Evaluating AI-generated claims
Verify AI output with sources suitable for the claim and decision.
What you will learn
- Break generated output into checkable claims.
- Select primary or authoritative sources.
- Record unsupported statements and uncertainty.
Confident wording is not evidence. Verification starts by isolating each material claim and asking what source would be authoritative for that claim, in that jurisdiction and at that date.
Core method
Confident wording is not evidence. Verification starts by isolating each material claim and asking what source would be authoritative for that claim, in that jurisdiction and at that date.
- Define the exact decision or claim before analysing detail.
- Use evidence that is relevant to the stated context.
- Record uncertainty instead of replacing it with confident wording.
Applied case
An AI assistant gives a plausible university application deadline without citing the admissions page. The date may be outdated or belong to another programme.
Apply the method
Describe the verification steps before advising an applicant.
Review answer guidance
Open the official programme and admissions pages, confirm the entry year, candidate category and time zone, check the page's update date, and preserve the link used. If official sources conflict, contact admissions rather than guessing.
Use in an assessment
Apply the method to the information provided in the task. A strong response makes its reasoning traceable and does not rely on specialist facts that are absent from the evidence.
- Read the requested outcome and constraints first.
- Eliminate responses that exceed the available evidence.
- Review whether the conclusion answers the precise question.