Extract Intelligence From Your AI Conversations

Follow the extraction process

1. 1

   Use the exact AI conversation when you can

   Use the original conversation when you can. If the chat happened in a different app, paste the transcript into a new chat. Built-in memory and search features can help you find older conversations, but they summarize loosely and miss details. Pick one conversation before running the extraction.

   Tip: If you need an accurate record, use the exact chat or a pasted transcript. Memory can miss details, merge conversations, or summarize the past too loosely.

   

2. 2

   Extract only what the AI conversation actually supports

   Copy the prompt below and paste it into the AI conversation or transcript you want to preserve. The prompt tells the AI to look for the specific kinds of intelligence worth saving: your preferences, decisions, standards, workflows, priorities, principles, open questions, and reasoning. It also tells the AI to skip any category that has no clear evidence in the conversation. Every extracted record should point back to a specific moment in the conversation, not just sound plausible.

   PromptExtract Intelligence From This AI ConversationOpen when you are ready to copy it.

   One prompt that extracts the preferences, decisions, standards, workflows, principles, open questions, and reusable thinking from your conversation.

   Review this AI conversation or pasted transcript and extract the intelligence worth saving: the preferences, values, standards, priorities, workflows, patterns, principles, decisions, reasoning, and open questions embedded in the exchange. Only include an item if you can connect it to something visible in this conversation or transcript. Do not fill a category just because it is listed. If a category has nothing clearly supported, write "No clear item found." For each item, include: - Record type: project, next step, decision, preference, standard, principle, workflow, problem/fix, open question, thinking pattern, or useful reference. - Record: the item worth saving, written clearly enough that it would still make sense later. - Source: the message, section, quote, or moment in the conversation that supports the item. - Confidence: high, medium, or low. - Related items: any project, person, decision, principle, or open question this item connects to. Extract these categories: 1. Core goal: What was this conversation trying to accomplish? Summarize it in one sentence and cite the evidence. 2. Projects: Active efforts with a specific goal or outcome. Note the stage and what each project needs next. 3. Ongoing responsibilities: Areas of work or life that came up repeatedly without a single finish line. 4. Next steps: Specific actions. Include the owner and deadline if the conversation mentioned them. 5. Decisions: For each decision, state what was chosen, what alternatives were considered, what tipped the choice, and the reasoning. 6. Preferences, values, and standards: How you prefer to work, the quality bars that came up, the rules or boundaries that were set, and the priorities that were expressed. "Review the source transcript before writing a summary" is a standard. "I prefer concise updates over detailed reports" is a preference. Write each as a standalone sentence. 7. Principles and lessons: Guiding insights, rules of thumb, or hard-won lessons that could apply beyond this conversation. Write each as a standalone sentence. 8. Workflows and processes: Any repeating sequence of steps, routine, or method that surfaced. Describe the trigger, the steps, and the expected result. 9. Problems and fixes: What went wrong, what resolved it, and whether the fix applies to similar situations. 10. Open questions: Unanswered questions, deferred decisions, untested assumptions, and follow-ups still needing action. Flag which ones need attention soon. 11. Reusable thinking patterns: Decision methods, structured approaches, or reasoning moves that could be reused in another context. 12. Useful references: Tools, sources, links, examples, names, files, or knowledge worth saving. At the end, list anything that seemed important but was too uncertain to save without human review. For items that connect to each other, note the relationship.

   

3. 3

   Review the extraction before anything gets saved

   Treat the extraction as a draft. A useful extraction captures the specific preferences, decisions, standards, workflows, principles, and open questions from your conversation, with each record pointing back to the moment that supports it. Sort each item into keep, revise, or delete. Keep records that are specific and grounded in the conversation. Delete anything that sounds invented or cannot be traced to a real exchange.

   Tip: If the extraction only says the conversation was about productivity, strategy, or memory, it is too shallow. Run the prompt again on one specific conversation or paste a fuller transcript.

