Find the unsupported claim
- verification
- research
An unsupported claim is a statement specific enough to be checked, consequential enough to matter — and attached to no evidence. The marker is not tone: fluent text styles invented sentences exactly like true ones. The load falls on you, so learn the shapes that deserve an immediate second look.
The shapes to catch
- Precise numbers with no source. “Grew 23%”, “saves 6 hours a week” — a number is a promise of measurement. Where is the measurement?
- Predictions stated as facts. “Three vendors will switch pricing by 2028.” Nobody knows this; it is a scenario wearing a date.
- Anonymous authority. “Experts agree”, “studies show”, “industry consensus” — which experts, which studies? A claim that names nothing cannot be checked.
- Superlatives. “The only solution”, “the best approach”, “the strongest effect in a decade” — usually someone’s framing, not a finding.
- Specifics about people and organizations. Roles, titles, dates, quotes — checkable in minutes and frequently wrong.
The shapes that are fine
- Qualitative generalities: “customers increasingly expect transparency” asserts nothing measurable.
- Statements that name their evidence: “our churn was 4.2% according to the analytics dashboard” — you can open the dashboard.
- Self-limiting lines: “check the official source before acting” sends you somewhere better.
The three-part test
For each claim, ask: is it specific? is it consequential? does it name its evidence? Specific + consequential + no evidence is the danger zone. Specific + consequential + named evidence is a five-second check, not a problem.
A bad example
A briefing says: “Our sector grew 23% last year, and experts agree the shift is inevitable.”
Two claims, zero sources, one meeting away from becoming the number your leadership quotes for a year.
A better example
“The sector grew, per the [industry association’s] 2025 report — the 23% figure comes from its page 4 table; the ‘inevitable’ line is our own framing and should be labeled as such.”
Evidence attached, opinion labeled. Both can stay in the deck.
Make the pass concrete: mark every number, date, name and prediction in the draft. Anything marked that cannot point at checkable evidence gets a “verify” tag before the document moves another step. On a single page that pass takes a minute — and it is the difference between catching a fabricated detail in private and discovering it in front of an audience.
Practice
Spot the unsupported claims
A strategy briefing goes to your leadership team tomorrow. You have one pass to mark the claims that need evidence attached before the deck ships.
Flag every claim that should not be trusted as-is.
The AI answer
Market update: The sector grew 23% year over year. Three of the top five vendors will move to usage-based pricing by 2028. Customers increasingly expect transparent pricing. Our churn was 4.2% last quarter, according to our analytics dashboard. Experts agree the shift is inevitable. The competitor's pricing page lists three tiers.
Hint
Scan for digits, dates and 'experts' — each one needs a source or a label.
One statement names where to find its own evidence. Is that a problem?
Why this is the answer
The flagged claims carry specifics — a percentage, a dated prediction, an unnamed authority — with nothing to check them against. The others assert nothing measurable, name their source, or point at a page you can open. Specificity plus consequence plus no evidence equals the danger zone.
Transfer
- Inbox triage: scan for digits, dates, names and futures — each needs a source or a label.
- AI-assisted drafts: run the three-part test before anything leaves your hands.
- Your own writing: unsupported claims are contagious — catch them in your sentences too.
Next
Next: Ask for uncertainty — instead of hunting for unsupported claims after the fact, ask the model to expose its own assumptions up front.