Photo: SHVETS production / Pexels
Ask an AI assistant whether your landlord can raise your rent this year, and you’ll probably get a clear, well-organized answer in seconds. It may even include a link.
It reads like it came from someone who knows.
But it might describe the rules in another province. Or rules that changed last year. Or the link might not say what the answer claims it does.
AI tools are useful for getting oriented quickly. They are less reliable as the final word.
The skill that matters is knowing when to check, and how.
Here’s the version that fits in your head.
Why can an AI answer sound right and still be wrong?
Most AI chatbots are built on large language models. They generate responses based on patterns learned from enormous amounts of text.
That makes them very good at producing fluent, plausible language.
It does not guarantee that what they say is true.
When a model produces something false but convincing, it is often called a hallucination. The Government of Canada’s Guide on the use of generative AI warns that these tools can produce content that appears “well developed, credible and reasonable” while actually being inaccurate. They can also invent sources that do not exist.
The important part is that a wrong answer usually does not look wrong.
It can have the same tone, structure and confidence as a correct one.
How sure an AI sounds tells you nothing about whether it’s right.
A link is not a source check
Many AI tools can search the web and show links alongside their answers.
That is useful. It gives you somewhere to check.
But the presence of a link does not prove that the answer is correct.
Open it, and you might find that:
The page supports the claim.
The page is real but says something different.
It supports only part of what the AI said.
It is out of date or applies somewhere else.
It is a weak source, such as a sales page, forum post or another summary.
There are really two questions to ask: is this a trustworthy source, and does it actually support the claim the AI made?
A credible source can still be the wrong source for a particular claim.
The link moves you from “trust me” to “here’s where to look.”
You still have to look.
Five checks before you rely on an AI answer
You do not need to verify every answer an AI gives you.
You should verify the ones you are going to rely on.
1. Open the source
If the answer provides a link, open it.
If there is no source, ask for one or look up the important claim yourself.
An answer with no traceable source can still be useful for orientation. It should not automatically be treated as a conclusion.
2. Find the actual support
Do not stop at the page title.
Find the sentence, table, rule or number that supports the claim.
Check that it actually says what the AI told you. Look for missing qualifications or context.
This matters especially with statistics and studies. A report can be real while an AI misstates one of its findings.
3. Check the place and the date
A correct answer for one place can be wrong somewhere else.
Much of the information online is American. Rules involving tenancy, employment, taxes, health care and consumer rights can differ by country and by province.
Ask: does this apply where I live? Is it still current?
That second question matters for more than laws. Prices, government programs, product features, deadlines and office-holders can all change.
4. Go to the primary source when it matters
For important claims, try to reach the source closest to the information.
For a government rule, check the government.
For a law, regulator or court decision, check the official text.
For research, look for the original study.
For a company policy, check the company’s own policy.
A news article explaining an official source is one step removed. An AI summarizing that article is another.
The higher the stakes, the closer you should get to the original information.
5. Match the checking to the stakes
Not every mistake matters equally.
A restaurant suggestion can be wrong without much harm.
A tax deadline, medication question, legal right, financial decision or contract term is different.
Slow down especially when an answer contains exact numbers, dates, quotes, citations or rules that affect something you are going to do.
The greater the cost of being wrong, the stronger the verification should be.
Sometimes that means checking the original source.
Sometimes it means asking a qualified person instead of relying on the AI at all.
Make the answer easier to verify
You can also ask the AI to structure its answer so checking is easier.
Useful prompts include:
“Give me the sources for the main claims in this answer.”
“Show me which source supports each important claim.”
“Does this apply in Ontario, and as of what date?”
“Which claims here should I verify independently?”
“Point me to the primary source rather than a summary.”
These prompts can make verification faster.
They do not perform the verification for you.
An AI can still misdescribe a source, attach the wrong source to a claim or reproduce a passage incorrectly.
The final step is still opening the source yourself.
What responsibility looks like in Canada
Two B.C. cases illustrate an important distinction.
In Moffatt v. Air Canada, a customer received incorrect information from a chatbot on the airline’s website about bereavement fares. The B.C. Civil Resolution Tribunal held Air Canada responsible for the information it provided through the chatbot, just as it was responsible for information elsewhere on its website.
The decision did not establish that the chatbot was a generative-AI system. Its relevance is simpler: using an automated tool did not transfer responsibility away from the organization providing the information.
A separate 2024 B.C. Supreme Court case involved generative AI directly.
In Zhang v. Chen, 2024 BCSC 285, a lawyer submitted two court cases that did not exist after ChatGPT generated them. Opposing lawyers spent time and money trying to find the cases. The court ordered the lawyer to personally bear the additional costs caused by the mistake.
Justice David Masuhara concluded that generative AI is no substitute for the professional expertise required of lawyers.
The situations are different, but the accountability principle is similar.
Responsibility stays with people and organizations. It does not transfer to the tool.
Our read
AI is good at the first part of research. It can explain an unfamiliar topic, summarize information and point you toward sources.
The risk is treating a fluent answer as a finished one.
The habit worth building is not distrust of AI. It is source discipline: know which claim matters, find where it came from, check that the source actually supports it, and make sure it applies to the right place and time.
That is not a new skill invented for AI.
Researchers, journalists, analysts and public servants have always had to separate what a source actually says from what someone says about it.
AI simply makes that distinction easier to forget because the answer arrives so quickly and sounds so complete.
AI can shorten the path to an answer. It cannot take responsibility for it.
That part stays with you.
What is AI literacy, and why does Canada suddenly care?
A plain-language guide to the skill Canada has made a national priority.
AI at work: what to learn first
The practical habits that help people use AI tools effectively, check their work and protect sensitive information.
AI agents and agentic AI, explained
What changes when AI moves from answering questions to taking action, and why checking matters even more.
