What is a conversational frequency cap?

Why the display model does not transfer

Display frequency capping counts impressions per user per period, and a pageview is a natural boundary: each one is a discrete event with its own slots.

A conversation has no such boundary. It is one continuous exchange that may last minutes or span days, and a brand appearing three times within it does not read as three impressions — it reads as the assistant pushing something. The unit that matters is the conversation, and often the session, rather than the render.

What gets capped

• Per conversation. Usually one brand appearance, occasionally two in a long exchange.
• Per session, across conversations in the same sitting.
• Per user per period, so a brand does not follow someone across a week.
• Per category, which is less obvious and matters: three different CRM vendors appearing in one conversation reads as advertising even though no single brand repeated.

Why it protects revenue rather than costing it

Frequency capping on display is a brand-side concern — advertisers cap to avoid wasting impressions and irritating people. On an AI surface it is primarily a publisher concern, and it is a yield decision rather than a courtesy.

An assistant that shows advertising repeatedly within a conversation stops feeling like an assistant. Users disengage, sessions shorten, and the supply of monetisable moments falls. The cap protects the asset generating the inventory, which is why publishers set it centrally and do not sell exceptions.

This is the same logic that governs prompt inventory eligibility: on this surface, restraint and revenue point in the same direction over any horizon longer than a quarter.