Fact-Checking AI Output
By the Syncfusion DevTutorials team · Last updated
Catch AI mistakes before they go out: compare answers with the source line by line, make the AI show where facts came from, check sums and dates yourself, and ask it to mark guesses with [CHECK].
Lakshmi's Quote Summary
Lakshmi works in Accounts at Kestrel. She has received a supplier quote and needs to send a short summary to her manager for review and approval.
She pastes the quote into an AI chat tool and asks for a summary.
At first glance, the summary looks good.
- It is clear.
- It is well written.
- It sounds confident.
But three statements in the summary are wrong.
Lakshmi is about to forward this summary to her manager, who may approve a payment based on it. Can you spot the errors before she does?
In simple words: AI creates the first draft. You check the facts before anything is sent.
TL;DR
- AI can make mistakes even when you provide the correct source material.
- Always check the things that cost money or trust: numbers, dates, names, prices, and promises.
- Compare the answer with the original source line by line.
- Asking "Are you sure?" is not a reliable check.
- Ask the AI to show where each fact came from.
- For calculations, ask for the working and check each step yourself.
- Ask the AI to mark uncertain information using [CHECK].
A Real-World Example
Imagine a new colleague summarises a supplier contract for you. They seem confident, so you forward their summary to Finance without checking the original document.
Two weeks later, Finance pays the wrong amount. The summary said ₹64,800, but the contract said ₹62,800. Nobody compared the summary with the original document.
AI can make the same kind of mistake. Its summary is a draft, not a final answer. If your name is on the email, report, or payment approval, checking the facts is your responsibility.
What You Will Learn
In this lesson, you will learn:
- What to check in every AI-generated answer.
- How to compare answers with source material.
- How to ask AI to show where facts came from.
- How to verify calculations.
- How to check dates and days of the week.
- How to make AI clearly mark uncertain information.
What Should You Check?
The Five Things That Matter Most
You do not need to verify every word.
Focus on the details that can cause real problems if they are wrong.
Check:
- Numbers: totals, quantities, prices, percentages, discounts.
- Dates and times: including deadlines and meeting dates.
- Names: people, suppliers, products, customers, and companies.
- Promises: warranties, guarantees, commitments, and deadlines.
- Extra information: anything that was not present in the original source.
The last one is particularly important because AI sometimes adds details that sound reasonable even though they never appeared in the source material.
Spot the Errors
Here is the supplier quote and the AI summary from the start of the lesson.
Click every line that does not match the original quote and then press Check.
Spot the errors in the summary
There are three mistakes:
- A wrong total
- A changed warranty period
- A free installation service that was never offered
Each error could have cost Kestrel money or damaged trust with the supplier.
Easy way to remember: Check the numbers, dates, names, promises, and anything extra. If it would matter during a disagreement, check it.
Make the AI Show Its Sources
"Are You Sure?" Is Not a Check
Many people respond to an AI answer by asking:
"Are you sure?"
That feels like fact-checking, but in reality the AI often repeats the same answer with the same confidence.
A much better question is:
"Compare each line of your summary with the source and identify anything that does not match."
This forces the AI to explain where the information came from.
The quote and summary are already loaded in the chat below.
First ask:
"Are you sure this is correct?"
Then ask:
"Compare each line of your summary with the quote and tell me which lines do not match."
The first question produced a confident yes. The second question exposed all three mistakes.
Even the comparison is created by AI. Use it to identify where to look, then check those lines yourself against the original source.
Check the Working
Your Calculator Still Matters
Earlier in the course, you learned to ask the AI to show its working.
Now you will use that working to verify the result.
Lakshmi asks the AI to add four amounts step by step.
One of the calculation steps is wrong.
Use a calculator and check every step.
When you find the mistake, tell the AI:
- Which step is incorrect
- What the correct result should be
Notice what happened. When asked to "check again", the AI repeated the same mistake. When you checked the calculation yourself, the error became easy to find.
In simple words: If the number matters, verify it yourself.
Check Days and Dates
The Day Must Match the Date
AI often adds a day of the week to a date.
For example:
"Monday 10 November"
The problem is that the day may be wrong.
Arun asked for a reminder about a stock count scheduled for 10 November 2026.
The AI added:
"Monday 10 November 2026"
But 10 November 2026 is actually a Tuesday.
Send the follow-up below and correct the mistake.
Arun only provided the date. The AI added extra information and got it wrong. If the reminder had been sent, people could have arrived on the wrong day.
Mark What AI Is Not Sure About
The [CHECK] Rule
Sometimes your notes do not contain all the information needed.
When that happens, AI often fills in the gaps.
To make those gaps visible, add this instruction:
"Mark anything that is not in my notes with [CHECK]."
That one sentence changes the behaviour completely. Instead of inventing details quietly, the AI labels them clearly.
Lakshmi's notes below do not include a delivery date.
Send the prompt with the [CHECK] instruction.
Then try it again without the instruction and compare the results.
Without the rule, the AI invented:
- A delivery date
- A delivery time
- A total amount
Those invented details looked just as believable as the real information. With the rule, every assumption was clearly marked.
The [CHECK] labels are intentional. They are not unfinished placeholders. Their job is to highlight information that still needs verification.
Common mistakes
| Mistake | What Happens | How to Fix It |
|---|---|---|
| Trusting a neat summary | Wrong numbers or promises get sent | Compare it with the source line by line |
| Asking "Are you sure?" | The AI repeats the same mistake | Ask it to compare its answer with the source |
| Asking the AI to check calculations again | The same error appears | Verify every step yourself |
| Not checking the day against the date | People arrive on the wrong day | Confirm dates using a calendar |
| Letting AI fill missing information | Invented facts look real | Use the [CHECK] rule |
Exercise
Complete each task below and then press Check.
What to Remember
- AI creates the draft. You verify the facts.
- Check numbers, dates, names, promises, and extra information.
- Compare AI answers with the original source line by line.
- "Are you sure?" is not a reliable fact-check.
- Ask where each fact came from.
- Verify calculations yourself.
- Use [CHECK] to highlight uncertain information before anything is sent.
Key Takeaway
A confident answer is not the same as a correct answer.
AI can help you summarise, rewrite, and organise information, but you remain responsible for checking the facts before they are shared with anyone else.
Next: Data Safety