AI vs. Human Writing - Can You Really Trust AI Chatbots?

You’re in the middle of a hectic day and have a dozen things to do: need a grocery list, write that email to your professor you’ve been putting off, and work on the deadline-nearing essay. 

Naturally, you turn to an AI chatbot and ask it to give you a grocery list based on some random recipe idea. Bam! List done. Need a quick email? No problem, it writes it up in perfect, polite business language. Can’t think of how to start that essay? The artificial intelligence is quick to suggest an intro.

And it feels great, right? You start thinking, "Wow, this thing can basically do all my homework." Big mistake. Big. Huge!

Let’s say you trust the AI to help you find credible sources for your paper. But later, you find out one of those sources doesn’t exist. The chatbot just made it up. And not just that, it even created a fake author and a fake journal. Turns out, the AI invented a whole new study out of thin air.

This actually happens a lot. It’s called AI hallucinations where the chatbot confidently gives you information that’s just… totally wrong. As reported in The New York Times, these bots can make things up anywhere from 3% to 27% (!) of the time. So, basically, about 1 in every 30 times you ask something, you could end up with fake data or sources. 

Let’s break down exactly why AI chatbots make stuff up and what’s going on behind the scenes when they do.

Why AI Hallucinations Happen in The Writing Process

Ever looked at the clouds and seen shapes or faces that aren't really there? Our brains are wired to make sense of chaos and find patterns even in randomness. AI programs do something similar.

What is AI Hallucination?

According to IBM, an AI hallucination happens when large language models (LLMs) generate information that isn’t true or real. As simple as that.

LLM is trained on huge amounts of text (books, websites, research papers — everything), so it can respond to questions in a way that feels natural. For example, when you ask a chatbot like ChatGPT to write a story, it’s using an LLM to pull from all the patterns and info it’s learned. But it doesn’t really understand like we, human beings, do; it just knows how to put together pieces it has seen before. 

So when you ask an AI chatbot to write an essay intro, it's using its pre-trained language model to mix and match phrases it remembers. And it works most of the time! But sometimes, when it doesn’t have the right pieces for the puzzle you give it, it cuts some cardboard, makes its own pieces, and then forces them together to make something that looks like it fits, even though it doesn’t. 

You might also hear people call this problem confabulation. In psychology, it happens when we unknowingly fill in memory gaps with things that didn’t actually happen. That’s what artificial intelligence is doing. It’s not like it’s trying to lie or gaslight you — it just fills in the gaps, without even knowing it’s wrong.

Why Does This Happen?

There is a bunch of reasons AI chat bots make up information, and here are 6 of the big ones:

1. Overfitting: You’ve studied so hard for a test that you memorize every single example from your textbook. But the test asks different questions, and suddenly, you're not sure how to apply what you know. This is what happens when AI is too closely trained on specific data. For example, if you train an image recognition AI on thousands of pictures of apples — all red and round like in “Snow White” — it might struggle to recognize a green apple because it's fixated on what it saw during training.
2. Training Data Bias: AIs learn from the data they’re trained on, and if that data has errors or biases, it will absorb them. This actually happened in real life. A study by MIT Media Lab found that some popular facial recognition systems worked almost perfectly for white men (like, less than 1% error). But when it came to identifying women with darker skin, those same systems made errors up to 34% of the time.
3. Lack of Context: AI doesn’t “get” context like we do. It’s just scanning for patterns in the words you give it. So, if you ask a vague question, it might throw out an answer that’s not what you meant. Imagine asking a chatbot, “Can I run with scissors?” and it responds with tips on marathons instead of the whole “safety” issue. 

4. Not Enough Data: If AI hasn’t been trained on enough examples of something, it will just start making stuff up. Have you ever tried writing an essay on a book you barely skimmed? You’d start guessing, filling in the gaps with whatever seems logical, but a lot of it would be way off. That’s exactly what happens when an AI gets a question it hasn’t seen enough examples of. It’ll try, but the guesswork can be obvious.
5. Outdated Info: AIs are trained on data, but the world changes. If the AI hasn’t been updated with the latest info, it might give you answers that are outdated or just plain wrong. 
6. Random Glitches: And then, there are the random weird moments. Every now and then, the AI glitches out and gives an answer that makes zero sense. There’s no big explanation for it; just think of it as an AI equivalent of your computer crashing for no reason.

If you’re worried about getting authentic content, why not let a human essay writer handle your next big project?

When AI Hallucinations Go Wrong: Real-World Consequences

Alright, so at this point, you might be thinking, “Okay, artificial intelligence makes stuff up sometimes, but it's not like it's hurting anyone… right?” Well, not so fast. When these hallucinations happen in more serious situations, they can cause real-world, big problems.

Let’s get into some examples of when AI led to some serious consequences:

• Sexual Harassment Scandal: Let’s say you’re a professor, and one day, out of nowhere, you’re accused of being involved in a sexual harassment scandal during a trip to Alaska — a trip you never even took. To make matters worse, ChatGPT cites a non-existent article from The Washington Post to back up its claims. Well, that’s exactly what happened to a law professor at George Washington University.
• Canada Lawyer Using Fake Cases: In Canada, one lawyer used ChatGPT to help with a court case. The chatbot came up with several legal cases to support the lawyer’s arguments, except none of those cases actually existed. Turns out, ChatGPT had just invented the cases, including the decisions and details. 
• Refunds That Didn’t Exist: Speaking of legal disasters, Air Canada found itself in quite a situation thanks to an AI hallucination. A customer service chatbot invented a refund policy that didn’t exist. The passenger took the airline to court, and guess what? The court ruled that Air Canada had to honor the fake policy the AI made up. 
• Crime Accusations Against a Journalist: Microsoft Bing Copilot pointed fingers at a German journalist, pinning some serious crimes on him (like child abuse and fraud) just because he had reported on them. 
• Healthcare Fails: Just check online, and you'll find heaps of examples where it got things wrong. Always double-check with a real doc!

If you're curious about just how weird AI can get, take a look at this example from Google’s AI that had some major fails (check it out here). It's one of those moments where you don’t know whether to laugh or be a little freaked out. 

AI vs Human Writing: Can You Rely on AI Chatbots for Accurate Information?

Okay, sure, AI models make things up sometimes. But how does that affect you in school? Students usually just need help with essays, assignments, and getting through the never-ending academic nightmare.

Well, artificial intelligence can mess up academic writing too. Badly. 

Tools like ChatGPT use something called predictive modeling, which is based on massive datasets of text that the algorithm has been trained on. Instead of knowing the answer, ChatGPT is predicting what the next word or phrase should be based on patterns it’s learned from all the data it’s been fed.

Because it relies so much on patterns, ChatGPT struggles with accuracy. It’s great for routine tasks like brainstorming or writing drafts, but for academic work that needs real facts, critical thinking, or specific references, it can lead to mistakes. 

Let’s break down a few specific ways AI can hallucinate in academic settings:

• AI Can’t Even Count: You’d think something as simple as counting characters and words would be easy for AI, but you’d be wrong. AI doesn’t actually count in the way we do (remember, it predicts based on patterns). This means that when you ask it to hit, say, a 2,000-word count for your essay, it might miss the mark by quite a bit. 
• Fake Statistics and Facts: Artificial intelligence loves to get creative. There was this wild example with Google’s AI where it claimed that Barack Obama was Muslim. And it’s not just about political figures — AI can easily invent numbers or dates that don’t exist. Using fake statistics in academic papers can absolutely destroy your credibility.

• Fake Articles and References: You’re almost done with your paper and you ask an AI tool to give you some references to back up your points. Then, your professor checks one of the references, and it turns out... the article doesn’t exist. Not even a trace of it anywhere online. 

This happens more often than you’d think. According to an article by Macquarie University, the reason ChatGPT generates fake references and false information is because it doesn’t have access to a database of facts, so if it hasn’t seen something in its training data, it might still try to predict an answer (and that’s where the AI hallucinations come in).

2023 Fake Reference Experiment

In the article above, the author also ran an experiment to understand why ChatGPT sometimes makes up fake references. The goal wasn’t to bash the tech (because, honestly, it’s incredible in many ways) but to understand how it works.

Here’s what they did: they asked ChatGPT to generate a list of academic references for a specific topic. The response was quick and confident: scientific publications, properly formatted APA citations, author names, journal titles... 

But when they checked the references, 5 out of 6 were fake. For example, in one case, ChatGPT gave them a reference to a legitimate book, but with the wrong authors. It also invented journal articles that never existed. 

Fast Forward to September 2024: Our Experiment

Inspired by this experiment, I decided to see if things had gotten any better in September 2024. I mean, a year and a half is a long time in tech, right? So I ran a similar test: I asked ChatGPT to generate academic references on a topic, expecting it might be a bit more accurate now.

Well, here’s what I got: every single reference was fake. 

It turns out, ChatGPT still uses the same guessing game when it comes to references. It doesn’t have access to databases with actual journal articles. Instead, it’s still using its predictive model, stitching together real-sounding names and journal titles in ways that are believable but totally inaccurate.

And this is with GPT-4, which is supposed to be the most accurate version yet!

AI Writing vs Human Writing: A Detailed Comparison

It’s not just about the fake facts or references. Sometimes, you’re sitting there reading something, and halfway through, you get this gut feeling: "Did a robot write this?"

You don’t even need those AI detectors, which, honestly, don’t always work (but for added peace of mind, here’s a list of the best AI content detection tools to help you find an effective AI scanner for essays).

It’s more about the way the writing feels. Sure, if you’re reading one AI-generated article, it might not stand out right away. But your professor is going through dozens, or even hundreds, of essays. The patterns and tells will start to pile up and become really obvious. 

Let’s look at some of the differences between human writing and AI writing, and why, even without tools, you might be able to tell the difference:

1. Artificial Intelligence is Predictable (Literally)

Perplexity is just a fancy term, but it’s one of the ways AI detectors often catch AI-generated content. It just means how predictable a piece of text is. High perplexity means the writing is more creative and surprising. Low perplexity means it’s predictable and boring. 

