AI Conversation Extraction Prompt Kit
Purpose
This kit upgrades your original prompt:
Provide me with a downloadable MD artifact for all of the key parts of this discussion that I would be able to provide as a guide/playbook to another AI.
The revised version is built to get more complete extraction, better verification, clearer uncertainty handling, and a stronger final Markdown artifact that another AI can actually use.
What was missing from the original prompt
Your original prompt is good as a starting trigger, but it leaves too much up to the model:
- It does not explicitly require verification of claims.
- It does not force capture of decisions, assumptions, open questions, risks, or contradictions.
- It does not tell the model how to distinguish confirmed facts from tentative ideas.
- It does not specify a useful output structure for downstream AI use.
- It does not require preservation of source references, links, or citations.
- It does not ask for a reviewer/self-critique loop before finalization.
- It does not ask for missing-but-important context that was implied rather than directly stated.
- It does not push the model to convert the conversation into an operational artifact rather than a loose summary.
Best overall prompt (recommended)
Copy and paste this as your default extraction prompt.
Create a **downloadable Markdown artifact** that extracts and organizes the most valuable information from this discussion so it can be used as a **high-quality guide, playbook, briefing memo, and handoff document for another AI or human operator**.
Your job is **not** to give me a light summary. Your job is to produce a **comprehensive, structured, verified, and operationally useful document**.
## Core objective
Turn this discussion into a Markdown document that:
1. captures the most important ideas, decisions, recommendations, arguments, tradeoffs, workflows, examples, and next steps,
2. preserves important context that another AI would need in order to continue the work well,
3. clearly separates **verified facts** from **tentative suggestions, assumptions, speculation, or open questions**,
4. includes **source references / citations / links / evidence** wherever possible,
5. is optimized to be useful as a **handoff artifact** for future AI agents.
## Required process
Follow this process before producing the final document:
### Phase 1: Extract
Identify and capture all high-value material from the discussion, including:
- goals and intent,
- key facts,
- constraints,
- preferences,
- decisions made,
- rejected options,
- tradeoffs discussed,
- workflows/processes,
- tactical recommendations,
- strategic insights,
- examples, templates, scripts, or frameworks,
- risks, warnings, and red flags,
- unresolved questions,
- implicit assumptions,
- anything that would materially help another AI continue the work.
### Phase 2: Verify
Verify every material claim as much as possible.
- Do **not** present unverified statements as facts.
- If something was stated in the discussion but not independently verified, label it clearly as one of:
- **User-stated**
- **Assistant-stated but unverified**
- **Verified**
- **Tentative / speculative**
- Where verification is possible, include the supporting citation, source reference, link, or note describing what was verified.
- Where verification is not possible, explicitly state the uncertainty.
- If claims conflict, call out the conflict rather than smoothing it over.
### Phase 3: Strengthen
Improve the usefulness of the document by adding:
- missing but important considerations that logically should be included,
- edge cases,
- implementation notes,
- “what to do next” guidance,
- pitfalls and failure modes,
- operational checklists,
- handoff notes for future AI agents,
- questions that should be answered before execution.
### Phase 4: Review
If multi-agent or reviewer-agent capability is available, launch a reviewer agent to critique the draft for:
- missing information,
- unsupported claims,
- weak structure,
- ambiguity,
- poor handoff quality,
- lack of actionability.
Then improve the document using that feedback.
If reviewer-agent capability is **not** available, perform a strong self-review pass and explicitly state that a self-review was used instead.
## Required output structure
Produce the Markdown artifact with these sections, unless a section is clearly irrelevant:
# Title
A precise title that reflects the discussion.
# Executive Summary
A concise but high-value summary of the discussion and its practical importance.
# Purpose of This Document
What this document is for and how another AI or operator should use it.
# Discussion Context
Relevant background, goals, stakeholders, constraints, and assumptions.
# Key Facts and Verified Findings
A structured list of facts, each labeled as:
- Verified
- User-stated
- Assistant-stated but unverified
- Tentative / speculative
Include citations, links, or evidence notes wherever possible.
# Major Decisions and Conclusions
What was decided, recommended, or concluded.
# Reasoning, Tradeoffs, and Why It Matters
Capture the logic behind the recommendations, not just the conclusions.
# Recommended Playbook / Process
Turn the discussion into an actionable process or operating playbook.
# Tools, Resources, Links, and References
Include relevant links, documentation references, products, services, and cited sources.
# Risks, Caveats, and Red Flags
What could go wrong, where the uncertainty is, and what needs caution.
# Open Questions / Unknowns
What still needs to be answered, verified, decided, or tested.
# Suggested Next Steps
Ordered next steps with practical priority.
