Progressive messaging works with the choir. We need to reach beyond it.
“What happens if you message to someone’s psychology instead of to their demographics?”
Demography isn’t destiny.
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.
Finding: The Backlash
The staff-written progressive frame was the only message that lost union support, and it lost the most ground with the people who decide elections.
Question shown: “Do you support or oppose the right of security officers to join a labor union?”
The staff-written frame was the only message that lost overall union support. Both AI frames held, at 65 and 64.
The staff-written frame moved them 9 points below the placebo group.
The staff-written frame moved them 10 points below the placebo group. The Personal Safety frame moved them to 57.
“WHAT HAPPENS IF YOU MESSAGE TO SOMEONE’S PSYCHOLOGY INSTEAD OF TO THEIR DEMOGRAPHICS?”
Finding: Hold the Base, Grow the Tent
over the staff-written frame, across the conservative values segments
With liberals and Democrats, the AI-generated frames performed about the same as the staff-written frame.
Union support. Green = best AI-generated psychographic frame · Red = staff-written progressive frame · Baseline = the placebo group who read an unrelated message. Significance as flagged by the vendor: 90% confidence versus the placebo. Respondents were general-population adults; the trial measured attitudes, not votes.
Officer pay, 44 to 66, Personal Safety frame against the placebo group.
Officer pay, 54 to 76, Personal Safety frame against the placebo group.
Officer pay, 54 to 73, Personal Safety frame against the placebo group.
We already know this works.
Relational Organizing / Deep Canvassing
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 UNSCALABLEOff-the-Shelf LLMs
Research 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 UNACCOUNTABLEThe Gap: Operationalization
We invest millions in research and message development. But implementing those learnings is riddled with pitfalls.
With the Hoplight Engine
- Deploy effective frames to all users instantly
- Generate bespoke variants without drift
One input can return 37 platform-ready copy types.
Success is deploying those best practices with high fidelity at scale
Four Auditable Layers
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.
What This Unlocks:
RECURSIVE MESSAGE REFINEMENT
Move the people other programs leave out.