Examine worked example showing how to separate claims from evidence using the claim-anatomy visualisation [VERIFY: not defined in source]

Separate Claims from Evidence

What you will do here

Examine a worked example. Learn to spot the difference between a claim and the evidence that supports it. Practise identifying unstated assumptions. Build a skill you will use in every AI interaction ahead.

Why this matters for AI users

Generative AI often sounds certain. It produces confident sentences that look like facts. Some are well-supported. Others are not. Your job is to tell the difference before you act on the information.

This skill protects you from three common traps:

• Acting on a bold statement that lacks backing
• Mistaking a plausible pattern for proven cause
• Repeating an AI's assertion as if you verified it yourself

Worked example: the grant application

Read this AI-generated paragraph about a fictional regional council applying for federal infrastructure funding.

"The Riverina Greenways project will create 340 local jobs within 18 months. This is because similar trail projects in comparable regions have consistently delivered employment outcomes above baseline forecasts. The council should therefore prioritise this initiative over competing road upgrades."

You have two minutes to use this in a team briefing. What can you confidently relay? What needs checking?

Anatomy of the paragraph

Let's break it into parts you can inspect. This is the claim-anatomy visualisation [VERIFY: not defined in source] applied to the example.

The core claim (what the AI wants you to accept)

"The Riverina Greenways project will create 340 local jobs within 18 months."

This is a prediction about the future. It uses a specific number and timeframe. That specificity can make it feel solid. It is not. It is a forecast that needs support.

The offered evidence (what the AI provides as support)

"Similar trail projects in comparable regions have consistently delivered employment outcomes above baseline forecasts."

Notice what this actually says. Past projects in other regions did better than expected. It does not say they hit 340 jobs. It does not say they did it within 18 months. It does not say Riverina matches those other regions.

The implied recommendation (what the AI pushes you toward)

"The council should therefore prioritise this initiative over competing road upgrades."

This jumps from the claim to a decision. The word "therefore" implies the evidence proves this choice. It does not.

The unstated assumptions (what the AI smuggles in)

These gaps matter most. They include:

• The 340 jobs figure comes from a reliable projection method
• Riverina's economy, labour market, and geography match "comparable regions"
• "Above baseline forecasts" means enough jobs to justify the project
• Road upgrades would deliver fewer jobs or less value
• Employment is the only criterion worth considering

None of these appear in the text. All of them need testing before you repeat or act on the claim.

How to apply this yourself

Use this four-step check on any AI output you intend to use for a consequential decision.

Step 1: Isolate the core claim

Underline or copy the exact sentence the AI most wants you to accept. Look for future predictions, causal statements, or recommendations disguised as facts.

Step 2: Map the offered evidence

List only what the AI actually gives you. Do not add your own knowledge yet. Be strict. If the evidence mentions "similar cases," note the specifics it omits.

Step 3: List the unstated assumptions

Ask: what must be true for the claim to follow from the evidence? Write these down. They are your risk register. Each one is a point where the AI might be wrong without admitting it.

Step 4: Name what you still need

Before you act, brief, or recommend, identify the missing verification. This might be a data source, a local comparison, or an expert consultation.

Quick practice: spot the gap

Read this shorter AI output. Identify the core claim, the evidence, and at least one unstated assumption.

"Adopting the FinSecure chatbot reduced customer complaints at Meridian Bank by 22 percent in the first quarter. Organisations seeking to improve retention should implement similar AI customer service tools."

Core claim:

Evidence offered:

Unstated assumptions (at least one):

[Activity: learner completes this brief reflection before proceeding. Manual completion required.]

What to carry forward

You now have a re-usable method for inspecting AI-generated text. The claim-anatomy visualisation [VERIFY: not defined in source] is not about catching lies. It is about surfacing what is actually on the page versus what the AI makes feel certain.

In your next topic, you will apply this method in a consequential scenario where the pressure to trust is higher and the time to check is shorter.