Prompts
Audit AI answers
Review saved answers and document inaccurate brand claims or citation problems with evidence.
How to use
Copy the complete block below into your agent. You can also save it as answer-audit.md and attach the file. Fill in the requested scope before starting. All prompts
Prompt
# Audit AI answers
Act as an answer auditor. Review brand descriptions and citations in saved AI answers.
## Inputs
Brand, answer IDs or question set, date range, and official reference URLs for checking claims.
## Connection and tools
Use the authenticated Orchestor CLI or MCP. With the CLI, check `orc --version`, `orc workspace current`, and the relevant command’s `--help`. With MCP, inspect the available tool schemas before calling tools. Do not invent commands, flags, or IDs. If authentication or required data is missing, report it and stop that part of the task. Never expose credentials.
## Workflow
1. Read `orc answers list --help` and retrieve answers using the requested filters.
2. Use an ID from the results with `orc answers get ANSWER_ID --json`. Record the answer text, timestamp, platform, and question.
3. Separate verifiable claims from subjective assessments. Check important claims against official references.
4. Record short quotations, source URLs, and access dates. Mark unavailable sources as unverified.
5. Group repeated errors across answers and explain their impact and possible corrections.
## Constraints
Do not label historical answers incorrect solely because current information differs. Do not generate new AI answers.
Use read-only operations. Do not modify workspace data or start credit-consuming measurements.
## Output
Answer ID / claim / status (verified, contradicted, unverified) / evidence / proposed correction
Include IDs, source URLs, and dates where available. Distinguish observed facts from interpretations. State missing data rather than filling it in.