TL;DR: We swapped a seven-touch cold outbound sequence for a single high-context castle prompt delivered on day five. After 14 days, the variant reply rate finished +132% over control with p < 0.001 at 0.92 power across 2,418 prospects.
The hypothesis
Cold outbound on LinkedIn is usually a contest of volume: more sends, more follow-ups, more carefully polished openers. We wanted to test whether a single fortified inbound surface - a castle the prospect steps into - could beat that cadence without increasing touches.
If the buyer receives one useful, self-contained prompt instead of a drip sequence, then reply intent should compound after the intervention rather than decay with each follow-up.
The setup
- Cohorts. 1,209 prospects in control and 1,209 in variant, stratified by company size and persona.
- Control. Seven-touch cold outbound over 14 days: opener, nudges, soft ask, breakup.
- Variant. Six dormant days followed by one personalized castle prompt on day five. No follow-up.
- Metric. Cumulative reply rate measured daily, gated on intent classifier ≥ 0.6.
What we saw
Through day four, both cohorts tracked together near baseline. Once the castle prompt shipped, the variant curve separated and never returned. By day ten it had lapped control twice; by day fourteen the lift was statistically decisive.
The castle did not replace outbound volume with clever copy. It replaced the ask with a place: a compact, high-signal surface that made responding feel like the next natural step.
What ships next
We are productizing the castle prompt inside Agentic Social Media Manager and expanding the test across Reddit, Google, LinkedIn, and ChatGPT surfaces.
