AINUA AI Academy
F4.2 · Foundations

Beginner · responsible practice · 20 min core lesson · 4 min quick note

Verify claims, sources and calculations

Open the evidence and do the arithmetic.

Information and dataHuman agencyEthics, safety and citizenship
Nia and Kio introducing Verify claims, sources and calculations
Cold open

There is a moment, just after a good answer arrives, when checking it feels like an insult to the answer. That moment is the entire subject of this lesson.

By the end, you can

separate the claims in a piece of work by how load-bearing they are, verify the ones that matter against real sources, and recalculate any figure that will inform a decision.

Quick note · 4 min

Not every claim deserves the same scrutiny. Verify the ones that would do damage if wrong, and be honest about the rest.

Do now: Take one document and mark its three most load-bearing claims, then verify those three.

A source you have not opened is a suggestion. A figure you have not recalculated is a rumour.
Why it matters

Verification is the skill that makes everything else usable. Without it you are forwarding confident text; with it you are producing work that survives a reader who checks. That is the whole difference between an AI user and a professional who uses AI.

01

Sort claims before you verify them

Verifying everything is impossible, so people verify nothing and call it pragmatism. The way out is triage. Go through the work and mark each claim by what breaks if it is wrong: nothing, an argument, a decision, or a person.

Claims that could affect a decision or a person get verified properly. Claims that only support an argument get checked lightly or softened. Claims that nothing depends on can often just be cut — a great deal of unverifiable filler exists purely because it sounded authoritative.

Also mark what kind of claim each is. A fact has a source. An estimate has a method and a range. An opinion has an owner. Presenting an estimate as a fact is one of the most common and least noticed forms of dishonesty in ordinary professional writing.

Triage is not laziness. It is how verification survives contact with a real workload.
02

Triangulate, and open what you cite

For anything that matters, find two independent sources. Independent means genuinely separate — not two outlets reprinting the same press release, and not a search result quoting the source you already had. If both trace to one origin, you have one source wearing two coats.

Prefer the origin over the summary. The regulator's own publication over an article about it. The company filing over the news story. The law itself over the explainer. Every layer of retelling introduces compression, and compression is where conditions and exceptions disappear.

Record the access date. For anything that changes — prices, rules, who holds a position, what a platform permits — a claim without a date is not verifiable at all, because neither you nor your reader knows what it was true of.

03

Do the arithmetic yourself

Numbers carry authority that sentences do not, and they are exactly where these tools are weakest. Any figure that will inform a decision should be recalculated independently — on paper, in a separate formula, or against the source document.

Check the units and the base as well as the arithmetic. A percentage of what? Per household or per person? Gross or net? Which currency and at what rate? Most numerical disagreements in real organisations turn out to be definitional rather than mathematical, and the definition is what nobody wrote down.

When a figure comes from a model rather than a source, treat it as a hypothesis. Sometimes it is a very good hypothesis that points you at the right document. It is never itself the evidence.

If a number will change what someone does, calculate it twice by two routes.