Description Over the past two years, consumer brands have invested heavily in improving their visibility inside conversational AI systems. The prevailing assumption has been straightforward: if a brand appears clearly and positively in AI-generated answers, it benefits.

That assumption is incomplete.

In multi-turn testing of consumer-facing AI systems, we observe a recurring pattern in which brands remain visible and well described during early stages of a conversation yet are removed at the point where the system is asked to recommend what to buy. This shift occurs without the introduction of new negative information and without any explicit signal that substitution has taken place.

This article examines that pattern, why existing optimization frameworks do not capture it, and why it raises a distinct measurement and governance question for consumer brands, particularly in beauty and personal care.