These are my own thoughts.
Let me start with what the paper found. In “Retrieval Collapses When AI Pollutes the Web,” researchers at NAVER gradually added AI-generated documents to the sources a search system could draw on, then tracked its search results and answers. When synthetic documents designed to rank well made up about 67% of that material, they accounted for about 80% of the top search results. Answer accuracy stayed roughly the same under those conditions. The answers still looked reasonable, even as more of the sources behind them came from AI. Read the paper
That's a controlled experiment, so it doesn't show that the internet as a whole is already flooded with false information. The synthetic material included accurate information too. When the researchers deliberately inserted errors, some search methods were better than others at filtering them out.
Reading it made me think about children growing up in this environment. When you're new to a subject, you may not know enough to spot a factual error. You may not yet know how to check the source, either. I worry that you could accept a false claim as fact, then use it to judge what you learn next.
We were taught to check our sources
Some of us were taught not to trust a Wikipedia entry without checking it, or copy it straight into an assignment. We were supposed to look for other sources and check the evidence behind the claims.
The TV drama The Newsroom put a similar emphasis on facts. Its creator, Aaron Sorkin, argued in an interview that people can disagree and debate, but they need to agree on the underlying facts. Aaron Sorkin's interview
To me, both come back to the same idea: polished writing and a confident delivery don't make a claim true. Even when I agree with an argument, I still need to check what supports it.
Those messages don't make anyone immune to misinformation. But they might leave you with a useful reflex: checking something before accepting it as fact.
Looking things up in print, or opening several web pages ourselves, may have helped build that reflex. At least we had a chance to notice when the sources disagreed.
I wonder how you learn that habit when you grow up with answers neatly summarised for you. The answer alone won't necessarily show which sources were left out, what caveats came with the originals, or how conflicting accounts were combined.
None of this proves that one generation is better at judging information than another. Older people fall for false claims too, and some students check their sources carefully. My concern is about people who haven't yet learned how to check what they read.
The first thing we learn could be wrong
If you already know something about a subject, an odd answer can make you pause. It doesn't fit what you know. You might be the one who's wrong, but at least you have a reason to look again.
If you're new to the subject, you may have nothing to compare it with. False information can become your starting point.
Suppose AI gets something wrong and other articles repeat the mistake. A later search turns up several of those pages, and their agreement can look like confirmation. Trace them back, though, and you may find they all borrowed from the same original.
You can question a claim, go looking for evidence and still come away with the wrong answer. If you then use that answer to judge other information, the mistake may be hard to shake.
I worry about how much false information is out there, but also about whether people get a chance to unlearn it. They might even distrust accurate information because it contradicts what they already believe to be true. That concern goes beyond what the paper tested.
Checking the facts is now a skill employers assess
That brings me to hiring and performance reviews.
As AI takes on more of our research and drafting, I think the ability to check its work becomes more important. We still have to decide whether it's accurate and whether we can rely on it in our work. Employers could look for that judgement when hiring and consider it in performance reviews. I touched on this in my earlier post: as output grows, someone has to review it.
A 2025 study asked 319 knowledge workers about their experiences with AI. It described a shift in critical thinking towards verifying information, incorporating AI responses into their work and overseeing the process as a whole. These were self-reports, so they don't establish that people's judgement has declined or that employers have changed how they hire and assess staff. They do give us a picture of how people review AI's output at work. Read the study
But what happens if employers expect that judgement before they hire someone, then keep assessing it once they're in the job? In Homework We Never Finished, I wrote that if we leave it to experts to decide what's true, everyone else remains dependent on their judgement.
That's why I argued for teaching people to trace claims back to the evidence and read the original sources, even when they aren't experts. If we assess that skill without asking whether they've had the chance to learn and practise, we risk treating a gap in opportunity as simply a gap in ability.
Writing a report and checking its claims take different kinds of practice. Someone who's mostly used AI to turn in finished work may struggle to explain which claims need checking, or how much checking is enough.
When an employer says, “You should be able to judge this for yourself,” where was that person supposed to learn how? They may not have been taught it properly at school or had a chance to learn it on the job.
AI can help people who already know how to judge its answers. We should also ask what opportunities there are for people who still need to learn.
We need more than a warning
That's why telling people “Don't blindly trust AI” only gets us so far.
People need to learn what to check: who wrote something, whether the evidence behind a claim exists, and whether supposedly independent sources all lead back to the same account. They also need practice explaining what they checked and why they decided to trust it.
Stanford reported on a study in which high school students got better at evaluating online information after lessons that included opening other sites to check a source's credibility. That suggests we can teach these habits, rather than wait for students to pick them up on their own. Stanford's account of the study
I'd like schools to look at how students checked their work, as well as what they handed in. New hires should get feedback on the reasoning behind their decisions, too. If we value judgement, we need to give people time to develop it.
I'm still optimistic about what AI can help us do. I just worry that some people won't get the chance to develop the skills they need to benefit from it.
The more we value good judgement, the more we need to think about how people learn it. How is anyone supposed to get their first job if we expect them to have that judgement before they've had a chance to develop it?