How to Fact-Check AI Answers (Because It Gets Things Wrong)
AI is brilliant, but it sometimes makes things up, and does it with total confidence. Here is exactly how to spot it, step by step, no technical knowledge required.
Let’s get something important out of the way: AI is incredibly useful. It can write, explain, brainstorm, and summarise better than most tools you’ve ever used. But it has one rather significant flaw: it sometimes makes things up. And it does it with absolute confidence.
If you’re going to use AI (and you should), you need to know how to spot when it’s wrong. This guide will show you exactly how to do that, step by step, no technical knowledge required.
Why Does AI Get Things Wrong?
Before we get into the “how to check” part, it helps to understand why AI makes mistakes. Don’t worry, we’ll keep this simple.
AI tools like ChatGPT and Claude don’t actually “know” things the way you and I do. They don’t have a database of facts they look up. Instead, they predict what words should come next based on patterns in the text they were trained on.
Think of it like the world’s most sophisticated autocomplete. When you type “the capital of France is…” your phone might suggest “Paris.” It doesn’t know geography; it’s just seen that pattern so many times that it predicts the right answer.
Most of the time, this works brilliantly. But sometimes, the prediction goes wrong. AI might:
- Invent facts that sound plausible. It might cite a study that doesn’t exist, name an author who didn’t write that book, or give you a statistic it made up. The AI community calls these hallucinations: the AI isn’t lying on purpose, it’s just generating text that sounds right.
- Mix up details. It might combine facts from two different events, get dates slightly wrong, or attribute a quote to the wrong person.
- Present outdated information as current. AI models have a training cutoff date. They might give you information that was true last year but has since changed.
- Be confidently wrong. This is the tricky bit. AI doesn’t say “I’m not sure about this.” It states wrong information with the same confidence as right information. There’s no hesitation, no uncertainty, just smooth, assured text that happens to be incorrect.
Robi saysHere’s my favourite way to think about it: AI is like a very confident friend who’s read a lot but sometimes misremembers the details. You’d still check before putting their advice in a work presentation, right? Same principle.
Red Flags: When to Be Extra Careful
Not everything AI says needs rigorous fact-checking. If you ask it to brainstorm blog ideas or help you reword a sentence, accuracy isn’t really the concern. But certain situations should trigger your fact-checking instincts.
Check carefully when AI gives you:
- Specific numbers or statistics. “According to a 2024 study, 73% of UK small businesses…” That number might be completely made up.
- Named sources or citations. “As reported by the BBC…” or “A study published in The Lancet found…” AI frequently invents citations. The source might not exist.
- Historical dates or events. It might get the year wrong, confuse similar events, or merge details from different incidents.
- Legal or regulatory information. Laws vary by country and change regularly. AI might give you US law when you need UK law, or cite regulations that have been updated.
- Medical or health information. AI is NOT a doctor. It can give dangerous advice if taken as gospel. Always verify with the NHS website or your GP.
- Current events or recent news. If the AI’s training data doesn’t cover the period in question, it might guess, or simply tell you something outdated.
- Quotes. AI loves to put words in people’s mouths. That “famous quote” by Einstein or Churchill? There’s a decent chance they never said it.
- Recommendations of specific products, businesses, or websites. It might recommend a business that’s closed, a product that’s been recalled, or a website that doesn’t exist.
Your 5-Step Fact-Checking Process
Here’s a simple process you can follow every time you need to verify something AI has told you. It takes a few minutes and can save you from embarrassment, or worse.
1Ask Yourself, Does This Matter?
Not everything needs checking. If AI helped you brainstorm a catchy headline, you don’t need to verify it. But if you’re about to put a statistic in a client proposal, quote a law on your website, or share health information, check it.
Quick filter: “Would I look bad if this turned out to be wrong?” If yes, check it.
2Google the Specific Claim
Take the exact claim AI made and search for it on Google.
“The UK small business sector contributes £2.3 trillion to the economy.”
