When AI companions should use memories and how they decide
Before Bringing It Up: When and How AI Companions Should Use Memor
Human-Computer Interaction
Summary
AI companions can remember past interactions to feel more personal, but using that memory at the wrong time can be awkward or unhelpful. The authors studied what factors matter before an AI shares something it remembers. They designed a process called Reconsider that checks several things before deciding how to use past information. Testing showed this method improved the way AI used memory in many situations, though results varied across AI models.
What this means in practice
- •For chatbot developers: Implement the Reconsider process to improve when and how chatbots share remembered information during conversations.
- •For customer service teams: Use guidance informed by this paper to help AI respond appropriately by balancing memory use and relevance to improve user experience.
Authors
Zihan Guo, Roxy He, Junwei Quan
Abstract
Memory can sustain AI companionship, yet even accurate recollection can be inappropriate to use. Two rounds of formative interviews with 14 users (n = 6 exploratory, n = 8 memory-focused) motivate asking what a companion should consider before using past information. Eight themes inform Reconsider, a single-call procedure with five checks and four handling modes, evaluated on 80 scenarios across five models over 400 blinded within-model pairs. Two LLM judges favored Reconsider by net margins of +15 and +23 percentage points, with bootstrap intervals excluding zero for three of five models but not for GPT or Claude. Evaluator analysis linked judge scoring differences to model family, and a preliminary matched-guidance control isolating memory-specific content gave positive margins. We contribute an interview-grounded design framework for memory use and an evaluation that scrutinizes its own evaluators.