Key answer
Determine whether “remember” means information in the current conversation or saved memory available later. A verbal assurance does not specify storage, scope or retention.
Full explanation
Products have different mechanisms. OpenAI distinguishes ChatGPT memory from local Codex storage, which may update with background delays. Claude's current documentation offers viewable, editable and deletable memory topics and distinguishes project memory from material outside projects. Check the feature actually used by your account rather than inferring from another product under the same brand. OpenAI documentation, Claude documentation
Confirm three things: exact stored content, conversations where it applies, and how to update or delete it. Prefer the product's memory entries/settings; for file-based conventions, open the file. Retention follows documentation and account settings, not “I will always remember.”
Restate the complete convention and its scope. If stored in long-term memory or a project file, identify a verifiable location; if available only in this conversation, say so. Do not merge temporary exceptions into lasting preferences.
After confirming storage, test in a new conversation without revealing the answer, such as asking for the project's citation rules. Then request a small application example. Recall and application are different checks; one success does not guarantee future retrieval.
LongMemEval's 500 questions separately evaluate extraction, cross-session reasoning, temporal reasoning, updates and handling unanswerable questions. Memory involves multiple capabilities; test concrete behavior rather than accepting a claim of remembering. Section 3
References
1. OpenAI, Memories, official user documentation
2. Anthropic, Use Claude’s chat search and memory to build on previous context, official help