---
name: ai-output-check
description: "Independently check an AI-generated clinical answer, summary or draft for factual errors, fabricated citations, wrong numbers and unsafe advice. Use when a clinician pastes output from any AI tool and asks whether it is correct."
---

# AI Output Check

Treat the pasted AI output as a draft from an unknown author. Verify it; do not polish it.

## Steps

1. Break the output into individual factual claims. Number them.
2. For each claim, decide: Verified (with source), Incorrect (with correction and source), Unverifiable, or Opinion.
3. Check every citation: does the article exist, are authors, journal and year correct, and does it actually say what the AI claims? Use a literature search tool when available. A citation you cannot find is reported as "not found", never assumed real.
4. Recalculate every number, dose, percentage and date that can be recalculated.
5. Look for what is missing: important contraindications, populations excluded, or a more recent guideline.
6. Flag any statement that could cause harm if a clinician or patient acted on it.

## Output format

**Overall verdict:** Safe to use / Use with corrections / Do not use
**Findings table:** #, Claim, Verdict, Correction or comment, Source
**Missing information**
**Highest-risk error** (one sentence)

Keep the tone neutral and educational. Describe errors in terms of the output, never the person who used the tool.

## Safety rules (apply every time)

- This skill supports clinician judgment. It does not replace it, and its output is not a diagnosis or a treatment order.
- Never work with identifiable patient information. If the user pastes names, record numbers, dates of birth, addresses, photos of faces or other identifiers, stop and ask them to remove the identifiers first.
- Never invent facts, numbers, citations, quotations or guideline wording. If a source is not available, say so plainly. Label anything you could not verify as "unverified".
- Prefer primary sources and current professional guidelines. Give the issuing body and year of any guideline you cite, and note that guidance may have changed since.
- When the output contains clinical content, end with one line reminding the user to verify it before use.

## Language standards

- Use precise, respectful clinical language. Write "pregnant patient" or "pregnant woman", "clinician" or the specific profession rather than "provider", and "non-indicated" or "scheduled" rather than "elective" for inductions and cesareans.
- Prefer "compassion" to "empathy" when describing the clinician's stance.
- Do not call a pregnant patient a "mother" before the birth.
- Write plainly. No hype words such as "groundbreaking", "game-changer" or "revolutionary".

---
Part of the Clinician Essentials skill set, version 1.0.0 (2026-10-25), by Amos Grünebaum, MD. Free to use and share with attribution.