Skip to main content

The Engagement Goes Off the Roof

Manipulation Breakdowns · 10 min read · By D0

A King, a Cheesehead, Two Babies Fighting Over a Toy

In the past several weeks, accounts belonging to Illinois Republican gubernatorial candidate Darren Bailey and his running mate Aaron Del Mar have posted a small library of AI-generated images and videos depicting Governor JB Pritzker: as a king on a throne, as an “evil oligarch,” dressed in Green Bay Packers gear with a cheesehead and a Brett Favre jersey, lounging on a beach as a “shady billionaire” funneling tax dollars to Hyatt, vacationing in the Bahamas over “offshore trusts,” moving out of the Governor’s Mansion with his belongings loaded onto a truck outside the Capitol, and — in one video — as an animated infant wrestling Chicago Mayor Brandon Johnson over a toy. Somewhere in the rotation, Pritzker also appears overweight, smoking, and in what one report described only as “toilet regalia.” The campaign’s logo has been rendered as the Bat-Signal over the Chicago skyline. A “say no to data centers” button was generated to match.

None of it is trying to convince anyone that Pritzker actually sat on a throne. That’s the point worth sitting with, because it means the usual manipulation-breakdown question — is this fooling people — is the wrong one to ask here. Nobody watching a cartoon governor brawl with a cartoon mayor over a toy believes they’re watching documentary footage. The mechanism operating on the viewer isn’t deception. It’s something else, and Del Mar named it himself, on the record, without being asked to.

“The Engagement Goes Off the Roof”

Asked why the campaign leans so heavily on AI-generated content, Del Mar didn’t reach for a message about voter education or issue clarity. He described a feed metric. “When you start posting some of these AI images that you’re describing, the engagement goes off the roof,” he told Capitol News Illinois. He added that the images cost less to produce than content from a human graphic designer — a second, related point, not an aside.

Put those two sentences together and you have the entire strategy stated plainly by the person running it: content that performs better on the algorithm, produced at a fraction of the cost of the thing it replaced. Nothing in that description requires the content to be true, or even to be believed. It requires only that it move through a feed faster than the alternative, and the alternative — a single human designer producing one image at a time — was never going to keep pace with an unlimited number of AI-generated variations tested against live engagement numbers.

This is worth separating cleanly from the deepfake cases Decipon has covered elsewhere, where campaigns generate synthetic audio or video of an opponent saying something they never said, wrap it in a small-print AI disclosure, and let the visual realism do the work of implying authenticity even after disclosure. Bailey’s content doesn’t attempt that. A king on a throne isn’t a claim about a fact. It’s closer to editorial cartooning than to fabricated testimony — which is exactly the category the campaign wants it filed under, and exactly the claim worth testing.

Two Labels, One Fight

Neither campaign is fighting over whether this content exists. They’re fighting over what to call it, because the label is doing real work in both directions.

Del Mar’s framing: “We look at those kind of like political cartoons that have been around for well over 100 years.” Political cartoons are a protected, well-understood genre — exaggeration in service of commentary, a tradition running from Thomas Nast to any newspaper’s op-ed page. Invoking it borrows a century of legitimacy and First Amendment protection in a single sentence.

Pritzker’s spokesman Alex Gough countered with a different label entirely: “We believe that is more compelling than Darren Bailey’s AI slop videos.” “Slop” does the opposite work — it recodes the same content as low-effort, algorithmically-farmed junk, the internet-era term for mass-produced synthetic filler nobody asked for.

Both labels can be applied to the identical set of images, and both are doing more than describing — they’re pre-loading how an audience should react before that audience looks closely. That’s worth naming as its own category of maneuver: institutional camouflage doesn’t only run one direction. An operator reaches for a legitimizing label (“political cartoon”) to armor content against scrutiny; a critic reaches for a delegitimizing one (“slop”) to strip that armor back off. Neither label, on its own, describes the mechanism actually operating on a voter’s feed. That mechanism is volume, not genre.

What a Political Cartoon Used to Cost

The “100 years of political cartoons” comparison is worth taking seriously enough to find where it breaks, because the break is the actual story.

A newspaper cartoon in 1926 or 1996 came with a built-in rate limiter: one artist, drawing by hand, whose work passed through an editor before it reached print. A paper ran one editorial cartoon a day, maybe two. That constraint wasn’t incidental to the genre — it was load-bearing. It meant every cartoon that reached an audience had survived a human editorial judgment call about whether the joke landed, whether the exaggeration crossed from commentary into something else, whether it was worth the paper’s name attached to it. The cost of production was also, functionally, a quality and accountability gate.

AI generation removes the rate limiter and the gate in the same move. Del Mar’s own reasoning — cheaper than a designer, and testable against real-time engagement data — describes a production model with no equivalent to an editor’s judgment anywhere in the loop. The metric that decides which image survives isn’t “does this land as commentary.” It’s “did this get more engagement than the last one.” A political cartoon optimized against an engagement dashboard, with unlimited daily output and no per-unit cost, is not a faster version of the old genre. It’s a structurally different one wearing the old genre’s name, precisely because the name still carries protection the new version was never built to earn on its own terms.

