Before You Act on an AI Answer, Check the Part That Carries the Risk

The AI answer is ready, someone is waiting for your reply, and you do not want to redo the whole task. Can we check the part that could actually change what happens next?

AI can turn a rough question into a polished answer in seconds.

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 simple language, an AI system can sound certain while part of its answer is incorrect. NIST also warns about over-reliance and automation bias, which means giving too much weight to an automated answer simply because a system produced it. NIST Generative AI Profile

OpenAI’s current ChatGPT guidance gives similar advice. 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 inspect every line in the same way.

Let’s begin with one question: What kind of result am I about to depend on?

Three checks for three different risks

• Let’s go back to the source when a fact could change our plan.
• Let’s try a small example when the answer has to work in practice.
• Let’s place the reasons side by side when AI recommends a choice.

1. If a fact could change your next move, trace it to the source

Imagine a notice from your training centre or workplace is forwarded into a WhatsApp group.

You ask AI for a quick explanation because the message is long and you mainly need to know whether Saturday’s session has changed and who the update applies to.

The AI may give you several clear paragraphs.

But perhaps only two details will affect what you do next:

• the date or time of the change;
• the people covered by the notice.

Those are the claims carrying the consequence.

To trace a factual claim means going back to the document or organisation that issued it, instead of checking one summary against another summary.

Open the original notice.

Find the date or time yourself, then find the line that says who is affected.

If the AI gives you a citation, open that source too. OpenAI’s current search guidance says search results and citations can be incomplete, outdated, or incorrect, so the cited source still needs to be checked when accuracy matters.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 tell a friend or colleague that the session has moved.

The AI summary says the change starts this Saturday, but the original notice gives a different date.

You have not repeated the whole reading task.

You have checked the fact that could have changed another person’s plan.

For factual claims, that is the useful habit: trace the detail that carries the consequence back to the original source.

2. If the result has to work, test it on a small case first

Now imagine AI helps you build a spreadsheet for a small chama contribution record, a neighbourhood activity, or a team expense list.

Here, knowing where a statement came from is not enough.

The formula itself has to work.

NIST’s Generative AI Profile recommends checking AI output against known correct information and using test methods rather than trusting a result only because it looks convincing. NIST Generative AI Profile

You can turn that principle into a small test before the sheet grows.

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

Try three made-up values first:

Trial value	What should happen
80	Leave the cell unchanged
100	Leave the cell unchanged
120	Add the mark

Because you already know the expected result, the three rows become a quick check rather than another full task.

If the 100 row receives a mark, you have found a problem before spreading the formula across the record.

The same pattern works elsewhere.

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

The point is not to do the entire job twice.

It is to create one small, safe place where a mistake can appear before it reaches work that matters.

3. If AI recommends a choice, make the comparison reasons visible

A recommendation needs another kind of check because “better” depends on what matters in your situation.

Imagine you are preparing an internship or entry-level job application in Nairobi and ask AI whether a simple sales-tracking project or a customer-feedback project should appear first.

The AI chooses the sales-tracking project because its final screen looks cleaner.

That reason might matter, but perhaps you want the first project to show how you understood a problem, made a trade-off, and checked whether your work helped.

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

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

What you want the first project to prove	Sales-tracking project	Feedback project
Shows the original problem clearly	Somewhat	Clearly
Gives you a decision or trade-off to explain	One	Several
Lets you describe how you checked the result	Limited	Stronger

Now the recommendation has something you can inspect.

The question is no longer which project looks more polished, but which one better proves the kind of judgement you want another person to see.

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 do not fit your situation. NIST Generative AI Profile

If the AI still chooses the sales-tracking project, you can compare that choice with the reasons on the table instead of asking for a more confident answer.

The same recommendation may survive the check, or the order may change once your real priorities are visible.

Either outcome gives you something better than borrowed certainty.

You can see what your choice rests on.

The answer does not need one giant trust label

The three checks belong together because each is aimed at a different possible failure.

Facts are about whether something is true or false, so the original source matters. Formulas, calculations, code, and processes are about whether something works or fails, so a small test matters.

Recommendations are about whether the reasons fit the decision you are making, so those reasons need to be visible.

A single AI response can mix all three.

For example, a work plan may include the date of a meeting, a calculation in a sheet, and advice about which task deserves priority.

Instead of deciding whether the entire answer is trustworthy, separate the parts.

Then give each part the kind of check that can actually reveal its weakness.

That preserves the speed of AI without asking polished language to do the job of evidence.

Move closer to the person you want to become

• Let’s notice the one part of the next AI answer that could alter what we decide.
• Let’s return to the source for a fact, or run a small check for something that must work.
• Let’s write the comparison reasons down when AI is choosing between options for us.