A few weeks ago, I intentionally started using ChatGPT to follow the latest news about the Iran war. This was also a test to see how chatbots perform compared to traditional news sites in providing real-time information. Another reason was the overwhelming pace of news at the time.
But at some point I realized I wasn’t clicking to confirm anything. I was just absorbing what ChatGPT was teaching me.
Article continues below
Tracking changes
Using AI for search isn’t always a good idea. Not so long ago, ChatGPT did not have access to real-time information. Google’s AI overview recommended adding glue to pizza to help the cheese stick and suggested eating a stone a day. The problems with relying on AI for real-time, accurate information were obvious and easy to spot.
However, many things have improved over the past year. Models are more accurate, information is more up-to-date (many chatbots now have real-time internet access), and sources are more likely to be cited.
AI search has recently moved towards what Ofcom calls ‘answer engines’ – tools that don’t just point you to information, but deliver it directly to you in plain conversational language.
This all sounds good, and in many ways it is. I believe that for low-stakes, simple queries like recipes, definitions, travel tips, and buying advice, AI search can really help. Its conversational style also helps you drill down and answer the right follow-up questions to find what you need faster than clicking through a list of links.
However, I think this improvement itself is problematic.
Rebuttal to a better answer
Even when AI search was clearly flawed, many of us remained vigilant. Now that it’s better and more reliable, I fear we’re less likely to question it. And conversational style plays a very important role in that.
We are wired to treat fluent and consistent language as reliable. When something seems confidently explained, it’s much harder to step back and interrogate it, even if you know you should. I’ve written about this same pattern in other areas of AI: therapy, relationships, and health advice. It is much easier to delegate our thinking to the AI tools we use, and we are much less likely to apply our own judgment.
Ellen Scott, who I talked to about this in the context of my work, called it a smooth out. This is a type of cognitive offloading where the effort of evaluating information is absorbed by the AI. It removes the friction that was making you think.
Traditional search wasn’t perfect, but it had built-in friction. Typically, I would scan a list of links, check the source, and make a quick judgment about trustworthiness. It was active even when it felt automatic. AI search replaces all of this with a single, synthetic answer delivered in a conversational (and sometimes flattering) tone. This means that you are just sitting and receiving information instead of evaluating it.
Pew Research found that when AI summaries appear in search results, users are significantly less likely to click through to the original source. This means that AI will effectively answer your questions and reduce the chances of confirming your questions.
remaining failures
Of course, AI search isn’t always completely reliable either.
The illusion of chatbots confidently producing something that is not true has not yet disappeared. Quotations can still be misleading or even broken.
And there’s another problem: I’m a picky eater. While this is something AI companies are actively working on, we know that AI systems are still more likely to agree with you. This is often because these systems are optimized to feel like a good, natural conversation, but don’t necessarily tell the truth.
What makes this even worse is that the increased accuracy makes it harder to spot any remaining errors. I think if a tool is clearly unreliable, we will continue to be more critical of it. But I’m worried that if it’s mostly true, I’ll just stop checking, as I did in my own experiments.
Build better systems and better decisions
The standard answer here is, and I believe, that people need better media literacy for the AI era. Understanding what these systems are doing, treating the output of AI as a starting point rather than a conclusion, and learning to question fluent and confident language are all very important.
But we are most likely to reach for AI search in fast-moving situations, when we need answers to high-stakes questions, or during emotionally overwhelming events—precisely when validation is most important and critical thinking is most difficult.
In a previous report, I spoke with therapists and doctors who noticed the same pattern: patients often turn to AI in moments of crisis or distress, when they are least likely to scrutinize exactly what is being said. That’s why we can’t completely shift the burden onto the users.
If AI tools are to become central to how people find information, their design choices will be critical. That should mean clear attribution, interfaces that encourage review and action when finding more information from other sources, and tools that tell you not just what’s included, but what’s not.
There is no doubt that AI search has gotten better. I think we need to be honest about what actually means better for the way we find, process, and understand information in the long term.
Follow TechRadar on Google News and Add us as a preferred source Get expert news, reviews, and opinions in your feed. Be sure to click the follow button!
Of course you can also do Follow TechRadar on TikTok Check out news, reviews, and unboxings in video format and stay updated regularly. whatsapp Too.
The perfect business laptop for every budget