“UK small business contribution to economy”
Look for results from official or reputable sources: government websites (.gov.uk), established news outlets (BBC, Guardian, FT), industry bodies, or academic institutions. If you can’t find the claim backed up by a credible source, be suspicious.
3Check the Original Source
If AI mentions a specific study, report, or article, try to find the original. Search for the exact title, author, and publication.
If you can’t find it, it probably doesn’t exist. This is one of AI’s most common tricks: fabricating realistic-sounding sources. A paper titled “Digital Transformation in UK SMEs: A 2024 Analysis” published in the “Journal of Business Innovation” sounds real. But if Google can’t find it, AI invented it.
4Cross-Reference With Trusted Sources
For different types of information, go to the most reliable source:
| Information type | Trusted source |
|---|---|
| UK statistics | ONS (ons.gov.uk), gov.uk |
| Health information | NHS (nhs.uk) |
| Legal information | gov.uk, Citizens Advice |
| Company information | Companies House |
| News events | BBC News, established newspapers |
| Scientific claims | Published journals, university websites |
| Financial data | Bank of England, FCA |
| Product information | Manufacturer’s official website |
5Ask AI to Check Itself
This is a surprisingly effective trick. If you’re unsure about something AI told you, ask it directly:
“Are you confident that statistic is accurate? Can you tell me exactly where it comes from?”
“I want to verify the claim you made about [X]. Is this something you’re certain about, or could you be generating a plausible-sounding answer?”
“That quote you attributed to Churchill, are you sure he said that? Misattributed quotes are common.”
AI will often admit uncertainty when directly challenged. It might say “I should note that I’m not entirely certain about that figure” or “I may have generated that citation, please verify it independently.”
This doesn’t replace proper fact-checking, but it’s a useful quick filter.
Robi saysI know what you’re thinking: “If AI can get things wrong, should I even use it?” Absolutely yes! You just need to treat it like a very helpful first draft, not the final word. Check the important bits, and you’ll be grand.
Real Examples of AI Getting It Wrong
These are common types of errors to watch for:
The invented statistic
“92% of consumers prefer personalised marketing emails.”
This specific number doesn’t trace back to any real study. AI generated a plausible percentage.
The fake citation
“According to a 2023 report by Deloitte titled ‘The AI Advantage for SMEs’…”
No such report exists. The title sounds real, Deloitte is a real company, but the report is fabricated.
The misattributed quote
“As Albert Einstein once said, ‘The definition of insanity is doing the same thing over and over and expecting different results.'”
There’s no evidence Einstein ever said this. It’s one of the internet’s most commonly misattributed quotes.
The outdated information
“The UK corporation tax rate is 19%.”
It was 19%, but it increased to 25% for larger profits in April 2023. AI might have older training data.
The confident hallucination
“The village of Thistlebrook in the Cotswolds is known for its annual cheese-rolling competition.”
Thistlebrook doesn’t exist. AI combined real elements (Cotswolds, cheese-rolling) into a convincing but fictional answer.
Building Good Habits
Fact-checking doesn’t need to be time-consuming. Once you get into the habit, it becomes second nature, like checking your mirrors when driving.
Daily habits:
- Quick scan everything AI writes before you use it. Do the facts feel right?
- Google any specific claims before sharing them publicly.
- Never copy-paste AI medical, legal, or financial advice without professional verification.
- Keep a healthy scepticism. If something sounds too perfect, too specific, or too convenient, check it.
Key Takeaway
Remember: AI is a brilliant first draft tool, an excellent brainstorming partner, and a fantastic time-saver. It’s just not a replacement for your own judgement and a quick Google search.
Knowing how to fact-check AI isn’t about being paranoid or distrustful. It’s about being smart. The people who get the most value from AI are the ones who know its strengths and its weaknesses. You now know both. Use AI freely, check the important stuff, and you’ll be in a better position than most. That’s not a bad place to be.
Put AI to work, safely. Our free prompt pack includes 20 prompts designed to get clear, useful answers, and some of them include built-in instructions that encourage AI to flag when it’s uncertain. For more prompts across every category, check out the prompt shop: packs from £4.99 to £14.99, all designed for real-world use.