendors are usually outside the statutes but inside the contract: the indemnity, the acceptable-use terms and the marketing claims on their own site are what determine where the loss lands after the fact.

What record should exist before an ad ships?

Per asset: what tool produced it, what prompt or source material it came from, whether it depicts an identifiable real person, and who approved it. Per placement: the states and windows the buy touches, which disclosure regimes those trigger, and a rendered proof showing both notices as the viewer will see them. That record is cheap to keep during production and effectively impossible to reconstruct after a complaint, which is the only time anyone will ask for it.

Two Disclosures Have to Ride on That Ad

The Two Tracks, Side by Side

They are easy to conflate because both end up as small type at the bottom of a frame. They are not the same obligation, they are not satisfied by the same text, and compliance with one is no evidence of compliance with the other.

Track one — funding

Who paid for this, and did a candidate authorise it?

Where it comes from: Campaign finance law: federal rules for federal races, a separate state regime for state and local races, plus municipal ordinances in some cities.
What it looks like: A prescribed disclaimer with specified wording, placement and often a minimum size or duration. Small-item and impracticability exceptions exist and are narrower than people assume.

Triggered by: Paying for a public communication. Applies whether or not a single pixel was generated.

Track two — provenance

Was this depiction fabricated or materially altered?

Where it comes from: State synthetic-media-in-elections statutes, most enacted since 2023, plus platform policies that operate faster than any statute.
What it looks like: A conspicuous notice that the media was generated or manipulated, usually with statutory wording and a legibility or audibility standard for the relevant portion of the ad.

Triggered by: A depiction of a candidate or, in some states, any real person, distributed within a pre-election window.

Which Assets Actually Trip the Second Track

The statutes are not about whether a model was used. They are about whether a viewer would believe a real person did something they did not. That distinction is what makes an asset-level inventory workable: most generated material in a political spot never comes near the line, and a small number of assets sit directly on it.

Synthetic voiceover reading approved script

Depicts: No identifiable person

Usually outside the synthetic-media statutes

Still inside platform disclosure policies on several networks, and inside right-of-publicity if the voice imitates a known one.

Generated b-roll of a crowd or streetscape

Depicts: No identifiable person

Usually outside

Watch composites that include recognisable faces from source footage.

Upscaled or colour-corrected real footage of the candidate

Depicts: A real candidate

Depends on materiality

Ordinary editing is typically excepted; alteration that changes what the person appears to say or do is not. Keep the original.

Composited image placing an official somewhere they were not

Depicts: A real official

Inside, and often the paradigm case

Label per statute, or do not run it. Satire and parody exceptions exist in some states and are drafted narrowly.

Cloned voice of an opponent

Depicts: A real candidate

Inside, and separately exposed

Right of publicity and state voice-cloning statutes reach this independently of election law.

Generated 'testimonial' from a fictional constituent

Depicts: No real person

Outside election statutes, inside consumer-protection thinking

Fabricated endorsement framing draws scrutiny even where election law does not reach it.

The Window Is the Part Everyone Gets Wrong

Synthetic-media election statutes almost universally operate inside a defined period before an election, and the periods differ by state — some measured in weeks, some longer, some with a separate shorter window carrying stiffer consequences. A national buy touching many states is therefore not subject to one rule with one deadline; it is subject to a set of overlapping windows opening on different dates.

That has a practical consequence for how creative is versioned. An asset cleared in March is not cleared for the same states in October, and a "final" master that gets re-trafficked into a late buy is the single most common way an untagged depiction ends up running inside a window. Tie the disclosure decision to the flight dates in the buy, not to the approval date on the asset.

Where Each Party Sits When Something Goes Wrong

The important distinction is between being named in a statute and being exposed through a contract. The first cannot be negotiated away; the second is what your master services agreement is actually about.

Party	Named in statute?	What it actually means
The committee or funder	Yes — campaign finance disclaimer	Owns the funding disclosure outright. Cannot contract it away; a vendor error is still the committee's filing problem.
The creative agency	Often — as creator or producer	Named directly by many synthetic-media statutes, not merely liable through the client contract. Also the party whose own site makes capability claims.
The placement or ad-tech vendor	Sometimes — as distributor, usually with a knowledge element	Exposure typically turns on what the vendor knew or was told. This is why intake questions about synthetic content are worth asking and logging.
The generative tool provider	Rarely	Sits in contract rather than statute: acceptable-use terms, indemnity scope, output-rights language, and whatever its marketing pages promise about provenance metadata.
The platform	Varies	Moves faster than any regulator. Policy enforcement pulls the buy mid-flight, which is usually the first real-world consequence anyone experiences.

What Enforcement Looks Like In Practice

The formal remedies in these statutes vary — injunctive relief, private rights of action for a depicted candidate, civil penalties, and in a few states criminal exposure for knowing distribution close to an election. But the mechanism that actually bites first is almost never a regulator. It is a takedown by the platform after an opponent reports the ad, followed by a press cycle, followed by a demand letter. The buy stops, the money is spent, and the asset becomes the story instead of the message.

That ordering matters because it changes what "compliance" needs to produce. A defence that requires litigation to vindicate is worth less than a provenance record that resolves a platform appeal in an afternoon. Keep the record in the form a reviewer can read: source material, tool, prompt or edit list, whether a real person appears, and the rendered proof of both notices.

A Pre-Flight Check That Takes Ten Minutes

List every asset in the cut and mark the ones containing an identifiable real person. Only those can trigger the second track.
For each marked asset, record what was done to it and keep the unaltered original in the same folder.
Pull the flight dates and the states from the buy — not from the brief — and check which pre-election windows they fall inside.
Render a proof at delivery size and confirm both notices are legible or audible as the statute describes, in the format the ad actually ships in.
Confirm the funding disclaimer matches the committee's current registered name, not the one from the last cycle.

Your Agency Site Is Also Making Claims

"AI-powered creative", "fully compliant political advertising", "automated disclosure" — capability pages and case studies are read as representations by exactly the people who later go looking for one.

See every claim your site is making in one pass. Run a free scan and check each against what your production process actually does.

This article is general information and not legal advice. Campaign finance disclaimer rules, synthetic-media election statutes, pre-election windows, exceptions and penalties vary by jurisdiction and change frequently, and several are the subject of ongoing litigation. Consult qualified election counsel before relying on any conclusion here.

Two Disclosures Have to Ride on That Ad. Your Tool Only Knows About One.