Most AI-assisted writing hides the human. The final post looks clean, confident, finished — but the prompts that produced it were hesitant, self-correcting, specific. Those are your voice. Trace is a format for publishing both: the final text, and the exact words you typed to get there.
Paste this at the end of any session in any AI tool — ChatGPT, Claude, Gemini, Copilot, Perplexity, anything. The AI already has your full conversation in context. It will reconstruct the Trace from what it can see. Then paste the output into the Trace creator to get a shareable page.
Format this conversation as a Trace document using the schema below. A Trace is a structured record of AI-assisted writing. It preserves the human's prompts verbatim so a reader can see how the final output was actually made. A Trace has two records: the session record (your messages in chronological order) and the unit record (the final output annotated paragraph by paragraph). Rules: - Do not paraphrase, clean up, or improve my prompts. Reproduce them word for word, including typos, hedging, and informal language. - Do not editorialize or add commentary about my process. - In the session record, list every message I sent, in order. Tag each as prompt (new instruction), reaction (response to your output), or edit (I changed the text directly). - In the unit record, map each paragraph of the final output to the prompt and reaction that most directly shaped it. - Any text in the final output that I wrote directly — not from an AI response — should be marked TYPE: human. - The final output appears first, followed by both records. Schema (use this exactly): --- TRACE [title or topic] [date] [tool used — optional] --- FINAL OUTPUT: [the finished text, exactly as it would be published] --- SESSION RECORD: [1] MESSAGE: [my exact words, verbatim] ROLE: prompt | reaction | edit [2] MESSAGE: [my exact words, verbatim] ROLE: prompt | reaction | edit ... [HUMAN ADDITIONS] TEXT: [anything I wrote directly into the final output, not via prompting] --- UNIT RECORD: [UNIT 1] TEXT: [sentence or paragraph from final output] PROMPT: [verbatim human input — reproduce exactly, do not clean up] REACTION: [human's response to AI output, before next prompt — omit if none] TYPE: prompted | human | hybrid ROUNDS: [session message numbers that shaped this unit] [UNIT 2] ...repeat for each meaningful unit of the final output... --- END TRACE
The human's messages, in the order they were sent.
Click any paragraph to see the exact words that produced it.
Trace is a plain-text open format. You can produce it by hand, with the closing prompt above, or by building it into any tool. The schema is intentionally simple.
--- TRACE [title or topic] [date] [tool used — optional] --- FINAL OUTPUT: [the finished text] --- SESSION RECORD: [n] MESSAGE: [verbatim human input] ROLE: prompt | reaction | edit [HUMAN ADDITIONS] TEXT: [anything written directly, not via prompting] --- UNIT RECORD: [UNIT n] TEXT: [sentence or paragraph from final output] PROMPT: [verbatim human input — reproduce exactly, do not clean up] REACTION: [human's response to AI output, before next prompt — omit if none] TYPE: prompted | human | hybrid ROUNDS: [session message numbers that shaped this unit] --- END TRACE
Once the AI produces your Trace output, paste it into the creator to get a formatted page with a shareable URL. No account needed, no data stored — the entire document lives in the link.
Five examples of Trace documents in both session and unit views. Click into one to read the finished text, then explore the prompts behind it.