AI Output Audit Workbench — CRAFT PD Series

Load an existing audit from Markdown

Paste Markdown from a previously exported audit (or from an LLM response). Fields below will be populated.

Context

Set up the audit — what are you checking and at what level?

Your Name Grade Level & Subject AI Tool Used Your Subject Expertise Level (honest self-assessment)

CtM Protocol

TASK — What did you ask the AI to do?

Copy your exact prompt. Specificity matters for auditing — a vague prompt explains a vague output.

Your Prompt (paste the exact text you sent) Intended Use (how will this output be used?)

BEFORE — Your Expectations

Write these before reading the AI output. This is the most important step — it activates your own knowledge and prevents you from anchoring to what the AI says.

Why this matters: A common misconception is that subject-matter expertise is sufficient to catch AI errors. But expertise creates blind spots — you're more likely to accept a confident-sounding claim in your own field because it "sounds right." The protocol is the upgrade.

2–3 things you already know to be true about this topic What do you expect the AI to produce? Where do you predict the AI might get it wrong?

AFTER — Examine the AI Output

Paste the AI's full response, then systematically audit it using the three panels below.

AI Output (paste the full response)

🔍 Panel A: Claims Inventory

Extract every factual claim, number, citation, or standard code. Rate each one.

Claim / Citation	Rating	Source / Note

🧠 Panel B: Reasoning Trace

Where does the AI's logic break? Look for: confident-sounding wrong statements, oversimplifications that distort meaning, and "sounds right but I can't verify" moments.

Logic breaks, gaps, or distortions

📐 Panel C: Standards Fit

Does the content actually address the standard it claims to align to? Or is it "standards-adjacent"?

Standard code(s) the AI cited Does the activity actually meet this standard? Notes

TAKEAWAY — Verdict & Classroom Protocol

Two takeaways: (1) your verdict on this output, and (2) the student-facing protocol you'll use in class.

Overall Verdict

✅ Keep
Usable as-is

⚠️ Fix
Usable with edits

❌ Cut
Too many errors

Errors Found (list each one) Edits Needed Before Classroom Use Student-Facing CtM Exercise (the audit task you'd give students) Self-Reflection (what did the comparison reveal about YOUR understanding?)

Iterate with an LLM

Have a second AI audit the first — or generate a fix.

Copy the audit prompt

We build a detailed prompt from every field above, plus this workbench's citation.

Paste the LLM's fenced ```markdown block back into "Load an existing audit from Markdown" above.

Export Your Audit

Your data stays in the browser — nothing leaves your computer unless you click an export button.

Google Docs · Word · Pages

Rich-text paste keeps headings and formatting. The .doc file opens in Word and Google Docs.

Markdown

Includes YAML front matter for repository integration.

Student Worksheet

Exports just the student-facing CtM exercise with blank fields. For a more stylized, print-ready handout, use the CtM Template.

📄 Open CtM Template

Print

Print-friendly version of your completed audit.