   Example outputExample extraction output from ChatGPT or ClaudeOpen when you want a reference shape.

   Your extraction will look different. What matters: each record is specific, traces back to the conversation, and is ready for your review.

   Core goal
   This conversation focused on designing a reusable system for pre-meeting preparation. Evidence: three separate exchanges returned to the question of what to review before a call. Confidence: high.
   
   Projects
   - Client prep system: Build a repeatable pre-meeting workflow combining discovery notes, meeting plans, and follow-up drafts. Stage: early design. Source: session planning thread. Confidence: high.
   
   Next steps
   - Draft a template for pre-meeting review notes. Owner: me. No deadline mentioned. Source: final exchange.
   
   Decisions
   - Save source transcripts before summarizing. Reasoning: future summaries need the original evidence, not a memory of the conversation. Alternatives considered: summarize immediately. Confidence: high.
   
   Preferences, values, and standards
   - Always review the source transcript before writing any summary. Source: decision thread about workflow order. Confidence: high.
   - Prefer concise two-sentence status updates over long written reports. Source: discussion about team communication. Confidence: medium.
   
   Principles and lessons
   - Store the reason alongside the action. A task without its reasoning becomes hard to reuse or delegate later. Confidence: medium.
   
   Workflows and processes
   - Pre-meeting prep: (1) Pull the client's discovery notes, (2) draft three questions from open items, (3) review last meeting's action items. Trigger: any scheduled client call. Source: session planning thread. Confidence: high.
   
   Open questions
   - Which folder should own client examples: the teaching library or client records? Needs a human decision before filing. Urgency: moderate.

   

4. 4

   Open only the knowledge folder the agent should change

   Switch to a folder-backed AI agent: Codex, Claude Code, or Claude Cowork. These agents can read and edit files on your computer, which is what lets them save your extracted preferences, decisions, and workflows into actual files you own. Open the agent inside your personal assistant or second-brain folder. Choose only the folder you want updated.

   Tip: Open only the folder this task needs. Do not give the agent access to your whole home directory, Downloads folder, financial files, health records, or passwords. Give it the smallest folder that contains the files it should update.

   

5. 5

   Give the folder-backed agent the reviewed extraction and filing instructions

   The agent needs two things: your reviewed extraction from step 3 and the filing instructions below. Paste the extraction first, then paste this prompt underneath it. The prompt tells the agent to reuse your existing files, connect related records, and place uncertain items somewhere visible instead of inventing new folders.

   PromptFile These Reviewed AI Conversation RecordsOpen when you are ready to copy it.

   Run this in Codex, Claude Code, or Claude Cowork with your personal assistant or second-brain folder open.

   The reviewed extraction output is pasted above this prompt. You are working inside my personal assistant or second-brain folder. File these records into the right existing files or folders. Create a new file only when no sensible home already exists. Rules: - Work only inside the folder I opened for this session. - Preserve the source date, conversation name, and confidence rating when provided. - Keep the human-reviewed wording unless a small edit makes the record clearer. - Connect each record (decisions, preferences, standards, workflows, principles) to the project, person, or open question it affects. - Add open questions somewhere I will actually see them later. - If a destination is uncertain, place the item in a visible "Needs review" section or file instead of guessing. - Do not create a new folder system unless I explicitly ask for one. Before editing, briefly list the files or folders you plan to update. Then make the changes. After editing, summarize: 1. Which files changed. 2. Which records were added. 3. Which items still need human review.

   

6. 6

   Check whether the saved records are well organized

   Inspect the changed files yourself. Check that decisions still include their reasoning, preferences and standards are clearly stated, workflows have their steps and triggers, and open questions landed somewhere you check regularly. Well-organized files are easier to maintain, review, and build on. If the agent invented folder names or created unnecessary structure, ask it to consolidate.

   Tip: Prefer fewer well-named files over a freshly generated folder maze.