AI ads are popular in the 2026 midterms, but state laws haven’t stopped a lack of disclosure
What it says
The article reports on the use of AI-generated imagery in 2026 campaign advertising, opening with a Bruce Blakeman gubernatorial ad in New York that depicts AI versions of Kathy Hochul and Zohran Mamdani without an AI disclosure, which the state Democratic Party has complained about. It cites Wesleyan Media Project data identifying at least 164 AI-generated or AI-enhanced political ads this cycle, 69% of which carried no disclosure, and notes that disclosure rates are not higher in states with AI-disclosure laws. Experts Travis Ridout, Scott Babwah Brennen and Basil Smikle discuss enforcement gaps, the patchwork of state laws, and research showing AI labels can reduce voter trust in the advertiser.
Why this verdict
The piece is densely and specifically sourced: named, verifiable experts in their actual institutional roles (Travis Ridout of the Wesleyan Media Project, Scott Babwah Brennen of NYU's Center on Technology Policy, Democratic strategist Basil Smikle), an identifiable dataset with a cited caveat about measurement difficulty, and a named secondary source (National Conference of State Legislatures). It includes original reporting detail — Ridout apparently cross-referencing his data at the reporter's request — and a standard 'did not immediately respond to a request for comment' line for the Blakeman campaign. The October 2024 Brennen-coauthored finding that AI labels reduce perceived candidate trustworthiness matches published research I am aware of. Examples (Mike Rogers vs. Abdul El-Sayed in Michigan; NC state Sen. Michael Lee) are consistent with known political figures. The AI-text scorer returned essentially no markers, and the emotional-loading score is near zero; tone is measured and hedged by experts rather than hype. I am overriding the low source_trust prior: the domain inference is unverified, and the Boston Herald is an established newspaper that routinely carries syndicated wire/nonprofit newsroom copy, which this reads like (missing byline is consistent with a feed-stripped syndication). Nothing in the text contradicts established fact. Several specifics — the August complaint to the NY Board of Elections, the 164-ad count and its partisan breakdown, Blakeman's candidacy and Mamdani's mayoralty — postdate my knowledge and are unverifiable to me, which is why I am not going above the high-0.7s. Lack of corroboration in this refresh is weak evidence given the piece's narrow, data-story nature.
No meaningful indicators of machine generation. High burstiness, ten distinct direct quotes attributed to three separately identified experts with institutional affiliations, no list-template structure, and zero LLM boilerplate phrases. The quotes contain conversational disfluency and hedging ('what are the norms surrounding this, what are the laws surrounding this?') that generated filler rarely reproduces, and the article includes reporter-initiated data work rather than keyword-assembled padding.
Claims examined
-
unverifiable
A Bruce Blakeman gubernatorial ad used AI-generated images of Hochul and Mamdani without an AI disclosure, while carrying a 'paid for by Blakeman for New York' line.
Specific, recent ad description the reporter appears to have viewed directly; plausible and internally consistent but not independently checkable here.
-
unverifiable
The New York State Democratic Party filed an August complaint with the state Board of Elections over Blakeman's AI ads.
Concrete, attributable, falsifiable claim of a kind routinely reported; no reason to doubt but beyond my ability to confirm.
-
unverifiable
The Wesleyan Media Project identified at least 164 AI-generated or AI-enhanced political ads this cycle, 69% without disclosure, 80% from Republican sponsors.
Attributed to a real, well-established ad-tracking project and a cited press release, with the source's own caveat about undercounting — consistent with how WMP publishes data.
-
supported
Scott Babwah Brennen directs NYU's Center on Technology Policy and co-authored an October 2024 report finding AI disclaimers made candidates seem less trustworthy, with a 'backfire effect.'
Matches published research and institutional affiliation I can corroborate from established knowledge.
-
supported
Tennessee's AI law covers only deepfakes impersonating candidates, and no federal AI-ad disclosure law or regulation exists.
Consistent with the known state-by-state patchwork and congressional/FEC inaction on AI political-ad rules.
-
unverifiable
Mike Rogers faces Abdul El-Sayed for an open Michigan Senate seat; Michael Lee is an NC state senator facing Democrat Jessica Bichler.
Both named figures are real and the matchups are plausible for 2026, but the specific race states postdate what I can verify.
Signals
| Signal | Strength | Score |
|---|---|---|
|
Source trust
Reputation prior for the publishing domain. A starting point, not a verdict.
|
|
0.28 |
|
Corroboration
How many independent outlets carried a matching story, discounted when they are all low-tier.
|
|
0.14 |
|
Specificity
Density of named people, direct quotes, attributed statements, figures and dates.
|
|
0.80 |
|
AI text markers
Stylometric markers of generated filler. Higher means more machine-like. A weak prior, never proof.
|
|
0.10 |
|
Emotional loading
Clickbait, outrage and hedging language. Higher means more loaded.
|
|
0.02 |
|
Byline
Whether a named human author is credited.
|
|
0.00 |
Also carried by
1 other domains in the current window carried what appears to be the same story: usnews.com.
Caveats
- The article is dated September 2026, after my training cutoff; the central news hook (Blakeman's candidacy, the Democratic Party complaint, the 164-ad tally) cannot be directly verified by me and is treated as unverified rather than false.
- The tier-4 source rating is explicitly flagged as inferred from the domain name and not from a registry; I weighted the text's sourcing quality over that prior, and a convincing impersonation of a syndicated news piece on a spoofed domain is the main scenario my verdict would miss.
- No byline was found, which slightly limits accountability, though this is common in syndicated copy stripped by feeds.
- No independent outlet in this run carried the same story, so corroboration is untested.
Provenance
| Assessed by | claude-opus-5 |
| Text available | Full article body |
| Found via | Search: campaign ad attack candidate 2026 (discovery feed) |
| First seen | Oct 24, 14:05 EDT |
| Composite signal prior | 0.41 |