# Handoff Notes for Another AI
Include:
- what context the next AI should know,
- what assumptions it should avoid making,
- what it should verify first,
- what outputs would be most useful next.
# Optional Appendices
Where relevant, include:
- glossary,
- templates,
- scripts,
- checklists,
- sample prompts,
- comparison tables,
- implementation notes,
- reviewer comments and what changed.
## Output quality rules
- Be comprehensive, but keep it organized and skimmable.
- Prefer structured bullets, tables, and headings over dense prose.
- Preserve nuance.
- Do not invent facts, links, or citations.
- Do not silently upgrade guesses into facts.
- Do not omit important disagreements or uncertainties.
- Include the most useful material for execution, not just what sounds polished.
- Write the final result so it is genuinely useful to another AI system with minimal extra context.
## Final instruction
Return the result as a **downloadable Markdown artifact**.
If possible, also save the Markdown in a way that preserves it in local/history workflows.Stronger version for research-heavy conversations
Use this when the discussion includes lots of factual claims, product comparisons, current events, laws, compliance, or technical recommendations.
Create a downloadable Markdown handoff artifact from this discussion.
This must be a **research-grade extraction**, not a casual summary.
In addition to extracting the key parts of the discussion, do all of the following:
- verify important claims against reliable sources where possible,
- preserve source references and links,
- flag anything outdated, uncertain, conflicting, or weakly supported,
- separate confirmed facts from hypotheses, opinions, and planning ideas,
- identify gaps where the discussion may have missed an important issue,
- include a reviewer pass (or self-review if reviewer agent is unavailable),
- optimize the final Markdown so another AI can continue the work without needing the original conversation.
Required output sections:
1. Executive Summary
2. Verified Findings
3. Unverified or Tentative Claims
4. Recommendations
5. Tradeoffs and Alternatives Considered
6. Risks / Caveats / Compliance Concerns
7. Open Questions
8. Action Plan
9. Source References
10. Handoff Notes for Another AI
11. Reviewer Notes and Revisions Made
Rules:
- No fabricated citations.
- No unsupported certainty.
- Label each important point by evidence status.
- Include exact references where available.
- If verification cannot be completed, say so plainly.
- Improve the structure and usefulness of the document beyond what was explicitly discussed.
Return the result as a downloadable Markdown artifact.Shorter everyday version
Use this when you want something strong but not massive.
Please turn this discussion into a downloadable Markdown handoff document for another AI.
Make it comprehensive and practical, not just a summary. Extract the key facts, decisions, recommendations, workflows, tradeoffs, risks, open questions, and next steps. Clearly label what is verified versus unverified or speculative. Preserve links, references, and citations where possible. Add any important missing considerations that should logically be included.
If a reviewer agent is available, use it to critique and improve the draft. Otherwise do a self-review pass and note that you did so.
Structure the document so another AI can continue the work with minimal extra context.“Strict handoff” version for feeding another AI directly
Use this when the output is mainly meant to become context for another model.
Build a downloadable Markdown handoff artifact from this discussion for direct use by another AI.
Prioritize:
- completeness,
- truthfulness,
- source traceability,
- operational usefulness,
- explicit uncertainty labeling.
The document must help a new AI quickly understand:
- what we were trying to do,
- what was learned,
- what was decided,
- what remains unknown,
- what should happen next.
Include these sections:
- Objective
- Context
- Confirmed Facts
- Claims That Still Need Verification
- Key Decisions
- Recommended Plan
- Constraints and Preferences
- Risks and Failure Modes
- Open Questions
- Next Best Actions
- Handoff Instructions for the Next AI
Important:
- Do not invent facts.
- Do not compress away nuance.
- Do not bury important caveats.
- Preserve exact terminology where it matters.
- Use reviewer-agent feedback if available; otherwise self-critique and revise before finalizing.
Return the final answer as a downloadable Markdown artifact.Extra clauses worth adding depending on your use case
You can append any of these to the main prompt when relevant.
