ChatGPT Redesign
A full redesign concept focused on clarity, discoverability, and productive AI workflows
Concept video
After more than two years of intensive use across multiple projects, we identified strong redesign potential in ChatGPT's interface. Several functional areas could be improved in structure and usability to increase writing and research efficiency.
The target was a clearer product architecture, stronger discoverability, and a more supportive interaction model for long-running work sessions.
Overview
Why a redesign and what needed to improve
UI mockup frame and overall screen composition
The redesign started from three recurring issues: imprecise information architecture, ambiguous icon logic, and a visually monotone interface with too little hierarchy. Together, these problems slowed orientation and hid important functions.
The project therefore framed one clear goal: build a structure that improves findability and efficiency while preserving a calm, modern visual system for everyday AI work.
Process
Predefined goals and concept framing
Wireframes: structure, flows, and early interaction framing
Before visual execution, six target principles were defined: room-based structure, source transparency, a dedicated canvas workspace, quick private mode access, stronger visual hierarchy, and measurable accessibility/efficiency criteria.
The concept was grounded in benchmark analysis across Gemini, Perplexity, Claude, DeepSeek, and Llama. From this, we set a four-phase process frame: research and analysis (2 weeks), concept and ideation (3 weeks), design and prototyping (4 weeks), and testing and iteration (3 weeks).
Key challenge definition focused on organization, collaboration, source traceability, and privacy handling. Each challenge was paired with a specific concept direction before implementation details were explored.
Outcome
Final feature system and design impact
Visual direction: banner exploration for the redesigned system
The final concept centers on four modules: Rooms for project-based organization, a dedicated Canvas workspace for longer flows, source-backed transparency in the conversation, and a global Quickchat (incognito) mode for short, low-stakes questions—plus a Library that keeps everything scannable.
Sources: in-context traceability and citation-backed passages
Transparency and direct traceability were a major priority. Models can always produce an answer—and may fabricate when needed—so we designed for explicit clarity: what is supported by sources is clearly marked in the text, and what is not remains visible as such.
Canvas workspace — primary layout
Canvas workspace — extended flows
Canvas: For people in demanding roles or in study contexts, Canvas is an all-round tool without constant tool switching: code, mind maps, and structured work happen directly on canvases inside the chat experience, with everything kept in one place. That leads to the next principle: keeping everything together.
Rooms — holistic project storage
Rooms — collaboration in shared project spaces
Rooms: Content can live in rooms in a holistic way—chats, canvases, sources, and more. A clear search and save pattern was important for a fluid workflow. Rooms also support collaborative work in shared project spaces.
Quickchat (incognito) for lightweight, ephemeral questions
Quickchat: Incognito-style chat existed before in spirit; we refined it. Not every exchange belongs in a shared project—some are quick word lookups or small info requests that do not concern everyone. Quickchat is invisible to collaborators, results are not persisted, and it stays reachable from anywhere for small questions in between.
Library: horizontally scrollable collections for clarity
Library: For overview, everything is organized in the library with horizontally scrollable lists of the different elements contained in a space. The color scheme is warm, friendly, and stimulating.
These modules are supported by a consistent visual system—typography hierarchy, component states, and restrained effects—so orientation stays clear without visual noise.