Before Bringing It Up: When and How AI Companions Should Use Memor

2026-10-07

AI Summary

This study addresses the issue of inappropriate memory utilization in AI companions. Based on formalized user interviews, it pioneers a memory usage design framework and proposes the Reconsider pipeline. This pipeline standardizes memory retrieval in large language models (LLMs) through five checks and four processing modes, while introducing a self-review evaluation mechanism. Blind comparative experiments demonstrate significantly positive evaluations across most models, effectively validating the framework's feasibility. The core contribution of this work lies in constructing the first empirically grounded paradigm for AI memory governance, offering a systematic solution to enhance memory safety and standardization in human-computer interaction.

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.

Problem

• AI companions
• memory utilization
• contextual appropriateness
• human-AI interaction

Innovation

• AI Companions
• Memory Utilization
• Reconsider Framework
• LLM Evaluation
• Human-AI Interaction