The Unhinged PR Playbook: A Tactical Guide to Controlled Chaos

The brands generating the most AI citations right now are not running the safest campaigns. They are running the weirdest ones. Muck Rack's May 2026 analysis found that 84% of AI-cited links come from earned media, the sources brands neither own nor pay for, and in Kevin Indig's analysis of more than 7,000 citations, brand search volume correlated with AI chatbot mentions at 0.334, the factor that stood out most among the metrics he tested. Unhinged PR is built to produce both at volume. Here is how to deploy it without losing control.

The ROI Case for Controlled Chaos

Polished campaigns generate polished results. Chaotic campaigns generate outsized ones. The data is not ambiguous.

Duolingo killed its mascot in September 2025. Within two weeks, the "Dead Duo" stunt generated 1.7 billion earned-media impressions, 450 news articles (60% from top-tier outlets), and a 38% lift in app downloads. Daily brand mentions spiked from 11,000 to 59,900. The cost structure for the campaign was negligible compared to a traditional media buy.

Burger King's Whopper Detour campaign returned a reported 37:1 ROI, producing 3.5 billion earned media impressions and $40 million in earned media value over its nine-day promotion, according to FCB New York's case. Chili's opened a fake "Fast Food Financing" shop next to a Manhattan McDonald's in 2025 to launch its Big QP burger, and Chili's parent company told investors that the pop-up generated more impressions than its Big Smasher launch.

These are not brand-awareness stunts. They are earned-media factories. And in 2026, earned media is the input layer for AI citations. Muck Rack's May 2026 What Is AI Reading? report analyzed more than 25 million cited links from ChatGPT, Claude, and Gemini. About 84% of citations came from its broad earned-media category; journalism alone supplied about 27% of cited links. When a stunt generates 450 articles, each of those articles becomes a potential citation source for every AI engine running a retrieval loop on the topic.

Why Unhinged Beats Polished for AI Visibility

The mechanism is straightforward. AI engines score sources on three criteria: entity clarity, third-party corroboration, and extractability. Unhinged campaigns compress all three.

Entity clarity spikes. When Duolingo kills its mascot, every headline names Duolingo. When IHOP temporarily rebrands to "IHOb", every article names IHOP. A polished product launch might earn five headlines. A weird one earns five hundred. Each headline is a named-entity signal that AI engines ingest.

Third-party corroboration multiplies. A standard press release might land three placements. The IHOb rebrand generated 36 billion earned impressions and quadrupled burger sales. Unhinged stunts do not need a distribution strategy because the distribution is built into the weirdness. Journalists cover it because it is a story, not because you pitched it.

Correction, November 8, 2026. An earlier version repeated The Digital Bloom's ChatGPT multiplier for brands mentioned on four or more platforms. Its report states the figure with no method, dataset or primary study, so it is removed rather than relayed.

Extractability is inherent. A stunt creates a concrete, specific, quotable event. "Duolingo killed its mascot and got 1.7 billion impressions" is exactly the kind of clean, extractable claim an AI engine can pull into a citation. Compare that to "Our Q2 brand campaign drove awareness across key demographics." The first sentence gets cited. The second one gets ignored.

The Machine Relations framework I use at AuthorityTech connects earned media directly to AI citation outcomes. The insight here is that unhinged PR does not bypass that pipeline — it floods it. More earned coverage means more third-party sources for AI engines to retrieve, score, and cite.

The Controlled Chaos Framework

"Unhinged" does not mean reckless. Every campaign that generated the ROI numbers above was engineered. Here is the operating framework I use when advising on high-impact PR.

1. Pick one absurd contrast. The best stunts create a single jarring juxtaposition. IHOP becomes IHOb. Duolingo kills its mascot. Chili's opens a fake payday lender next to McDonald's. Blendtec puts an iPhone in a blender. One contrast, one headline. If you need to explain the stunt in more than one sentence, it is too complicated.

2. Engineer the share mechanic before the stunt. Duolingo did not just kill Duo — they created a cascade: the announcement, the user-generated tributes, the merchandise, the resurrection arc. Graza olive oil put fake lost-pet flyers on city streets with QR codes. The stunt must contain its own distribution trigger. If people cannot participate, photograph, or react, it stays local.

3. Set the blast radius. Define your acceptable risk surface in advance. Duolingo's team ran at "80% unhinged, 20% wholesome" for three years before recently recalibrating. The ratio matters. You need enough chaos to generate coverage, enough control to stay on-brand. I recommend starting at 60/40 for brands that have never done this — enough to surprise, not enough to apologize.

4. Build the measurement stack. Track earned media impressions, brand mention velocity, brand search volume lift, and — most importantly — AI citation presence for your branded queries within 30 days of the stunt. In Kevin Indig's analysis of more than 7,000 citations, brand search volume was the factor most strongly correlated with AI chatbot mentions (0.334). If a stunt spikes your branded searches, it is moving the signal that tracks AI visibility most closely in that data.

5. Time the stunt to the news cycle. Duolingo rode the TikTok ban news to a roughly 216% year-over-year rise in new U.S. Mandarin learners. A24 launched the Marty Supreme blimp campaign during the Christmas movie season. The best stunts attach to an existing conversation. They do not create one from scratch.

The Budget Math

The objection I hear most: "We do not have Duolingo's budget." You do not need it.

Blendtec's "Will It Blend?" series cost $50 per video to produce and generated 6 million YouTube views in five days with a 700% sales lift. Graza's lost-pet flyers cost a few hundred dollars in printing. The Whopper Detour was a geo-fenced mobile feature, not a Super Bowl buy.

