AI Signal 498
Repeating a system-prompt instruction improves compliance up to four repetitions, then yields no further gain
↗
Repeating a rule in the system prompt raises rule compliance from 74% with one copy to 97% with four copies, with no benefit beyond that.
Engineers can boost prompt-driven behavior without extra model tuning, but only up to a small number of repetitions. Adding more copies consumes tokens and costs money (about a dollar for the test) without improving results, so prompt length should be kept minimal.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
One repetition of the quote-style rule yields about 74% compliance, while four repetitions raise compliance to roughly 97%.
Eight or sixteen repetitions fall in the same performance range as four repetitions, offering no additional gain.
The effect matches prior research on position-dependent repetition, with stacked copies climbing then flattening.
THE CLUSTER