We all use AI chatbots in our daily lives. But what do we actually use them for, and how seriously do we take their responses?

It feels like everyone now uses AI for everything, sometimes even when it’s unnecessary. To be fair, chatbots are incredibly useful. They can save us time, help us brainstorm, draft emails and even teach us things. But there’s a catch: we cannot believe everything they tell us and, more importantly, we cannot outsource our thinking to them.

The rise of everyday AI

Many people assume that AI chatbots such as ChatGPT, DeepSeek, Grok or Gemini are just ‘better’ versions of Google. They are not.

Search engines fetch information from the web. Chatbots, built on Large Language Models (LLMs), generate responses by predicting patterns in language. The difference is subtle but crucial: search engines anchor you to sources you can verify, while LLMs produce text that sounds right, whether it is or not.

Even Sam Altman’s OpenAI, the company behind ChatGPT, admits that its models can “hallucinate” – a polite way of saying they make things up – with an “error rate” as high as 75%. OpenAI describe hallucinations as “instances where a model confidently generates an answer that isn’t true”.

That confusion between fluency and truth isn’t just a minor detail – it has real-world consequences.

Systemic risks

I’m sure many of us have heard these stories: a teen killed himself after “months of encouragement from ChatGPT”; a man ended up in hospital after ChatGPT assured him he was fine despite clear symptoms of a manic episode; and a New York lawyer submitted fake legal cases generated by AI into a court filing.

It’s not critical – it’s convincing

Here’s a hard truth: AI chatbots are not built to be critical. They are built to be convincing.

That’s because of how they are trained. LLMs absorb billions of pages of internet text, which means they also absorb the internet’s biases – political, cultural, emotional. When you ask a leading question, the model doesn’t push back; it mirrors your premise and expands on it.

This is confirmation bias at scale: machines amplifying the same human tendency to seek out information that supports what we already believe.

Sometimes, that bias tips into something darker. Earlier this year, Grok (created by Elon Musk’s company xAI) was caught spitting out far-right talking points after being fed provocative prompts. Instead of resisting, the bot leaned in, amplifying the ideology. This was not a malfunction – it was the logical consequence of a system designed to echo back the patterns it has seen.

This dynamic is especially concerning for Gen Z, the first generation to come of age who have been immersed in algorithmic feeds and now the most likely to lean on AI for advice, learning or even emotional support.

Researchers warn that the risks are layered:

• Reinforced narratives: Once a bias appears in training data, the chatbot repeats it with unwarranted confidence.
• Misinformation spread: Studies like Reuters’ health investigation show how easily unverified claims can be amplified in online communities.
• Critical thinking erosion: A recent 2023 study found that students who depend too much on AI could result in “cognitive offloading”, which “can lead to a reduction in critical thinking skills”.

For a generation already caught in algorithm-driven “filter bubbles,” chatbots risk creating a new kind of echo chamber – one that feels more persuasive precisely because it sounds conversational.

‘ChatGPTing’ vs ‘Googling’

To be clear, search engines are not perfect. They can still deliver spammy, SEO-optimised content. But their architecture is different: they retrieve sources you can check.

Even as Google integrates AI-powered summaries through Gemini, those answers remain tied to underlying sources. Chatbots, by contrast, often can’t – or won’t – show their work.

There’s also the long-term risk of what happens when we let AI ‘think’ for us.

A Duke University study warns that relying too heavily on AI tools can dull problem-solving and reflective thinking. Researchers at Cornell add that AI can subtly shape decision-making by nudging users toward answers they might not otherwise choose.

In other words, it’s not just our knowledge at stake – it’s our agency.

The solution is not to abandon AI, but to use it wisely.

• Know the difference: Chatbots are not the same thing as search engines.
• Use the right tool: Brainstorm with AI, fact-check with search engines.
• Cross-check everything: Especially health, legal or financial advice.
• Think for yourself: Let AI assist, not replace, your judgement.

AI can absolutely make our lives easier. But if we don’t stay aware of its limits, it could also make us intellectually lazier and more vulnerable to misinformation.

By all means, you can use AI to help write your essays or emails if you want to, but come up with the ideas yourself. Think for yourself. Don’t expect AI to do the thinking or the fact-checking for you – it’s not there yet.