The economics work because earned media is free distribution. You pay for the stunt. The media covers it because the stunt is a story. Each article that results is a source AI engines can cite. Stacker and Scrunch's 87-story, 30-brand study reported a 239% median citation lift from distributing stories through Stacker's network of third-party news outlets. Earned media from a stunt is third-party distribution at scale, at near-zero marginal cost.

At AuthorityTech, we run a results-only model for exactly this reason. The earned media placement is the asset. The AI citation is the downstream compound.

What Not to Do

Three failure modes I see repeatedly:

Chaos without a thesis. If the stunt does not reinforce a specific brand claim, you get attention without equity. The IHOb campaign worked because it made people talk about IHOP's burgers — the actual product they were selling. A random viral moment that does not connect to your business claim is a meme, not a strategy.

Overfitting to social virality. Social engagement and earned media are not the same thing. A TikTok video that gets 10 million views but zero press coverage generates zero new sources for AI engines to cite. The goal is editorial articles, not engagement metrics.

One stunt, no system. Duolingo did not do one weird thing. They built a three-year content engine around a chaotic persona. The sustained approach grew them from 40 million to 100 million users while spending almost nothing on traditional advertising. Single stunts decay. A system compounds.

The AI Visibility Connection

Here is the part most PR teams miss entirely. Earned media from stunts does not just generate short-term buzz. It creates a permanent source footprint that AI engines retrieve and cite for months or years after the initial coverage.

Traffic from generative AI platforms grew 796% in two years, from January 2024 to December 2025, in WebFX's analysis of more than two billion site sessions. When your stunt generates 450 articles, those articles do not disappear after the news cycle. They sit in the retrieval corpus of every AI engine, available to be cited the next time someone asks a related question.

This is why I track share of citation as the primary metric at AuthorityTech. The question is not "did the stunt get coverage?" It is "did the stunt increase the percentage of AI-generated answers that cite this brand?" That metric connects the stunt directly to revenue through the AI discovery channel.

FAQ

What is an unhinged PR playbook?

An unhinged PR playbook is a structured framework for deploying unconventional, high-contrast campaigns that generate outsized earned media coverage. The "unhinged" part is the creative execution — stunts, absurd contrasts, participation mechanics. The "playbook" part is the controlled measurement and risk framework that keeps the chaos productive. Brands like Duolingo, Burger King, and Chili's have used this approach to generate billions of earned media impressions at a fraction of traditional campaign costs.

Does unhinged marketing work for B2B brands?

The mechanics are identical. B2B buyers are humans who read the same outlets, use the same AI search tools, and respond to the same attention triggers. The contrast needs to be relevant to your buyer's world — a SaaS company opening a fake "legacy software repair shop" at a conference works the same way Chili's fake financing shop worked. The metric that matters is whether the stunt generates earned coverage that AI engines cite when your buyers ask questions in your category. Brand search volume was the factor most strongly correlated with AI chatbot mentions in Kevin Indig's analysis of more than 7,000 citations; Indig's data does not split B2B from B2C.

How do you measure the AI visibility impact of a PR stunt?

Track four metrics independently: earned media article count, brand mention velocity (daily mentions before vs. after), branded search volume lift (Google Trends or GSC data), and AI citation presence for your target queries within 30 days. The last metric is the one most teams skip. Go to ChatGPT, Perplexity, and Google AI Mode, search the queries your buyers use, and check whether your brand now appears in the answer. At AuthorityTech, I use share of citation to track citation presence per query cluster.

Who is Christian Lehman?

Christian Lehman is a campaign strategist and editorial voice at AuthorityTech, the AI-era PR firm founded by Jaxon Parrott. He writes about earned media execution, campaign measurement, and the tactical frameworks operators need to build AI visibility through the Machine Relations discipline.

Correction, November 11, 2026

This page cited 5W's State of AI Citations 2026, a synthesis of earlier studies, for a count of tracked AI citations, for the claim that the sources engines cite are overwhelmingly editorial coverage, and for a finding on how few AI-cited links rank in Google's top 10. The count belongs to Profound's dataset, not to 5W, and the studies do not show that cited sources are overwhelmingly editorial: Muck Rack's study, which the page now cites, found most cited links came from sources brands neither own nor pay for, of which journalism is one part. The section on social virality no longer uses the overlap finding, which did not bear on its point, or the claim that engines build their source lists from editorial coverage rather than social platforms, which the studies do not support.

This page also credited Burger King's Whopper Detour results to a guerrilla-marketing roundup that does not mention the campaign, and put its earned media value above the figure reported; the results now cite FCB New York's case as Direct New Ideas published it, which gives the impressions, the earned media value and the return over the nine-day promotion. It credited the impressions from Chili's Fast Food Financing pop-up to a StoryChief roundup that no longer states them, and dated the pop-up to 2026. It ran in 2025, and the page now gives what Chili's parent company told investors, that it drew more impressions than its Big Smasher launch. And it credited Duolingo's rise in new Mandarin learners to a Brand24 post that does not state it; Duolingo gave the figure to TechCrunch in January 2025, and the page now cites that report.

Correction, November 12, 2026

This page said traffic from generative AI platforms grew nearly eight hundred percent year over year, and linked a statistics roundup for it. The figure is WebFX's: in its analysis of more than two billion site sessions, generative AI traffic grew by that amount over two years, from January 2024 to December 2025. The sentence now cites WebFX's study and gives its period.