ChatGPT Projects — Product Teardown

The product question

This September 2026 independent teardown examined a long-running project workflow: users accumulate chats, files, changing decisions, rejected ideas, and unresolved questions. Returning to the work can require manually reconstructing what matters now.

The analysis distinguishes finding historical information from understanding the current project state. It is a product hypothesis and concept proposal, not a shipped ChatGPT feature or a claim about every user's experience.

Proposed experience: Project State

I proposed a lightweight view of the current objective, key decisions, open questions, active tasks, important constraints, recent changes, and supporting sources.

User controls would let people confirm final decisions, mark information as outdated, retain open questions, and exclude items from project memory. Source provenance, freshness, and decision status would help separate approved information from old brainstorming.

How I would evaluate it

The proposed experiment compares the existing workflow with a Project State view. The primary metric is time to meaningful continuation after reopening a project. Supporting measures include old chats reopened, searches, task continuation, and acceptance or correction of suggested state items.

Quality guardrails include stale information, incorrect memory, unsupported statements, and user correction rates. These are proposed metrics, not results from a completed experiment.

Risks & supporting analysis

Risks include treating brainstorming as a final decision, retaining stale context, losing nuance, reducing user control, and adding interface complexity. The central design principle is to preserve relevant, current, trustworthy context rather than simply retain more history.

Read the original Notion teardown

This internal overview keeps the analysis available even if the external Notion page requires access.