HomeCasa Prompts

Hallucination check for an AI answer, claim by claim

Review an AI-generated answer: break it into claims, mark which are supported, which cannot be verified and which look wrong, and get a corrected version plus a checklist of what to verify. Designed for ChatGPT, Claude, Gemini and Perplexity. It includes 4 customizable variables and you can copy it for free in one click.

Intermediate · Text · Pending editorial review

Designed for:

Prompt

Act as a fact-checker experienced in spotting errors and fabricated content in AI-generated text.

Material:
- Original question: {{question}}
- AI answer I want to check:
"""
{{ai_answer}}
"""
- Sources I have: {{sources}}
- What I will use the answer for: {{use}}

Deliverables:
1. A numbered list of atomic claims taken from the answer (one checkable idea per line).
2. Type of each claim: fact, number or date, quote or bibliographic reference, opinion or inference.
3. Status of each: supported (with the exact quote from the source), contradicted (with the quote), not verifiable with the material available, or probably wrong (explain why: internal inconsistency, reference that looks fabricated, suspicious precision, information that may be outdated).
4. Overall red flags in the answer.
5. A corrected version that keeps only what is supported and marks anything doubtful as [TO VERIFY].
6. A list of pending checks: what to verify, in which kind of primary source, ordered by importance for {{use}}.

Constraints:
- If you have web search, cite the URL for each check; if you do not, say so clearly and do not present your memory as verification.
- Always distinguish "verified against a source" from "seems right to me".
- Do not invent sources or references to fill gaps.
- For health, legal or financial topics, remind me that the final decision should be checked with a professional.

Format: a table with columns #, claim, type, status and evidence, followed by the corrected version and the list of checks.

How to use it

Paste the full question and answer exactly as the AI gave them. If you have sources, paste or upload them; if not, use a tool with web search, such as Perplexity, so it can cite links. Start with the claims marked as probably wrong.

Usage example

Question: what tax credits exist for home solar panels in Texas? · Answer: ChatGPT text with amounts and deadlines · Sources: none · Use: personal purchase decision.

Sample output

Assuming the answer as pasted, and with no web search in this session: nothing below is verified against a source; I only flag what needs checking.

#ClaimTypeStatusEvidence
1FMLA gives up to [N] weeks of leaveNumberNot verifiableNo sources provided
2The leave is fully paidFactProbably wrongContradicts claim 4 in the same answer
3"Under section [X] of the act…"ReferenceNot verifiableSection number with no link
4Employers may require using paid time off firstFactNot verifiableNo sources provided
5"Most employees qualify"OpinionNot verifiableNo data cited

Red flags

  • Internal inconsistency on pay (claims 2 and 4).
  • Precise legal reference with no source.

Corrected version (excerpt)

Under FMLA, eligible employees can take up to [TO VERIFY: number] weeks of leave for the birth of a child. Whether it is paid depends on [TO VERIFY]…

Pending checks

  1. Leave length and eligibility → U.S. Department of Labor official pages.
  2. Pay rules → the act's text and your state's labor department.

As this is an employment-law topic, have the final version checked by a professional before publishing…

Key facts

Category
Research · Source analysis
Output type
Text
Level
Intermediate
Designed for
ChatGPT, Claude, Gemini and Perplexity
Verified in
Pending editorial verification
Variables
Original question, AI answer, Available sources and Intended use
Language
English
Price
Free, no sign-up
Updated
Nov 7, 2026

Frequently asked questions

Can one AI check another AI’s hallucinations?

It can help spot doubtful claims and inconsistencies, but it only truly verifies when it compares against specific sources. Without sources or web search, the result is a list of suspicions, not a verification.

What are the most common errors in AI answers?

Non-existent references, very precise figures with no source, outdated regulations or prices, and data from different countries mixed together. The prompt explicitly asks to flag these patterns.

What is this prompt for?

Review an AI-generated answer: break it into claims, mark which are supported, which cannot be verified and which look wrong, and get a corrected version plus a checklist of what to verify.

Which AI tools does it work with?

It is designed for ChatGPT, Claude, Gemini and Perplexity. We have not verified it in these tools yet, so results may vary between them.

How do I customize the prompt?

Click “Customize” and fill in the variables (Original question, AI answer, Available sources and Intended use). The text updates instantly and you can then copy it in one click. Values are stored only in your browser.

Do I need experience to use it?

It is intermediate: it helps to know the topic so you can provide good context and review the result critically.

Is this prompt free?

Yes. Every HomeCasa Prompts prompt is free, with no sign-up or subscription. You can also use them in professional projects.

Can I trust the AI output?

Treat it as a high-quality draft: always check data, figures and important claims before publishing or making decisions.

Related prompts