The Disclosure Law That Doesn’t Care Whether It’s AI

Illinois considered a law this year that would have required specific AI disclosure on political ads — written statements on graphics, a spoken statement on audio, both on video — sponsored by state Sen. Mary Edly-Allen as part of a broader election reform package. That bill was filed in the spring session, never advanced, and missed the mid-June deadline that would have let it apply to the 2026 election. As of this writing, Illinois has no AI-specific disclosure requirement for political content. Bailey’s campaign is not violating a law that does not exist.

That’s the answer to the question most coverage of this story reached for. It’s not the more interesting question sitting underneath it. Illinois election law already requires — independent of any AI-specific provision, and regardless of how the content was produced — that political advertising disclose who paid for it. A state election board spokesman put it plainly to reporters: the law “does not differentiate campaign content created with or without AI,” and all of it requires proper funding disclosure. Reporters reviewing Bailey’s posted images found no such disclosure on them.

That gap deserves precision, not overstatement. Whether an individual organic social media post — as opposed to a paid advertisement — falls under the same disclosure statute is a real legal question this reporting doesn’t resolve, and campaign finance disclosure requirements often turn on exactly that distinction. What can be said cleanly: the disclosure debate playing out in public is about a bill that failed and therefore imposes no obligation yet. The disclosure question that’s actually live is a boring, pre-AI transparency requirement nobody is currently discussing in connection with these images at all — which is itself worth noticing. A fight over a hypothetical AI law can absorb all the attention that a real, existing, unresolved compliance question would otherwise draw.

What This Is Not

Caricature of a sitting governor is core protected political speech, AI-assisted or not, and treating exaggerated mockery as inherently manipulative would prove too much — it would indict two centuries of editorial cartooning along with it. Nothing here establishes that any individual image is illegal, that the campaign has broken the funding-disclosure law rather than merely testing its boundary, or that voters are being deceived about who is depicted or what is being claimed about them. Del Mar has been unusually candid about the strategy rather than concealing it, which cuts against reading this as a covert operation.

What’s demonstrated is narrower and still worth the scrutiny: a campaign has stated outright that its content is selected and produced for algorithmic performance rather than commentary quality, has reached for a legitimizing genre label that doesn’t obviously survive contact with how the content is actually made and tested, and has done so inside a disclosure environment where the one rule that would clearly apply — regardless of AI — doesn’t appear to have been followed, while the rule that would specifically target AI conveniently doesn’t exist yet.

Key Findings

  • The goal was stated, not inferred. Del Mar’s “engagement goes off the roof” is a direct description of algorithmic optimization as the design target — not persuasion through belief, but amplification through the feed.
  • Cost removal is the mechanism, not a footnote. Eliminating the human-designer cost also eliminates the editorial judgment that historically rate-limited political cartooning and gave the genre its accountability.
  • Two competing labels are doing rhetorical work, not description. “Political cartoons” and “AI slop” are both pre-loaded reactions applied to the same content — neither one names what’s actually different about it: volume and cost, not genre.
  • The live disclosure question isn’t the one getting attention. The failed AI-specific disclosure bill imposes no legal obligation; the pre-existing, AI-agnostic funding-disclosure requirement is the one reporters found apparently unmet, and it’s drawing far less scrutiny.
  • Candor coexists with the tactic. Del Mar’s openness about strategy doesn’t neutralize the disclosure gap or the engagement-optimization design — it just means this operation isn’t hiding behind the deception that other cases in this series depend on.

Implications

The deepfake cases the Influence Tactics Protocol usually flags share a dependency on realism: the manipulation needs the audience to at least momentarily register the content as authentic, or the whole mechanism collapses. This case doesn’t have that dependency, which makes it a preview of a durable second track rather than a subtype of the first. As detection improves and audiences grow warier of anything that looks too real, content engineered to be obviously fake — but produced at unlimited volume, tuned against live engagement data, and wrapped in a century-old genre’s legitimacy — doesn’t need realism to work. It needs only reach, and reach is exactly what an unconstrained production pipeline is built to maximize.

That should shift where scrutiny goes. Detection tools built to catch synthetic realism will find nothing to flag in a cartoon that was never trying to pass as real. The signal worth tracking instead is structural: has the human cost and editorial judgment that used to gate a genre been removed, and has volume become the strategy in its place. Disclosure law drafted around “is this AI” answers a narrower question than the one this case actually raises, which is closer to: who is checking any of it before it ships, and does the disclosure law that already exists still apply once nobody is.

Conclusion

Nobody needs to believe Pritzker sat on a throne for this to work. They only need to see him there often enough, across enough variations, tuned against engagement data with no per-unit cost to slow it down, wrapped in a label borrowed from a genre that used to come with a human editor attached. The AI disclosure law that might have applied here never passed. The plain funding-disclosure law that was never about AI at all — the one that doesn’t care what tool made the image — is the one nobody’s asking about yet.

This article is part of Decipon’s Manipulation Breakdowns series, examining specific influence operations through the Influence Tactics Protocol.

Sources: