When searching for basic factual information, instructions, or ideas for what to cook for dinner, "I asked ChatGPT" has become as ubiquitous as "I Googled it." Despite starry-eyed discourses about how AI is going to transform our lives and usher in a limitless future, more and more studies confirm that AI is making us stupider in an era where it's crucial to maintain our critical thinking skills. Here's why generative AI might seem like an efficient means to an end, and why it's worth taking a longer look at.
Why AI Generated Content Looks Legit
It sounds authoritative.
While you can ask an AI to present its output in the style of Henry VIII or Luigi or Daffy Duck if you want to, its default is a detached, semiformal cadence that sounds confident and academic-ish, if we don't look too closely.
But this authoritative tone often masks claims that are either nonsensical, incorrect, or simply empty. Large Language Models (LLMs) like ChatGPT, Claude, or Gemini create text by ingesting massive amounts of data (from where, we don't know) then regurgitating words and sentences most likely to appear in relation to whatever prompt you enter. It's good at sounding good without actually saying anything.
It cites sources.
Most LLM results now include links to sources that ostensibly support the information in the output. But AI isn't a search engine. It's a probability machine. Asking ChatGPT a question, however simple or complex, doesn't trigger it to find the most relevant and reliable sources and synthesize them into a clear, accurate summary. Rather, it generates a response based on training data and inserts links that, based on that training data, seem like the most likely match. Often, even if the sources are real, they don’t say what the output says they do.
Here's the dangerous part, though: it's not always wildly off base. It gives you real sources most of the time, and it’s more or less accurate often enough that it's easy to trust it, especially for low-stakes queries. Eventually we assume it's fine, skip the tedious step of verifying the information, and eventually lose the cognitive capacity to distinguish real from fake in the first place.
Okay...so AI is an unreliable source?
It's a little more complicated. Here's why I argue it's not a source at all.
AI has no context.
Information is created by people. We can tell if a source of information is relevant and reliable based on its context: who created it, when, why, for whom, what the circumstances were around its creation. These factors are how we tell the difference between a legitimate news article and clickbait, between serious and satire, between a research article and a sales pitch.
Even a lie is information—it's just bad information. But it has context. A specific person told you that lie, likely for a specific reason, and that in turn allows you to make inferences about that person's character and motives.
AI-generated content, in contrast, is severed from any context: created on demand by no one for no purpose than to resemble as closely as possible what the prompt describes. Because it relies on probability rather than context, the output is predictable and formulaic. It's how you can look at patterns in language and tell immediately that it wasn't written by a human. It's not information; it's just words.
Did you notice this section uses an em dash—this punctuation mark—and includes multiple sentences structured like "It's not X. It's Y." Did you wonder, when you read that section, if it was written by AI? It wasn't, but that's why it might sound like it was.
AI is a people pleaser.
LLMs are programmed to give you what, according to your prompt, you most likely want to see. And they flatter you while they're doing it, complimenting your great questions and insights.
The fact that it's often somewhat accurate is less intentional than it is merely incidental, a function of the fact that most of the data it's trained on says similarly mostly correct things. It's runs on probability, not reality.
AI is ethically questionable.
AI data centers have a huge impact on their surrounding areas, spurring protests in areas where their construction is planned. It muddies the waters of copyright and authorship. It collects data about you that can then be sold to shady surveillance companies. It's making it harder and harder for college grads to get entry-level jobs. And it's making us stupid.
The bottom line
To think critically about AI-generated content, you need skills you can only develop without AI.
People spend careers learning to use information to generate new ideas based on existing ones. It takes practice, and it's messy. That's why research takes time, why people change their minds, why learning requires mistakes. If the only "information" you ever interact with is simulated and flattened of context, you're missing out on opportunities to develop this skill, which makes it harder to recognize bullshit when you see it.
Here are a few tips to find reliable information...and keep your brain sharp enough to understand it.
Rely on search engines, not AI.
This way you can trace information back to individual sources. Can you trust the person or organization who created or shared it? What is the context of the information? Is the source trying to sell you something?
Cross reference results.
Do multiple sources corroborate your findings?
Read whole articles, not just headlines.
Do you understand what you read? Does the headline accurately represent the gist of the article, or is it misleading? Are you reading with an open mind, or just looking for a source that supports what you already believe?
Conclusion
We've been navigating social media for decades before AI and have had to learn to distinguish reality from falsehood from "truthiness," But when a certain incontinent, dementia-ridden dictator who long ago redefined fake news as "any media outlet that says things that do not flatter me" routinely posts AI slop on his personal Truth Social and from official White House accounts and has seemingly no awareness of the irony, it's worth pausing to think about the effects of this technology.
I'm not saying there is never a use case for instances of AI. But remember, when you "Ask ChatGPT," no one is answering.