How to export and share your AI traces
An AI trace is the saved conversation behind the work: your prompts, the model replies, and the follow-up instructions. Submit the actual back-and-forth, not a summary of it.
Menus move. If a named export control is missing, copy the full thread. Do not use a public Share link.
Submit the conversation, not a recap.
Counts as a trace
- Include your prompts and the model replies. Output-only pastes look polished and score poorly.
- Add each conversation as a separate trace, up to eight.
- Stay above 500 characters and under 240,000 characters across all traces.
- Name the tool you actually used. The report does not infer it.
Does not count
- A summary of what you asked the model to do.
- A public Share link from ChatGPT, Claude, or Gemini.
- A screenshot of one answer.
- An account-wide data export that includes unrelated chats.
- Sessions that never touched this assignment.
ChatGPT, Claude, and Gemini
Most candidates work in a browser chat. Copy that thread. A Share button is for publishing, not for this interview.
ChatGPT
Copy the conversation you used for this assignment. Do not send a public Share link.
- Open the ChatGPT conversation that shaped the deliverable.
- Start at your first prompt and select through the last reply.
- Copy the text and paste it in Interview Lab as one trace. Label it with what that chat was for.
- Skip Share and skip Settings → Data controls → Export. Share can publish the chat. Account export is too broad and too slow for a take-home.
Paste as text, or save the copied chat as a .txt / .md file and import it.
Claude
~/.claude/projects
Export the Claude.ai chat, or import Claude Code session files if you used the terminal.
- Open the Claude.ai conversation used for this assignment.
- Select from your first message through the last reply, then copy.
- If the conversation lives in a Claude Project, export that project chat — not the project knowledge files.
- Do not send a Claude Share link. Paste the transcript into Interview Lab instead.
Paste as text. Claude Code users can import the .jsonl files from the local session directory.
Gemini
Copy the Gemini chat used for the assignment. Leave unrelated Gemini Apps Activity out.
- Open gemini.google.com and the chat that shaped the work.
- Select from your first prompt through the last reply, then copy.
- Paste that conversation as one trace in Interview Lab.
- Do not upload a Google Takeout or Gemini Apps Activity dump. Those include chats that are not part of this assignment.
Paste as text, or save the copied chat as a .txt / .md file and import it.
Cursor and other coding agents
Coding agents already write session files on disk. Import those files, or copy the Agent chat. Only send the sessions that shaped this assignment.
| Agent | Local session directory |
|---|---|
| Cursor | ~/.cursor/projects |
| Claude Code | ~/.claude/projects |
| Codex | ~/.codex/sessions |
Cursor transcripts are under ~/.cursor/projects/<project>/agent-transcripts. Claude Code and Codex write .jsonl in the directories above.
Cursor
~/.cursor/projects
Copy the Agent or Chat panel, or import the local transcript files.
- In the Agent or Chat panel, open the conversation used for this assignment.
- Copy the full thread — your instructions, the agent replies, and the tool results that changed the work.
- If copying is awkward, ask Cursor to export that transcript as text, then import the file.
- Local files also live under the project transcript directory as .jsonl. Import only the sessions that shaped this assignment.
Paste as text, or import .jsonl / .json / .md files from agent-transcripts.
Claude Code
~/.claude/projects
Import the JSONL session files Claude Code already writes locally.
- Finish the assignment in the same working directory you used with Claude Code.
- Open ~/.claude/projects and find the folder whose name matches that project path.
- Import the .jsonl session files that cover this assignment. Each file is one trace.
- Redact secrets in the file before you upload it. Command output can include keys and private code.
Import .jsonl session files. Text and Markdown exports also work.
Codex
~/.codex/sessions
Import the local Codex session directory.
- Open ~/.codex/sessions on the machine where you ran Codex.
- Find the session files for this assignment. They are .jsonl.
- Import those files in Interview Lab. Add each session as its own trace.
- Drop sessions that did not change the deliverable.
Import .jsonl session files from ~/.codex/sessions.
Redact secrets before you share.
Traces can include prompts, tool inputs, command output, private code, and credentials. Remove anything the hiring team should not hold. The quality gate still runs after redaction.
- API keys, tokens, passwords, and cookies.
- Customer data, employee records, and anything covered by an NDA.
- Private repository contents that are not part of the brief.
- Personal messages that rode along in an agent terminal log.
Upload inside Interview Lab.
On the assignment page, paste each conversation or use Import trace files. Accepted types: text, Markdown, JSON, JSONL, log, and CSV. Maximum 1 MB per file.
Thin traces come back before scoring. If the paste is only the final answer, add the prompts that produced it.
Open Interview Lab if you are practicing. Hiring-team invitations use the link in your email.