ChatGPT · AI product · Senior · 60 to 90 minutes

ChatGPT Memory Controls and Trust Model

ChatGPT is a conversational AI product from OpenAI that helps people write, learn, create, and get answers across text, image, and voice experiences. Answer this product management case and get it graded line by line against a weighted rubric.

Trust product design
User segmentation
Metric design
Prioritisation
Executive alignment

The situation

Your GM for ChatGPT and the Head of Safety have asked for a proposal on memory controls and the trust model for consumer ChatGPT. Memory can make replies more useful because ChatGPT may remember details from past interactions, and users can already view saved memories, delete them, or turn memory off. But support tickets, social posts, and recent research show a trust gap: many people appreciate personalization once they notice it, while others feel watched, do not understand why something was remembered, or are surprised when an old detail appears in a new chat. Assume the following numbers are illustrative: after memory reached 22% monthly active use among eligible users, complaints about unexpected or creepy responses rose 18%, while retention for active memory users was 9% higher than for similar users without memory. Leadership does not want to slow a capability that clearly improves usefulness, but they also do not want ChatGPT to gain a reputation for remembering too much or in ways people do not expect. You need to design the controls, product rules, and rollout plan that increase trust without making memory so hard to use that it loses value.

What to produce

A 1 to 2 page product proposal.

Your answer must

Identify at least three user segments or use cases with different trust needs, including one high sensitivity segment.
Define a memory trust model that covers defaults, consent moments, explainability, review and delete flows, and clear rules for when ChatGPT should not save or use memory.
Propose a first release and a second phase over two quarters, with rationale for sequencing and major tradeoffs.
Specify one north star metric and at least three guardrail metrics, including a way to measure surprise or creepy failures.
Describe the key cross functional alignment needed across Product, Design, Safety, Policy, Legal, Support, and Comms, and name one decision you would explicitly escalate to leadership.

Constraints

You have 12 weeks for a first release, with one PM, one designer, and up to eight engineers shared across web and mobile surfaces.
You must build on existing ChatGPT settings, saved memory management, and Temporary Chat behaviors in this phase, not a brand new privacy center.
The solution must work across web, iOS, and Android, and cannot depend on human review of individual memories or responses.
The proposal must respect current privacy and safety expectations that users can inspect, delete, and disable saved memory.

Format

Use sections for user problem, trust model, proposed controls, rollout, metrics, and risks. A small table for segments and memory rules would help.