Tại sao chúng ta tin tưởng chatbot và làm thế nào để sử dụng chúng khôn ngoan hơn? Một chuyên gia tâm lý giải thích

Sometime this week, you probably did it: A question came up, and instead of searching for the answer online or reaching for a book, you asked a chatbot for it.

If this sounds familiar, you’re not alone. By February this year, ChatGPT alone had 900 million weekly users, and a large study found that people mostly use it for searches under the broad topics of “seeking information,” “practical guidance” and “writing.”

As a researcher with a PhD in developmental psychology, studying how people learn and stay informed in digital environments, I believe there is a lot we don’t yet understand about what this means for our societies.

This is a shift in how we meet information, how we trust it and how we learn.

One fluent answer

In the past, finding information involved consulting various sources like books, newspapers, encyclopedias and people. These sources often disagreed, and assembling an answer required effort. Our judgment was formed in the gaps between them.

Then, search engines emerged. For about two decades, they provided ranked arguments for questions. While they offered shortcuts, they still showed that answers had rivals. Even if you chose the first option, you could still explore the alternatives.

AI chatbots changed this dynamic. Now the chatbot gives you one paragraph, sounding fluent and settled.

Do you click through?

The Pew Research Center tracked real browsing behaviour and found that when Google shows an AI summary above the results, people clicked on the actual source roughly half as often — eight per cent of visits instead of 15 per cent.

Even more strikingly, people clicked on the links inside the AI summary only one per cent of the time. Often, if the question is answered, the sources go unchecked.

And yet research shows that if we read conflicting information, it enhances our recall of the sources, compared with consuming information that simply agrees.

The Reuters Institute’s Digital News Report for 2026 finds 10 per cent of people worldwide now use an AI chatbot for news every week, where this rate increases to 16 per cent for individuals under 35. And only 42 per cent of those users report they always, or often, click through the links to check the actual sources.

Easy answers feel true

So why do we trust AI chatbots without fact–checking? Decades of research on processing fluency show that the easier something is to take in, the truer it feels. Pre-existing biases, repeated exposure and ease of processing can all play a role.

Even statements printed in colours that are simply easier to read are rated as more accurate. A chatbot’s answer is instant, clean, confident and effortless, pressing on our psychological tendencies to accept its claim as truth.

Additionally, the convenience of getting a single answer or solution, especially from a source that is perceived as fluent and authoritative, may be another important reason we trust and use AI chatbots. Disagreement between sources is uncomfortable.

In fact, in certain experiments, when people faced different types of answers, they were overly confident in the truthfulness of single answers and under–confident in that of plural answers, even when the plural answers were more accurate.

When 10 sources become one paragraph, the disagreement does not disappear from the world, but it disappears from view. Our certainty rises, not because the answer got better, but because the argument got hidden.

Machines seem objective

A third mechanism could also explain this. Scientists call it the machine heuristic — our tendency to assume machines are objective with no agenda of their own.

In an era when trust in certain institutions, in particular news organizations, is low, partly over perceived agendas, an answer with no visible author slips past defences that a byline might trigger.

None of this would matter much if the answers were reliable. But they are not. One of the largest assessments of its kind, co-ordinated by the European Broadcasting Union and led by the BBC across 18 countries and 14 languages, found that almost half of AI assistants’ news answers contained at least one significant problem.

Sourcing was the most common error, while significant portions contained misleading, missing, outdated details along with information the system simply invented.

Pair this with the small share of people who check sources, and you end up with a misinformed audience, who may feel they are highly informed without realizing their information is wrong.

In a democracy, the cost is not just personal. A misinformed audience, is also a misinformed electorate. Therefore, having access to accurate information is a necessity to make sure we are making correct decisions about our leaders, and our society as a whole.

Add one deliberate habit

So what’s the solution? AI tools can be extraordinary, and telling people to stop using one of the most convenient information technologies ever created will work about as well as forcing them to give up their cars and ride horses instead. The realistic goal is to add one deliberate habit to an automatic one.

As the first step, we can become aware of our psychological tendencies to accept certain claims at face value. This can then help us become more aware of the pitfalls, so we can avoid them.

Second, we need to treat a chatbot answer as a lead, not the end point. When a chatbot responds, this is where your understanding should start, not where it ends. Ask for sources and make sure to check them. Examine the subject from different angles and from opposing sources.

Most of the time, the truth is in what we assemble from different angles, not in the single line a chatbot hands to us.

Ali Jasemi does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.