Before You Take Advice From AI, Ask It These Five Questions
Keep up with IT Club
Add IT Club as a preferred source in Google Search.
AI is useful for explanations, brainstorming and drafting, but confident language is not proof. Five questions help everyday users and small UK businesses check context, freshness, sources, assumptions, consequences and privacy.
Imagine asking an AI assistant: “My partner did this. Am I being unreasonable?” It may explain what could be going on, suggest different interpretations and even draft a message to send. That can be genuinely useful when you are trying to organise your thoughts.
But the assistant has heard one person's version of events: yours. It does not know what you have left out, what happened before, or how the other person would describe it. A calm, polished answer can feel like a verdict even though it is working from an incomplete account.
That is the wider lesson behind the relationship-advice story that prompted this article. The same gap can matter when you ask about health, money, work, contracts, purchases, technology, security or current events. A good-sounding answer is not necessarily a correct one.
Why AI can sound so convincing
ChatGPT, Copilot, Gemini and similar assistants are very good at producing fluent language. They use patterns in their training and the conversation to generate a response; they are not a separate fact-checking service that certifies each sentence before it appears. Some assistants can search the web or use other tools, which may help with current information, but those tools do not guarantee that every claim is right.
The US National Institute of Standards and Technology calls one risk “confabulation”: confidently stated content that is erroneous or false. OpenAI also warns that ChatGPT can sound confident when it is wrong and may produce fabricated facts or citations. The important point is not that every answer is unreliable. It is that confidence and accuracy are different things.
The way you phrase a question matters too. If you write, “My broadband company is ripping me off because…”, you have already supplied a conclusion. The assistant may follow that framing instead of testing it. Research has also found examples of AI systems agreeing too readily with a user's stated view, a behaviour researchers call sycophancy. These studies examine particular systems and tasks; they do not mean every assistant always agrees with everyone.
Try a more open question instead: “Help me work out objectively whether my broadband deal is good value. What information do you need, and what might challenge my first impression?” The answer can still be wrong, but you have invited it to examine the issue rather than simply reinforce your opening claim.
Five questions to ask before trusting an AI answer
1. Does it have enough information?
Have you given it the details that could change the answer? For a relationship question, that might include what was said, what you want to happen and what you are unsure about. For a business problem, it could be your contract terms, software version, budget or country. If important facts are missing, ask the assistant to list them before it recommends what to do.
2. Is this information current?
Prices, laws, product features, software versions, security threats and news can change. Ask what date the answer applies to and whether the assistant checked live sources. If it has not, verify the detail on an official, up-to-date website. Even when an AI provides web links, open them yourself and check that they support the claim.
3. Where did this come from?
Ask for links to the original evidence, not just a list of website names. Open the sources and look for the specific detail being claimed: a headline or summary may leave out an important qualification. AI systems can also produce convincing-looking references that do not exist, so a citation is a lead to check, not a stamp of approval.
4. What would change the answer?
This question helps expose gaps. Ask what information might change the conclusion, which assumptions it has made, and what a reasonable person who disagrees would say. You can also ask it to separate what is supported by the information you gave from what it is inferring. Treat the result as a way to find questions—not as proof that every side has been covered.
5. Does this matter enough to verify?
Use common sense about the cost of being wrong. A dinner idea is usually low consequence. Deleting business data, changing a security setting, signing a contract, moving a pension, making a medical decision or ending an important relationship based on one chat is not. For a major relationship decision, speak with the other person if it is safe, or a trusted professional. AI can help you understand an issue and prepare questions, but important decisions deserve an independent check.
Useful prompts to copy and paste
You can ask an assistant to test its answer instead of accepting its first response. These prompts are useful starting points:
What information might I be missing that could change your conclusion? Ask me questions before advising me.Separate what you know from what you are assuming. Give me the strongest reasonable argument against your recommendation.Which claims here need current information? Give me links to original sources and say what you could not verify.What could go wrong if I follow this advice, and how should I verify the important parts independently?A prompt can improve the conversation, but it cannot make an assistant impartial or infallible. Asking the same assistant to prove its own answer is not an independent check; open original sources or ask an appropriate person. Verify for yourself when the answer could affect someone's wellbeing, rights, money, work or security.
