Before You Rely on an AI Answer: Use the Check That Matches the Risk

The AI draft is ready, your next meeting or reply is close, and starting over would waste time. Can we check only the part that could change the outcome?

AI can turn an unfinished task into a polished-looking answer very quickly.

That polish can make a date, a formula, and a recommendation look equally dependable even though they can fail in different ways.

The U.S. National Institute of Standards and Technology, or NIST, uses confabulation for AI content that is confidently presented even when it is false or wrong. In ordinary language, an AI system can sound certain while part of the answer is incorrect. NIST also warns about over-reliance and automation bias, which means giving an automated answer too much weight simply because a system produced it. NIST Generative AI Profile

OpenAI’s current ChatGPT guidance makes a similar point. As checked on 30 August 2026, it says ChatGPT can sound confident while being wrong and recommends checking important facts, quotations, data, technical information, and external references. OpenAI accuracy guidance

So the useful goal is not to treat every sentence as equally risky.

Let’s first ask: What kind of result am I about to depend on?

A three-part check

• Let’s trace a factual claim when it could change what we do.
• Let’s test an output when it has to work correctly.
• Let’s show the reasons when AI recommends one option over another.

1. If one fact can change your next step, trace it to the original source

Imagine a circular from your condominium management, school, or workplace is shared in a WhatsApp or Telegram group.

You ask AI for a quick summary because you mainly need to know when the change begins and whether it applies to you.

The answer may contain several useful paragraphs, but perhaps only two claims will affect what happens next:

• the start date;
• the people covered by the notice.

Those claims carry the consequence.

To trace a factual claim means returning to the document or organisation that issued it, rather than comparing one summary with another.

Open the original notice.

Find the date yourself, then find the line that explains who is covered.

If the AI gives you a citation, open that page as well. OpenAI’s current search guidance says search results and citations can be incomplete, outdated, or incorrect, so the cited source still needs to be checked when accuracy matters. OpenAI search guidance

Picture yourself about to forward the summary to a colleague or classmate before leaving home.

The AI says the change starts next Monday, but the original notice gives a later date.

You have not repeated the whole reading task.

You have checked the fact that could have changed someone’s next move.

For factual claims, that is the useful habit: follow the claim with consequences back to the original source.

2. If the result has to work, try a small case whose answer you already know

Now imagine AI helps you build a spreadsheet for a neighbourhood event, a small online shop, or a team record in Singapore.

Here, finding a source is not enough.

The formula itself needs to behave correctly.

NIST’s Generative AI Profile recommends checking AI output with known correct information and using testing methods instead of trusting a result because it looks convincing. NIST Generative AI Profile

You can turn that principle into a small desk-level test.

Suppose the formula should add a fee only when a value is above 100.

Before you copy the formula down a long sheet, try three made-up values:

Input	Expected result
80	No fee
100	No fee
120	Fee added

You already know what all three answers should be.

If one result is wrong, you can stop while the mistake is still small.

The same pattern works elsewhere.

If AI calculates a percentage, calculate it again from the original numbers. If it writes code that will rename product images, try copies of a few unimportant files before touching the real folder.

The aim is not to do the entire job twice.

It is to create one safe place where a failure can appear before the output reaches real work.

3. If AI recommends a choice, put the comparison reasons on the table

A recommendation needs a different kind of check because the word “better” depends on what matters in your situation.

Suppose you are preparing an internship application in Singapore and ask AI which of two project examples should come first.

The AI chooses the cleaner-looking project because it seems easier to explain.

That may be reasonable, but perhaps you care more about showing how you solved a real problem, what changed after your work, and whether you can explain your decisions clearly.

Here, the useful check is to make the comparison rules visible.

That simply means writing down the reasons used to judge one option against the other.

What matters for this choice	Project A	Project B
Can I explain the problem clearly?	Easier	Harder
Can I show what changed after my work?	Less clearly	More clearly
Can I explain my main decisions?	Partly	More fully

Now the recommendation has something you can inspect.

You can see why one option wins and decide whether those reasons fit what you actually want to show.

This is where NIST’s warning about automation bias becomes practical. A recommendation can feel stronger simply because an automated system produced it, even when the reasons underneath it are weak or do not fit your situation. NIST Generative AI Profile

If the AI still says, “Choose Project A,” you do not need to ask it to repeat the answer more confidently.

Put the reasons beside both projects and look again.

The choice may stay the same or change.

Either way, you can now see what the judgement rests on instead of borrowing the AI’s conclusion as if it were self-proving.

One AI answer can contain more than one kind of risk

These three checks belong together because they answer three different failure questions.

A factual claim can be true or false, so we trace it to the original source. A formula, calculation, code result, or process can work or fail, so we test it on something small.

A recommendation can depend on reasons that fit your situation or reasons that do not, so we bring those reasons into view.

One answer can contain all three at once.

A project plan, for example, might contain a date from a notice, a spreadsheet calculation, and a recommendation about which task to start first.

You do not need to label the whole AI answer “trusted” or “untrusted.”

Separate the parts, then use the check that fits each one.

That keeps the speed that makes AI useful without turning polished language into automatic proof.

Move closer to the person you want to become

• Let’s pause at the next AI answer and name the part that could actually change our decision.
• Let’s trace it when it is a fact, or test it when it has to work.
• Let’s place the reasons beside the options when AI recommends a choice for us.