Summary
I propose introducing a visual, tree-structured conversation interface in ChatGPT, allowing users to highlight specific passages within a response and open supplementary conversations (branches) without interrupting the primary conversation (trunk).
These branches would remain attached to their exact originating passages and could be accessed through an interactive visual map showing the main conversation and all its branches in their original chronological and contextual positions.
This would be particularly valuable for education, research, technical discussions, and other lengthy conversations where users frequently need clarification without losing their place.
1. The Problem
Currently, ChatGPT conversations follow a predominantly linear structure.
Suppose a student is working through a university-level economics examination with ChatGPT. The examination contains eight questions, and the student wants to work through them sequentially.
While reading ChatGPT’s answer to Question 2, the student encounters the following explanation:
“Under rational expectations, anticipated monetary policy changes may have no effect on real output in certain economic models.”
The student understands the overall answer but wants to investigate what “rational expectations” means.
They ask ChatGPT for an explanation, which becomes the next message in the main conversation. ChatGPT provides a detailed response, prompting two further questions.
After resolving their confusion, the student must scroll back through several messages to find where they originally left off in Question 2.
The same problem occurs repeatedly as the student progresses through the examination.
Consequently, what should be one clearly structured discussion covering eight examination questions becomes a lengthy, fragmented conversation containing dozens of supplementary explanations interspersed between the principal answers.
This makes it difficult to follow the original discussion, review the material later, or distinguish essential content from tangential explanations.
2. Proposed Solution: Inline Conversation Branching
Users should be able to highlight any text within a ChatGPT response and select an option such as “Ask in new branch”.
This would open a supplementary conversation in a side panel, allowing the user to ask questions specifically about the selected passage.
For example:
Main conversation:
ChatGPT explains how rational expectations influence the effectiveness of monetary policy.
Highlighted passage:
“Under rational expectations, anticipated monetary policy changes may have no effect on real output…”
Supplementary branch:
-
User: “What exactly are rational expectations?”
-
ChatGPT: Provides an explanation.
-
User: “How are they different from adaptive expectations?”
-
ChatGPT: Explains the distinction.
-
User: “Does this mean monetary policy is always ineffective?”
-
ChatGPT: Explains the relevant theoretical qualifications.
Once satisfied, the student closes or minimises the branch and immediately returns to the exact passage in the main conversation.
Crucially, the supplementary discussion remains accessible from that passage indefinitely.
3. Visual Conversation Tree
The second, equally important feature would be a visual representation of the entire conversation structure.
Instead of displaying separate branches as an unstructured collection of similarly named conversations, ChatGPT would display a central trunk representing the primary discussion, with supplementary conversations branching off at the exact points where they originated.
For example:
MAIN CONVERSATION
│
├── Question 1: Supply and Demand
│ │
│ ├── Branch: Price Elasticity
│ │
│ └── Branch: Consumer Surplus
│
├── Question 2: Monetary Policy
│ │
│ ├── Branch: Rational Expectations
│ │ │
│ │ └── Sub-branch: Adaptive Expectations
│ │
│ └── Branch: Policy Neutrality
│
├── Question 3: Economic Growth
│ │
│ ├── Branch: Solow Model
│ │
│ └── Branch: Technological Progress
│
└── Question 4: Inflation
│
└── Branch: Inflation Expectations
This diagram is a simplified representation. Ideally, the actual interface would provide an interactive graphical tree with expandable and collapsible branches.
The position of each branch should correspond to its original position in the main conversation, not merely the time at which the branch was created.
For instance, if a user later revisits Question 1 and opens a new branch concerning a particular sentence, that branch should appear beside the relevant passage in Question 1, rather than at the end of the overall conversation.
This distinction is essential to the usefulness of the feature.
4. Suggested Functionality
The proposed interface would ideally include:
-
Text-anchored branching: Highlight any word, sentence, paragraph, equation, or code snippet to open a supplementary conversation.
-
Side-by-side viewing: Display the branch alongside the originating response, so the user can refer to the original explanation while asking questions.
-
Persistent contextual links: Preserve the relationship between each branch and its originating passage, allowing branches to be reopened directly from the main conversation.
-
Interactive visual tree: Provide an overview of the entire discussion, showing branches and sub-branches in their correct locations along the main trunk.
-
Collapsible branches: Allow users to hide or expand supplementary discussions, preventing the interface from becoming visually overwhelming.
-
Return-to-position functionality: Closing a branch should return the user to exactly where they were reading.
-
Independent branch histories: Each branch should maintain its own conversation history without automatically inserting its messages into the main discussion.
-
Optional integration of insights: Users could choose to incorporate a summary or conclusion from a branch into the main conversation when that information is important to subsequent discussion.
5. Benefits and Applications
Although the example above concerns economics education, the feature would have applications across many areas.
Education: Students could investigate unfamiliar concepts without disrupting a structured lesson or examination walkthrough.
Academic research: Researchers could explore methodological questions, definitions, or alternative interpretations while keeping the primary research discussion coherent.
Programming: Developers could ask for explanations of individual functions, errors, or implementation decisions without interrupting the main development process.
Document analysis: Users could ask supplementary questions about particular passages in a lengthy document while preserving the structure of their overall analysis.
Creative writing: Writers could explore individual plot points, character motivations, or alternative scenarios without disrupting the main narrative-development conversation.
6. How This Differs from Existing Conversation Branching
The central idea is not simply the ability to fork an existing conversation.
It is the ability to create small, contextual, persistent conversations attached to particular pieces of content within a larger conversation.
A conventional conversation fork creates an alternative continuation of the discussion.
The proposed system instead allows users to temporarily explore a particular concept and then return to the original discussion without changing its structure.
The visual tree would also serve as a navigational and organisational tool, rather than merely displaying a list of separate chats.
7. Conclusion
ChatGPT is increasingly used for lengthy, complex discussions that do not fit naturally into a strictly linear messaging interface.
A visual conversation tree with inline, text-anchored branching would allow users to distinguish between the primary discussion and the supplementary questions needed to understand it.
The result would be a more intuitive interface for learning, research, problem-solving, and revisiting previous conversations.
The underlying principle is simple: users should be able to explore a tangent without losing the original thread, and every tangent should remain visibly connected to the exact point from which it emerged.
(written with help form ChatGPT 6)