A useful second brain—not your only brain
AI is particularly handy for explaining a complicated topic in plain English, brainstorming, summarising material you are allowed to share, drafting a message, comparing options, preparing questions and helping you work through a technical problem. It can save time and give you a first structure when you are not sure where to begin.
For higher-consequence subjects—medical, legal, financial, cybersecurity, employment or major relationship decisions—use it to help you think and prepare, not to make the decision for you. Check the underlying evidence and involve the right person when the stakes call for it.
Privacy matters too
Before pasting anything personal or commercially sensitive, understand which AI service and account you are using, what its current data settings say, and what your workplace allows. Different services and account types can handle conversation history, retention and model improvement differently; do not assume one tool's settings apply to another.
- Do not enter passwords or banking details.
- Avoid identifiable customer data, medical records and confidential company documents unless the tool and use have been approved.
- When you only need help with the wording or structure, remove names and details or use a made-up example.
For a small business, a personal account may not be suitable for work information. Check the organisation's policy and the service's current terms before using real data. The Information Commissioner's Office guidance on AI and data protection is a useful UK starting point, but its page notes that the guidance is under review and may change.
The IT Club rule
USE AI TO THINK BETTER — NOT TO STOP THINKING.
The most useful question is not “Tell me what to think.” It is “Help me think this through.” AI can explain, challenge, draft and save time. When something genuinely matters, check the answer.
You do not need to be suspicious of every response or avoid AI altogether. Use it, ask it questions and get it to challenge you. Just remember that confident and correct are not the same thing.
More from IT Club
AI confidence and human verification for small businesses →
ChatGPT Temporary Chat and sensitive business information →
AI privacy and business data controls →
How a trusted Google link can still lead to a phishing page →
How to prioritise cybersecurity risks →
Stay connected
For more straightforward technology advice, follow the free IT Club WhatsApp Channel.
Follow the IT Club WhatsApp Channel →
Found this useful? Forward it to someone who might be relying on an AI answer. If you would like to support IT Club, use the on-page “Buy us a coffee” button and choose an Espresso.
Sources and further reading
The relationship-advice story is an example, not evidence that every AI answer behaves the same way. The factual claims about AI limitations, sycophancy and data protection were checked against the sources below.
SuperAging News — Why more people are turning to AI first for relationship advice →
The Conversation — original relationship-advice article by Maha Khawaja →
NIST — Artificial Intelligence Risk Management Framework: Generative AI Profile →
OpenAI Help Center — ChatGPT accuracy and limitations →
Anthropic — Towards Understanding Sycophancy in Language Models →
Association for Computational Linguistics — LLM sycophancy under user rebuttal (2025) →
Information Commissioner's Office — Guidance on AI and data protection (under review) →
Plain-English Takeaway
Use AI to think better, not to stop thinking. When the answer matters, check the evidence before you act.
Frequently asked questions
Can you trust ChatGPT?
You can use ChatGPT as a helpful assistant, but do not treat a fluent or confident answer as proof. Check important claims against reliable, original sources, especially when the consequences of an error are serious.
Can I ask AI to challenge my assumptions?
Yes. Ask it for missing information, alternative explanations and the strongest argument against its recommendation. This can improve your thinking, but it does not guarantee that the response is balanced or correct.
Should I use AI for medical, legal or financial advice?
AI can help explain unfamiliar terms or prepare questions for a professional. It should not replace an appropriate healthcare, legal or financial professional when a decision could materially affect your health, rights or money.
Is it safe to paste confidential information into an AI tool?
It depends on the service, account, settings and your organisation's rules. Do not paste passwords, banking details, identifiable customer information, medical records or confidential business material unless you understand the service's data handling and have permission to use it.
Related Articles
AI can sound certain and still be wrong. That's the bit businesses need to remember.
A reported military intelligence near-miss shows why confident AI output is not evidence. For an SME, the practical answer is proportional human verification.
Read articleYour watch is becoming an AI meeting assistant. Does everyone else know?
Apple's new Audio Intelligence features show why workplace rules need to cover what a wearable can hear, process and summarise, not just what it looks like.
Read articleYour AI Did Exactly What You Asked. Just Not How You Expected.
If an AI agent has an objective, credentials and access to real systems, businesses should not assume it will stay within the boundaries they had in mind.
Read article