1) Preserve exact wording where strategically important
Where exact phrasing matters (for example, scripts, messaging, positioning, or legal/compliance wording), preserve the original wording or provide a clearly marked near-verbatim version.2) Capture implied strategy, not just explicit statements
Also extract important implied insights, priorities, and assumptions that were not always stated directly but are strongly supported by the discussion.3) Make it useful for execution
Whenever possible, convert ideas into checklists, SOP-style steps, templates, or decision frameworks.4) Make it useful for future prompting
Include a section titled “Reusable Prompting Notes” that explains how another AI should approach this topic, what mistakes to avoid, and what additional prompts would be most useful next.5) Prevent fake confidence
If a point cannot be verified, do not soften that uncertainty. Be explicit about the limitation.6) Preserve references for auditing
Include a source trail or evidence note for each major claim or recommendation whenever possible.7) Add a machine-friendly appendix
Add a final appendix with a structured YAML or JSON-style summary of the key facts, decisions, risks, open questions, and next steps so another AI can parse it more easily.8) Include “what changed during review”
After the review pass, include a short section documenting what the reviewer flagged and what improvements were made.My recommended final prompt for you
This is the one I would actually use most often in your shoes:
Provide me with a **downloadable Markdown artifact** that extracts, organizes, verifies, and improves all of the key parts of this discussion so it can serve as a **guide, playbook, briefing memo, and handoff document for another AI**.
Do not give me a light summary. Build a **comprehensive, structured, operationally useful document**.
Requirements:
- Extract the most important facts, decisions, recommendations, tradeoffs, workflows, examples, risks, caveats, open questions, and next steps.
- Preserve the context, goals, constraints, preferences, and assumptions that another AI would need.
- Clearly label each important point as **Verified**, **User-stated**, **Assistant-stated but unverified**, or **Tentative / speculative**.
- Verify material claims wherever possible and include citations, links, references, or evidence notes.
- If something cannot be verified, say so explicitly.
- Do not invent facts or citations.
- Do not hide disagreements, uncertainty, or conflicting claims.
- Add important missing considerations, edge cases, pitfalls, and implementation notes that should logically be included.
- Convert the discussion into an actionable playbook or handoff artifact, not just a narrative recap.
- Include a section specifically called **Handoff Notes for Another AI**.
- Include a section called **Open Questions / What Still Needs Verification**.
- Include a section called **Reviewer Notes and Improvements Made**.
Process:
1. Extract the high-value material.
2. Verify and label evidence status.
3. Strengthen the document with missing but important considerations.
4. If reviewer-agent capability is available, use a reviewer agent to critique the draft and then improve it.
5. If reviewer-agent capability is not available, perform a serious self-review pass and state that you did so.
Preferred output structure:
- Title
- Executive Summary
- Purpose of This Document
- Discussion Context
- Key Facts and Verified Findings
- Major Decisions and Conclusions
- Reasoning, Tradeoffs, and Why It Matters
- Recommended Playbook / Process
- Tools, Resources, Links, and References
- Risks, Caveats, and Red Flags
- Open Questions / What Still Needs Verification
- Suggested Next Steps
- Handoff Notes for Another AI
- Reviewer Notes and Improvements Made
- Optional Appendix with structured YAML/JSON-style summary
Return the result as a **downloadable Markdown artifact**.Why this version is materially better
It forces the downstream AI to do five things your current prompt does not reliably force:
-
Separate truth from conversation residue
A lot of AI-generated summaries accidentally blur together facts, guesses, and brainstorming. This prompt blocks that. -
Preserve handoff quality
Another AI needs more than a summary. It needs context, decisions, assumptions, risks, and what to verify next. -
Produce an operational document
This pushes the result toward a playbook, not just notes. -
Include review pressure
The reviewer/self-review step catches missing sections and fake certainty. -
Make the output auditable
The verification labels and evidence notes make it more trustworthy.
Practical note on “reviewer agent” language
Not every AI environment actually has true multi-agent support.
So the best wording is:
- “If reviewer-agent capability is available, use it”
- otherwise “perform a serious self-review pass and state that you did so”
That keeps the prompt strong without depending on a capability that may not exist.
Best practice workflow
For your use case, this is the cleanest workflow:
- Have the AI do the main conversation/research/work.
- At the end, run the recommended final prompt.
- Save the Markdown artifact locally.
- Feed that artifact into another AI as context.
- Optionally ask a second AI:
“Critique this handoff document for missing context, false certainty, and weak next steps.” - Save the improved version as the durable record.
Optional follow-up prompts you can chain after the artifact is generated
Critique prompt
Critique this Markdown handoff document for missing context, unsupported claims, weak structure, bad assumptions, poor prioritization, and anything that would make it less useful to another AI or operator. Be tough and specific.Compression prompt
Now create a second version of this document that is shorter and optimized for fast ingestion by another AI, while preserving all critical facts, decisions, risks, and next steps.Structured parsing prompt
Now extract the key facts, decisions, risks, open questions, and next actions from this Markdown artifact into a clean YAML block for machine-friendly reuse.Final recommendation
Use the My recommended final prompt for you section as your default.
That version is strong enough to:
- get more out of the conversation,
- force verification discipline,
- create better local historical records,
- and make the artifact much more reusable by future AI agents.