01
The newsroom verification checklist
Adapted fact-check practice: every number, quote, and named entity confirmed against a primary source.
Pillar 03
Critical Thinking & Verification
Evaluate AI output before trusting or using it, the muscle every recruiter looks for first.
You are here · pillar 03 of 07
Critical Thinking
← Interaction · Ethics →
A working definition
What this pillar is, what it isn’t, and why it matters.
What it is
+
Treating AI output as a draft, not an answer.
Verification means knowing which claims in an AI response need a source, how to check them, and what to do when the citation looks plausible but is fabricated.
What it isn’t
−
Distrusting every output by default.
Skepticism that blocks any use is just as unhelpful as blind trust. The skill is calibrated trust, knowing what to check and what to let through.
Why it matters
→
This is where most professional AI mistakes happen.
Hallucinated statistics, invented citations, and confidently-wrong summaries are the failures that get junior hires noticed, for the wrong reason.
1 in 4
AI-generated claims in business writing contains at least one verifiable error. (Stanford HAI, 2025)
What’s inside this pillar
The sub-skills the assessment is actually measuring.
Each pillar breaks into 4 sub-competencies. The assessment scores each one independently so you can see what to work on, not just an overall band.
02
Detecting hallucinated citations
Common patterns, plausible journal names, real authors paired with non-existent papers, well-formed DOIs that resolve to nothing.
03
Cross-checking with a second tool
When and how to use a second model or a search engine as a verification layer, not a co-author.
04
Calibrating your own confidence
When to ship a draft, when to add an asterisk, and when to scrap and start over.
Try a sample
A real question from this pillar.
The full assessment has 41 of these for Pillar 03. There’s no time limit and you can pause anytime.
Sample · 1 of 41
Which is the best reason to request the AI’s assumptions explicitly?
One correct answer
Proficiency levels
Where the assessment will place you, and where to aim.
“Proficient” is the workplace-ready bar. Most graduates leave school at Developing; recruiters look for Proficient or above.
Level 01
Novice
Trusts most AI output at face value. Doesn’t routinely fact-check claims or verify citations. Surprised when an output turns out to be wrong.
Score · 0–49%
Level 02
Developing
Knows verification matters, checks ‘important’ claims, but inconsistently. Sometimes misses fabricated citations and overconfident statistics.
Score · 50–74%
Level 03
Proficient
Verifies on a structured checklist: every numeric claim, every quote, every named entity. Catches most hallucinations before they ship.
Score · 75–89%
Level 04
Advanced
Calibrated trust, verifies the right things at the right depth, uses second-tool cross-checks fluently, teaches verification practice to peers.
Score · 90–100%
Reading list
All articles →
To go deeper on this pillar
Trust Takes Years to Build and One Unverified Send to Burn
A senior colleague forwards your memo to a client without reading every line. That forward is the whole game. It means your name is safe …
The $440,000 Citation: How Fabricated Sources End Up in Real Deliverables
A consulting giant handed a government a report full of sources that did not exist. Here is how AI fabricates citations that look flawles…
Four Outages in 90 Days: What Amazon’s Vibe-Coding Mess Teaches Every New Hire
When the people building AI tools skipped the review step, things broke in public. Amazon's run of outages holds a lesson that has nothin…
Start with the assessment
Find out where you stand across all 7 pillars.
The free AI Literacy Assessment scores you independently on each pillar and shows you exactly where to focus next.
✓
10 minutes · No email required to see results