Long AI conversations can become difficult to navigate. Starting a new chat gives you a cleaner workspace, but the new conversation may not automatically inherit every decision, preference, draft, constraint, and discarded idea from the original thread.
The most reliable solution is to use the platform’s built-in continuity features when available, supported by a strong context-transfer prompt.
How Claude and ChatGPT differ
Claude can search and reference previous conversations when its past-chat search feature is available and enabled. This means you can ask Claude to locate a relevant conversation and continue from it.
ChatGPT can use information from previous conversations through Memory, Reference chat history, Projects, and conversation branching. However, it may not recall every detail from an unrelated previous chat.
For important creative, technical, or business work, neither platform’s automatic memory should be treated as a perfect archive. A structured handoff prompt remains the safest option.
The simple Claude prompt
When Claude can search previous chats, start a new conversation with:
Continue from my conversation about [topic]. Pull the relevant context from that chat and continue from where we left off.
For more control, ask Claude to explain what it recovered:
Continue from my previous conversation about [topic or project name].
Search and reference any relevant past chats you can access. Recover:
- What we are working on and why
- The current status
- Key decisions and the reasoning behind them
- Approaches we tried, rejected, or ruled out
- My preferences and constraints
- Important terminology, drafts, files, and reference material
- Unresolved questions
- The most logical next step
Before continuing, briefly summarize what you recovered and clearly identify anything you could not verify.
Do not invent missing details.
The manual Claude handoff prompt
Before leaving the original Claude conversation, ask it to prepare a briefing for the new chat:
Summarize this conversation as a briefing for a fresh instance of you.
Include:
- What we are working on and why
- Key decisions we made and the reasoning behind them
- Things we tried that did not work
- Approaches we rejected or ruled out
- My preferences and constraints
- Important terminology, facts, links, files, and reference material
- The current status
- Unresolved questions
- What should happen next
Be thorough. Assume the reader knows nothing about this conversation.
Clearly distinguish between established facts, decisions, assumptions, and unresolved questions.
Do not invent missing information.
Paste the resulting summary into the new conversation with:
Here is context from a previous conversation.
Read it carefully, treat it as the working context for this chat, and continue from where it leaves off.
[PASTE THE HANDOFF BRIEF HERE]
The ChatGPT branching prompt
When using ChatGPT’s Branch in new chat feature, the earlier context is carried into the new branch. Use this prompt to establish where the work stands:
Continue from here in a fresh direction.
Before proceeding, briefly identify:
- The current objective
- Decisions already made
- The reasoning behind those decisions
- Important constraints
- Approaches already rejected
- Unresolved questions
- The most logical next step
Preserve the established terminology, preferences, and requirements from the conversation so far.
Do not reopen settled decisions unless new information makes that necessary.
The ChatGPT Project prompt
For ongoing work, keep related conversations, instructions, and files inside the same ChatGPT Project.
Start a new Project conversation with:
Continue the work from the previous conversations in this Project about [topic or project name].
First reconstruct the relevant context, including:
- What we are trying to accomplish
- Why the work matters
- The current status
- Decisions already made and the reasoning behind them
- Approaches already tested, rejected, or ruled out
- My preferences, constraints, and recurring instructions
- Important terminology, files, drafts, and reference material
- Open questions, risks, and dependencies
- The next recommended action
Use the conversations, files, and instructions in this Project as your primary sources.
Do not invent missing details or silently override earlier decisions. Flag uncertainty or contradictions before continuing.
The complete ChatGPT handoff prompt
For the safest transfer between separate conversations, ask the original ChatGPT thread to create a structured handoff brief:
Create a comprehensive handoff brief for continuing this conversation in a new chat.
Capture everything the new chat needs to work as though it had followed the full conversation, including:
- My goals, background, preferences, working style, and relevant personal or business context
- The project’s purpose, current state, terminology, characters, systems, and important reference information
- Decisions already made and the reasoning behind them
- Constraints, requirements, boundaries, and things explicitly rejected
- Established tone, style, formatting, and quality expectations
- Completed work, current drafts, unresolved questions, and next steps
- Any recurring instructions or corrections I have given you
- Important names, dates, numbers, links, examples, and source material
- Assumptions that have been made, clearly labeled as assumptions
- Contradictions or uncertainties that the next chat should resolve
Organize the handoff so it is easy for another AI conversation to use. Prioritize accuracy and operational usefulness over brevity.
Do not invent missing information or collapse meaningful distinctions.
End with:
1. A concise “Current Situation” summary
2. A prioritized “Next Actions” list
3. A “How to Work With Me” section
4. A ready-to-paste opening prompt for the new chat
Paste the handoff into the new ChatGPT conversation with:
The following is a handoff brief from a previous conversation. Treat it as the working context for this chat.
Read it carefully before responding. Preserve its established decisions, terminology, constraints, tone, and unresolved work.
Do not silently override it. If something in the brief is ambiguous or contradictory, flag it rather than guessing.
[PASTE HANDOFF BRIEF HERE]
First, briefly confirm your understanding of:
- The main objective
- The current state of the work
- The most important constraints
- The next action you recommend
Then continue from where the previous conversation stopped.
One prompt that works with both Claude and ChatGPT
The following prompt combines automatic conversation retrieval with a manual handoff fallback:
Continue my previous work about [topic or project name].
First, search, reference, or reconstruct any relevant previous conversations you can access.
Recover:
- The main objective and why it matters
- The current state of the work
- Key decisions and the reasoning behind them
- Approaches tried, rejected, or ruled out
- My preferences, constraints, and recurring instructions
- Important terminology, files, drafts, examples, and reference material
- Unresolved questions, risks, and dependencies
- The most logical next step
Do not invent missing details. Clearly identify anything you could not recover or verify.
If you cannot access enough of the previous conversation, use the handoff brief below as the source of truth:
[PASTE THE HANDOFF BRIEF HERE]
Before continuing:
1. Summarize your understanding of the objective and current status.
2. List the most important decisions and constraints.
3. Identify any missing or contradictory information.
4. Recommend the next action.
Then continue from where the previous conversation stopped.
Preserve the latest working material
Always include the latest version of important code, drafts, outlines, documents, or other working material.
A handoff brief preserves the history of the work. The latest artifact preserves the work itself.