Resource
Fix Invalid JSON from AI
The JSON won't parse and you can't see why. Deterministic cause-sniffing — trailing commas, single quotes, unclosed brackets — and the repair prompt that fixes it.
Hallucination Detection Prompt
Catch the confident invention — check an AI output's claims against its source and flag every statement that isn't supported, with the unsupported span quoted.
Overview
The dangerous AI failure isn't being wrong loudly — it's being wrong confidently, with an invented fact that reads exactly like a real one. This prompt audits an output against its source material: it extracts the checkable claims, marks each as supported, contradicted, or unsupported, and quotes the exact span that isn't backed by the source — so hallucinations are caught before a user trusts them.
How to use this resource
-
Assemble the output and its source
Gather the AI output you want to check and the source material it should be based on. The audit judges claims only against this source, so include everything the answer was meant to draw from.
-
Open this resource in AI Output Validator
Load the prompt into AI Output Validator and paste in the source and the output. The tool runs the check so every claim is classified without copying spans by hand.
-
Review the classified claims
Read each claim marked supported, contradicted, or unsupported with its quoted span, focusing on the unsupported claims a user would act on.
-
Feed the unsupported claims back
Use the flagged spans to correct the output or tighten the prompt that produced it, then re-run the check to confirm the inventions are gone.
Why This Works
- Classifying every claim catches the plausible invention, not just obvious errors
- Quoting the unsupported span makes the finding actionable, not vague
- Judging only against the source is what separates hallucination from 'sounds right'
Best for
- RAG systems and summarizers grounded in source text
- High-stakes outputs where a wrong fact is costly
- Evaluation pipelines needing a hallucination check
Not for
- Open-ended generation with no source to check against
- Grounded-answer verification specifically — use the Groundedness Check Prompt
Use cases
- Checking a RAG or summary output against its source
- Catching invented facts before they reach a user
- Auditing agent answers for unsupported claims
FAQ
How do I use a hallucination detection prompt to check an AI answer?
Paste the SOURCE (the ground truth the answer should be based on) and the AI OUTPUT into the prompt, then run it in your own AI tool. It extracts the output's checkable claims and marks each SUPPORTED, CONTRADICTED, or UNSUPPORTED, so you review the flagged spans instead of re-reading the whole answer. The key rule: it judges only against your source, never against the model's own knowledge.
What should I paste into the hallucination detection prompt?
Two things: the SOURCE material the answer was supposed to be grounded in, and the OUTPUT you want checked. The audit compares one against the other, so without a source there is nothing to verify against — it can only flag claims your source doesn't back, not facts about the world. Include everything the answer was meant to draw from, or supported claims may look unsupported.
What output does the hallucination detection prompt produce?
It returns a list of the output's discrete factual claims, each classified SUPPORTED, CONTRADICTED, or UNSUPPORTED against the source. For every contradicted or unsupported claim it quotes the exact span and notes what the source actually says (or that it is silent), then ranks the unsupported claims a user would act on as high-risk. The result is a triaged list of what isn't backed, not a rewrite.
Does this prompt guarantee the AI answer is accurate?
No — it checks whether each claim is backed by the source you provide, not whether the answer is true in the real world. A claim can be SUPPORTED by a wrong source, or flagged UNSUPPORTED yet still correct, so a clean audit is a grounding signal, not an accuracy guarantee. You run it in your own AI tool, and the final factual review stays with you.
More resources from AI Output Validator
Resource
Validate AI Output — Catch Format Violations
Paste the response, get the verdict: real JSON parsing, missing-field detection, and a repair prompt for everything found.
Resource
Agent Evaluation Scorecard Prompt
Grade agent output the same way every time — a rubric scoring correctness, grounding, safety, tone, and completeness, with a pass threshold instead of a gut call.
Resources that pair well
Resource
Force JSON Output from AI
Stop getting 'Sure, here is the JSON…' — the output-contract pattern that forces models to return only parseable JSON: schema, example, and a strict rule block.
Resource
Extract Data From Text with AI
Free text in, named fields out. The extraction prompt pattern that turns any unstructured text into consistent, parseable records.
Resource
Classify Support Tickets with AI
Billing, Technical, Account, How-To, Feature Request — ticket triage with definitions that decide the borderline cases for the model.
Related tools
Tool
AI Output Validator
Paste an AI's output and validate it against the expected format — with a repair prompt for every failure found.
Projects that use this resource
Project
Build an AI Support Agent with AI
The full path to a support agent you can put in front of customers — write its instructions, ground it in your docs, route and handle tickets, then evaluate and cost-control it before it goes live.
Project
Build an AI Research Assistant with AI
The full path to an AI research assistant — define its scope, organize the source corpus, ground responses in references, extract key facts, synthesize findings, check groundedness, then validate it for use.
Project
Build a RAG System with AI
The full path to a retrieval system that returns grounded answers — understand the corpus, chunk and ground it, extract and classify the metadata, then evaluate that retrieval actually works.
Workflows that use this resource
Workflow
AI Agent Evaluation Workflow
Find out whether an AI agent behaves before users do — define what correct means, build test scenarios with expected outputs, catch failures and hallucinations, then regression-test each version.
Guides for this resource
Guide
How to Stop AI From Inventing Missing Data
When a source is missing a field, AI tends to fill the gap with a plausible guess instead of saying it isn't there. Here's how to make the model mark missing data explicitly — and check the result before you trust it.
Guide
How to Review AI Output Against a Source Document
An AI summary reads perfectly, but you can't tell if every sentence is in the source or whether a "may" hardened into a "will." Here's how to review AI output claim-by-claim against the document it came from, so unsupported and overstated claims surface before you rely on them.