PARADOX
Check for sense.
Brainstorming, rewording, agendas and early drafts.
- Read it critically
- Remove invented specifics
- Confirm it suits the purpose
Confident is not
the same as correct.
AI can produce useful work without knowing whether that work is true. Verification is how a business decides what can be used, what must be checked and who owns the final call.
Match the check to
the consequence.
Not every AI answer needs an investigation. A list of meeting-title ideas is not the same as a safety instruction, contract clause, cost estimate or client recommendation.
The higher the consequence of being wrong, the stronger the evidence and human review should be.
A PROPORTIONATE APPROACHThree levels of
Three levels of
verification.
Check the evidence.
Quotes, calculations, proposals, policies and operational advice.
- Trace important claims to sources
- Recalculate numbers independently
- Record assumptions and uncertainty
Require accountable review.
Safety, legal, financial, employment, privacy and regulated decisions.
- Use authoritative sources
- Obtain qualified specialist input
- Keep a clear human decision owner
THE METHODFrom plausible answer
From plausible answer
to usable decision.
01
Define the claim.
Separate the actual fact, calculation or recommendation from the fluent explanation surrounding it.
02
Trace the evidence.
Identify where the information came from, whether the source exists and whether it genuinely supports the claim.
03
Test independently.
Recalculate, compare against primary material, use a second method or ask someone with the relevant expertise.
04
Make the decision visible.
Record important assumptions, unresolved uncertainty, the reviewer and the person accountable for acting.
WHAT THIS LOOKS LIKEEveryday AI.
Everyday AI.
Real consequences.
“AI checked my tender.”
The review sounds thorough, but it never saw an appendix containing a critical exclusion.
Verify document completeness, trace every conclusion and retain a human release decision.
“The calculation looks right.”
The arithmetic is presented cleanly, but an incorrect unit or hidden assumption changes the result.
Recalculate independently and expose the assumptions before relying on the number.
“It cited a source.”
The source may not exist, may be outdated or may not support the statement attributed to it.
Open the primary source and check the exact claim—not merely the presence of a citation.
A verification system
your team can use.
A defined list of outputs requiring checks, evidence standards, review thresholds, accountable owners and a repeatable verification checklist.
This is not an automated stamp of accuracy. It is a practical control that keeps important decisions attached to evidence and human judgement.
Explore AI output verification ↗Find a way
forward.
Take the free AI Risk Check ↗
A proportionate check.A decision someone owns.