Use the ladder with any generative AI tool. The point is the workflow, not the brand.
Clarify
Define the question before prompting
Write down what you actually need to know, what kind of evidence would count, and what constraints matter.
Generate
Ask for a useful starting point, not the final answer
Use AI to surface possible concepts, vocabulary, counterarguments, or directions. Treat the response as a hypothesis generator.
Extract
Separate the response into checkable claims
Underline or list every factual, causal, statistical, historical, or interpretive claim that would need support.
What is the AI asking you to believe?
What kind of source would actually support it?
Verify
Leave the AI response and search independently
Use library databases, primary sources, official data, scholarly literature, or other sources appropriate to the question.
Does the cited or implied source actually exist?
Does it support the specific claim?
Is it appropriate for this question?
Is it current enough for the topic?
Interrogate
Look for what the response leaves out
Ask about uncertainty, disagreement, missing perspectives, bias, power, context, and consequences.
Decide
Keep, revise, or reject
After checking the evidence, decide what survives. Replace unsupported claims with sourced material, revise oversimplified reasoning, and reject anything that cannot be defended.
Supported, relevant, and appropriate.
Partly supported or missing needed context.
Unsupported, fabricated, misleading, or inappropriate.
Explain how AI contributed when required or useful.
The ladder changes the center of gravity.
The AI output is not the evidence. It is the beginning of a research process that returns authority to sources, context, and human judgment.
