01
Precise numbers
“73.4% of firms…” with no source is a red flag, not a proof point.
Lesson 04 / 07
Trust · Verify · Gate · ~9 min
When AI is confidently wrong
Spot fabricated certainty, then use a verification checklist before anything ships.
Confidence is not evidence
Models are trained to produce plausible language. Plausible is not the same as true.
Hallucinations show up as invented stats, fake citations, wrong CRM fields, and “policies” nobody wrote — often in a calm, authoritative tone.
- Precise fake stats92%
- Invented citations84%
- Wrong CRM fields71%
- Grounded summary of pasted notes18%
Fig. 04 — Relative risk patterns (illustrative)
High-risk patterns
02
Fake citations
URLs, paper titles, and “studies” that sound real but do not exist.
03
Silent invention
Filling gaps in CRM notes instead of saying “not in the input.”
04
Policy fiction
Inventing internal rules or legal positions with total confidence.
Check
Which sentence is most likely a hallucination risk in a client email?
Checklist
Verify before it leaves the building
Check
Best default when the model is unsure?