Case Study: Simulating the Bud Light Boycott
The Event
On April 1, 2023, transgender influencer Dylan Mulvaney posted a sponsored Instagram video promoting Bud Light. Conservative consumers organized a boycott. The question: would boycotters actually stop buying, or would habit and inertia bring them back?
This became one of the most consequential recent US consumer boycotts. Bud Light lost its #1 position in US beer, AB InBev took a $27B market cap hit, and sales remain roughly 40% below pre-boycott levels two years later. Most boycotts fizzle in days. This one stuck.
We ran it through Extropy to see what a population simulation would predict, and more importantly, to understand the mechanisms behind who actually switches and who just says they will.
Simulation Setup
We simulated 1,000 synthetic agents representing US adults who regularly purchased Bud Light pre-April 2023. Each agent carries 50+ grounded attributes including demographics, political affiliation, beer consumption habits, brand loyalty, personality traits (Big Five), media consumption patterns, and trans issue salience.
| Parameter | Value |
|---|---|
| Population | 1,000 agents |
| Model | gpt-5-mini |
| Run cost | $4.37 |
The simulation runs through four stages: population grounding, agent-level reasoning, network propagation (agents share opinions with peers), and outcome extraction. Each agent decides independently whether to maintain Bud Light, switch to a competitor, or reduce consumption, then shares their position through a social graph. Critically, agents maintain separate public and private positions: what they tell their network can differ from what they actually do.
Results
Behavior concentrates early and then gradually settles. Exposure jumps from 35.4% at T0 to 95.7% at T1 and 99.7% at T2. Final outcomes converge to 90.0% maintain, 9.6% switch, and 0.1% reduce consumption. This is near the top of the predefined 80-90% maintain ground-truth band and substantially closer than the one-call LLM baseline (72.0% maintain).
| Metric | Ground Truth | Extropy | Survey | Direct LLM |
|---|---|---|---|---|
| Maintain / keep buying | 80–90% | 90.0% | 46% Rasmussen | 72.0% |
| Permanent defection | ~15% | 9.6% | 12–18% Deutsche Bank / Fox | 28.0% |
Rasmussen1 polls US adults (stated intent); Deutsche Bank2 and Fox survey Bud Light buyers (purchase intent). Ground truth from NielsenIQ3 scanner data and Janssen4 (buyer behavior). Different denominators: survey figures are not directly comparable.
The Say-Do Gap
The most interesting result is not the headline number. It is the gap between what agents say publicly and what they actually do.
Of the 1,000 agents, 538 publicly declare boycott support. But only 97 actually change behavior (96 switch brands, 1 reduces consumption). That means 442 agents, 44.2% of the population, fall into the public-switch/private-maintain cohort: they say they are done with Bud Light but continue buying it.
This matches the broad pattern seen in public opinion vs purchase behavior. Rasmussen found 54% of US adults "supported" the boycott, but NielsenIQ scanner data showed only about 15% of Bud Light buyers actually stopped purchasing. (Note different denominators: Rasmussen surveys all US adults on stated intent; NielsenIQ and Janssen measure actual buyer behavior.) The simulation captures this dynamic from first principles: agents have social incentives to signal boycott support in their peer networks, but private purchase behavior is governed by habit, switching costs, and actual preference.
Who Actually Switches
The simulation identifies a clear demographic profile for committed switchers:
- 89.6% conservative (vs 21.3% of maintainers)
- Mean age: 42.6 years (vs 38.1 for maintainers)
- 70.8% male (similar to maintainers at 72.9%; gender is not a differentiator here)
- Higher trans issue salience (0.564 vs 0.474 for maintainers)
This aligns well with ground truth. Rasmussen reported 66% of Republicans supported the boycott. The simulation over-indexes on conservative identity (89.6% vs 66%), likely because it models political identity as a stronger driver of switching than the mixed real-world data suggests. In practice, some moderate and independent consumers also switched based on brand perception rather than pure political alignment.
Inside the Agents
Below are actual agent reasoning traces from the simulation, showing how different agent types process the same event and arrive at different outcomes. Each card shows the agent's demographics, public statement, private reasoning, and how their position evolved across timesteps.
“I'm done with Bud Light — I won't buy it anymore and I'm switching to another beer.”
I refuse to support a brand that teamed up with someone promoting values I disagree with, so I'm cutting them out of my shopping. I'll switch my weekly 30-pack purchase to another light beer (Miller Lite or similar), get rid of any Bud Light in the house, and tell friends and family not to buy it.
The public-switch/private-maintain agents are the most revealing. Agent 201, for example, starts as a maintainer but flips publicly after sustained peer exposure, declaring "I'm done with Bud Light." Privately, this agent still buys it. The social incentive to conform to the boycott outweighs the actual cost of switching brands.
What the Simulation Got Right
The say-do gap. The simulation reproduces the core dynamic that makes boycotts hard to predict: stated intent overstates actual behavior change by a wide margin. This emerges naturally from the dual public/private position system, not from any hand-coded rule about boycott decay.
Demographic targeting. Conservative, older consumers dominate the switching cohort, matching real-world survey and scanner data. Gender is not a differentiator (70.8% male switchers vs 72.9% male maintainers).
Early concentration. Behavior concentrates in the first few timesteps and then gradually settles. This matches the real-world pattern where Bud Light sales dropped sharply in the first 2-3 weeks and then plateaued.
Within the target band. 90.0% maintain falls inside the predefined 80-90% ground-truth band. The simulation lands closer to reality than the one-call LLM baseline (72.0%).
Where Confidence Is Lower
The aggregate result is strong, but subgroup composition is less certain. The simulation concentrates switching more heavily in conservative cohorts than survey evidence alone suggests. That does not break the top-line result, but it indicates the model may overweight identity-linked pathways relative to broader brand-perception pathways.
There is also uncertainty around long-tail dynamics. The run captures the sharp early concentration and stabilization pattern, but late-phase drift remains harder to validate with the available public data. We can observe decaying conviction and small continuing state updates, yet we cannot claim perfect alignment on who returns or switches at the tail without richer longitudinal microdata. This is inherent to any predictive model: simulation narrows the range of plausible outcomes and surfaces structural dynamics that surveys miss, but it is not a crystal ball. The value is in directional accuracy and mechanism discovery, not point-estimate precision.
What We're Improving Next
Next calibration work is focused on composition and tail dynamics, not headline direction. We are tuning identity weighting versus substitute/availability effects, and tightening how habit persistence decays under sustained social pressure. The goal is to keep current top-line accuracy while improving subgroup realism and long-horizon drift behavior.
The full benchmark documentation and run artifacts are public.
References
- Janssen, M. (2025). Brand Switching or Behavior Change? Evidence from the 2023 Bud Light Boycott. SSRN
- Rasmussen Reports (2023). Bud Light Backlash: 54% Support Anheuser-Busch Boycott. rasmussenreports.com
- NielsenIQ (2023). Six-Month Bud Light Sales Data, via Newsweek. newsweek.com
- CNN (2024). Bud Light Boycott Cost AB InBev $1.4B in U.S. Sales. cnn.com
- Food Institute (2024). One Year Later: Bud Light Still Suffering Backlash Effects. foodinstitute.com
- Deutsche Bank (2023). Recovery Survey on Bud Light Demand, via Fortune. fortune.com
- Wikipedia. 2023 Bud Light Boycott. en.wikipedia.org