CRAAP Test
A systematic evaluation of each AI response using the classic academic framework: Currency, Relevance, Authority, Accuracy, and Purpose — adapted specifically for AI-generated content.
How it works
CRAAP is an established source-evaluation method originally developed by librarians at California State University, Chico. In SeekBox, the analysis model (default: Groq) applies the five CRAAP criteria to every model’s output:
- Currency – How up-to-date is the information? Does it reference recent events or data?
- Relevance – Does the response directly address your query and your context?
- Authority – What credentials or knowledge base does the model draw from? Does it cite verifiable sources?
- Accuracy – Are facts verifiable? Are there internal contradictions or known hallucinations?
- Purpose – What is the apparent goal of the response? Is bias detectable?
When to use it
- Academic research or fact-heavy reports where source quality matters
- Medical, legal, or financial queries (always consult professionals, but use CRAAP to triage AI answers)
- Evaluating marketing claims or product reviews generated by AI
- Teachers assessing student work that used AI assistance
- Anyone who wants a standardized, transparent rubric instead of subjective opinion
This mode organizes model output for review. Its conclusions still require appropriate source checking and human judgment.
Sources
- NIST AI Risk Management Framework — National Institute of Standards and Technology. Accessed 2026-07-20.