An AI tool can fill 90 days of slots in under a minute. That's the problem.
A calendar that compounds isn't a list of topics, it's a forecast. Each slot answers three questions: which pillar does this serve, which campaign does it support, and which metric should it move? Slots that can't answer all three are filler, and filler decays positioning faster than no posts at all.
The structure that works: pick 3 5 pillars tied to your ICP and value prop, assign a weekly cadence per pillar, then let the AI populate inside those rails, never outside them. The AI handles variety within a pillar; you handle the pillar choice.
Re-score every 4 weeks. Drop the pillar with the lowest saves-per-post and replace it. Keep the engine narrow, keep the inputs sharp, and the calendar starts working for you instead of the other way around.