PM AI Rate Limits
(2026 Edition)
Design rate limits around the 95th-percentile user, communicate caps before anyone hits them, and default to soft limits with overage warnings rather than hard caps that surprise paying users during high-value moments. Soft limits convert better and reduce churn than hard stops; save hard caps for genuine cost protection, not as a disguised nudge to upgrade.
By Naman Goyal · Product manager · Builder of PM Streak · Updated July 3, 2026
5 principles and 4 traps for AI rate limits.
Build AI Rate Limit PM Skills — Free →5 Principles
Default limits aimed at 95th percentile usage
Communicate limits clearly before users hit them
Soft limits with overage warnings beat hard caps
Reset cadence (hourly vs daily) shapes perception
Enterprise tiers with custom limits reduce friction
4 Traps
Surprise rate limits with no warning — trust killer
Vague limits ('fair use') without specifics
Limits that hit during high-value moments (deadlines)
No way to upgrade in-flow when limits hit
FAQ
Hard or soft limits for AI products?
Soft limits with overage warnings work better in 2026. Hard caps frustrate paying users at the worst moment. Soft limits with clear pricing for overage convert better and reduce churn. Reserve hard caps for true cost protection scenarios, not nudge-to-upgrade tactics.
Keep learning