“What happens if you message to someone’s psychology instead of to their demographics?”
And it may not even be the optimal targeting methodology.
Pushing progressive-coded moral language on non-progressive audiences backfires with the voters who decide elections.
Persuasion messaging from progressive human communicators underperformed the placebo message about Morton Salt.
Issue tested: “Many security officers are poorly paid and resourced.”
The progressive frame moved them 5 points below placebo.
The progressive frame moved them 4 points below placebo.
“It’s better for them to hear about Morton Salt than to hear from us.”
“WHAT HAPPENS IF YOU MESSAGE TO SOMEONE’S PSYCHOLOGY INSTEAD OF TO THEIR DEMOGRAPHICS?”
across all conservative segments
when using AI-generated psychographic messages
Green = psychographic message · Red = human-generated persuasion message · Baseline = the Morton Salt placebo
These tactics succeed by meeting people inside their worldview. A canvasser who listens, adapts, speaks in the listener’s values: that works.
But these tactics are prohibitively expensive to scale.
PROVEN BUT UNSCALABLEResearch shows us large language models are good at persuading people. But if the model changes, the message changes. No audit trail. No persistent replicability. No institutional knowledge.
A billionaire’s black box isn’t a scalable strategy.
SCALABLE BUT UNACCOUNTABLEWe invest millions in research and message development. But implementing those learnings is riddled with pitfalls.
One input can return 37 platform-ready copy types.
Success is deploying those best practices with high fidelity at scale
Built on established psychometric measurement traditions. The four layers are the public map, while the proprietary pipeline spans eight stages.
Measures who the listener is psychologically. Takes signals in and outputs a factor vector.
Projects the profile into felt worldview: archetype, antagonists, fears, and internal story.
Maps worldview directly to frame, narrative, metaphor, register, and appeal type.
LLM writes prose to spec. The moat is the spec it receives from Layers 1-3.
Move the people other programs leave out.