Do not ask whether AI is trustworthy in the abstract. Identify the claim, support, recency, uncertainty, decision, affected people, and cost of error, then choose a proportionate verification route.
Plan for this page
Understand the idea, test it actively, then explain it in your own words.
You will leave with
separate factual claims from ideas, style, opinions, and instructions distinguish source presence from exact claim support match verification and human review to uncertainty and consequences
Time
35 min
Before you begin
AI from Zero is useful preparation, but it is not a locked prerequisite. Each unit states the concepts it uses.
Do this now
Read the outcomes and useful terms before the first section.
separate factual claims from ideas, style, opinions, and instructions
distinguish source presence from exact claim support
match verification and human review to uncertainty and consequences
Terms you will use
These definitions prepare you for the reading; you do not need to memorize them.Confabulation / hallucination
Generated content that is false, erroneous, unsupported, or inconsistent, without requiring an intention to deceive.
Claim support
The relationship between an exact statement and evidence that is relevant, authoritative enough, current enough, and correctly interpreted.
Verification
A scoped process for checking claims, sources, dates, uncertainty, and fitness for the decision—not a promise of absolute certainty.
01
Classify before you verify
A brainstormed slogan, a summary of supplied text, a claim about law, a medical recommendation, and a fictional story need different checks. First mark statements about the world, time-sensitive details, calculations, quotations, causal explanations, and professional interpretations.
Then identify the intended use. A private idea draft may tolerate uncertainty that a public report, financial decision, safety instruction, or application affecting another person cannot.
02
A real source can still fail the claim
Links may point to search results, secondary summaries, old pages, unrelated passages, or documents that contradict the generated sentence. Even a primary source has scope, definitions, date, jurisdiction, methods, and limitations.
Open the source. Locate the relevant passage or data. Check whether it supports the exact wording, whether an omitted qualifier changes the meaning, and whether independent confirmation is needed.
03
Verification effort follows stakes
For low-stakes ideation, recognizing uncertainty may be enough. For current schedules or prices, check the official current source. For consequential medical, legal, financial, employment, security, or public claims, use stronger primary evidence and appropriate qualified human review.
Verification also includes knowing when not to act. Preserve unresolved uncertainty, record conflicting evidence, narrow the claim, ask for help, or delay the decision when the evidence threshold is not met.
Interactive model
The CLAIM verification loop
Selected layerC · Classify
What exact claim is being made: fact, forecast, quote, calculation, interpretation, advice, or idea?
Read the complete text alternative
C · ClassifyWhat exact claim is being made: fact, forecast, quote, calculation, interpretation, advice, or idea?
L · LocateFind the closest relevant primary or authoritative evidence and its date, jurisdiction, and method.
A · AssessDoes the evidence support the exact wording? What conflicts, limitations, incentives, or uncertainty remain?
I · ImpactWho may be affected and what happens if the claim is wrong, incomplete, outdated, or misapplied?
M · Make the decisionUse, revise, qualify, seek review, independently confirm, or do not act; record the route when it matters.
Misconception to correct
“Verification means finding any second page that repeats the answer.”
Repeated wording may come from one weak origin. Verification looks for relevant primary evidence, independence where useful, exact support, current scope, and consequence-aware review.
Active checks
Decide, then compare the reasoning
No grade, score, or streak. Progress records completed learning objects, not your worth or ability.01
An AI cites a genuine government report, but the report discusses a different year and population. What is the evidence status of the generated claim?
02
An assistant summarizes a new employment-law obligation for a company decision. Which route is proportionate?
Explain back and transfer
Build a verification route for one claim
Choose a consequential claim you might encounter. Classify it, identify the closest source, state the scope and date, name the cost of error, and decide what confirmation or human review is required.
Answer both checks before completing.
Source context
What supports this unit and where it stops
Evidence status: established
Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1)U.S. National Institute of Standards and Technology
Uses confabulation as a defined generative-AI risk and supplies organizational risk actions; not a substitute for domain-specific evidence or professional review.