AI-generated text usually has low perplexity because it’s designed to be as average and non-risky as possible.

Let’s say you write the sentence:

I grabbed my keys and headed out to the...

• A high-perplexity answer might be "circus" or "moon" (weird, but unexpected).
• A medium-perplexity one could be "train station" (makes sense, but still a little unusual).
• But AI, with its low perplexity, would go for something like "car" or "store" — predictable and safe.

2. We Mix It Up

Sometimes you’ll write a long, complex sentence with tons of detail for your research paper. Other times, you need something short and sweet. It’s how we talk, too. Some thoughts need unpacking. Others? Just a simple "yeah" will do. This variety is called burstiness, and that’s the human language for you.

The paragraph above has a burstiness of 150, mixing up sentence lengths, while the ChatGPT example below sits at 50, making it feel a bit too smooth and predictable.

3. We Don’t Overgeneralize

Let’s say you’re writing an essay about political compromise, but instead of sticking to the dry examples, you add a story about personal relationships. You could start with something like:

The whole idea of political compromise reminds me of this time my girlfriend and I couldn’t agree on where to go for vacation. I wanted a mountain retreat, she wanted a beach resort, but we settled on a boring, lake cabin filled with mosquitos (which honestly sounds like a scenario of a horror movie!). We both ‘won,’ but honestly, we both lost.

Artificial intelligence has no real-life connection, no emotion. That’s why it tends to overgeneralize and make statements that don’t really add value, like “Politics is complicated and requires compromise.” It prefers generating safe, neutral responses that won’t offend or stray too far from commonly accepted knowledge.

As a result, AI writing falls flat in creative essays, as it can’t add funny, real stories, personal opinion, human judgment, and anecdotal evidence that make your reader nod along and think, “Yeah, I get that.”

4. AI is Too Perfect

Most ironically, AI follows grammar rules to a T. No weird phrasing and accidental typos. And while grammatical perfection might sound like a good thing, it can actually make writing feel less human. We’re messy when we write, we make mistakes, and we let contractions, colloquialisms, or sentence fragments slip. 

5. AI Doesn’t Know the Sweet Spot

Artificial intelligence is either too formal or too simple. When it goes full-on academic and you ask it to make it simpler, natural, and more engaging, the AI-generated essays might come across as awkward because it doesn’t truly understand humor or human touch. As a result, you get sentences that sound a bit off or forced.

Often, when you ask AI to write casually, it might spit out something like: 

Hey there, friend! Let's embark on a wild journey of learning!

Most people wouldn’t say that. We know how to adjust. We know when to be formal and when to relax a little, depending on who we’re talking to.

6. AI Technology Gets Stuck

For me, one of AI’s biggest giveaways is repetitiveness. When humans write, we instinctively know to change up our words to keep things fresh. We don’t repeat the same phrases over and over because we know it gets boring fast.

Artificial intelligence, though, loves to "dive" or "delve" into topics so much, you'd think it was training for the Olympics.😅

AI models, like ChatGPT, also have a habit of sticking to specific transitions, like "in conclusion," "moreover," or "however." So, if you’re reading something and notice “it is important to note” in nearly every paragraph, it’s a sign you might be looking at AI-generated text.

To really show you what I mean, I’ve put together a list of 100+ words that AI seems to love using over and over. This list isn’t just from my own experience — it’s the experience of our entire team who work with writing and editing, and we’ve all noticed the same patterns:

These are the words AI relies on the most, and once you spot them, it gets a lot easier to recognize when you're reading something written by a machine versus a person.

ChatGPT vs Human

Let’s put all this into action. Below, I’ve added a side-by-side comparison of AI vs human writing: on one side, we’ve got an AI-generated text where you’ll see all these issues highlighted. On the other, I have a version written by one of our pro writers, with that natural flow, variety, and coherence. 

Take a look:

The AI version feels pretty robotic: it’s formal, repetitive (look at all those words highlighted in purple), and doesn’t give any real-world examples to make the points stick. It talks about big ideas like "social development" and "emotional intelligence" but never shows how these actually play out in everyday life. 

In contrast, the human text feels like a real conversation. It has specific examples, like Finland’s education system, which makes it easier to understand. Instead of just talking in general terms, it uses stories to show how ECE works in real life, making the whole thing more engaging and, well, human. 

What do you prefer more, human or AI text?

One More Thing: ChatGPT is a Thief

That’s right: it’s not as private as you might think. 

According to OpenAI's legal documents (the ones most of us skim over), ChatGPT can actually use the data you enter to improve its models. In simple terms, this means that anything you type in could, in theory, be used later on to train the AI. 

What Kind of Data is Vulnerable?

If you’re just asking harmless questions about general knowledge or fun facts, you’re probably fine. But things get a little risky when you start feeding personal information or confidential data into the system.

For example, let’s say you’re working on a project and you type in some proprietary business statistics or company strategies for help with analysis. That data, even if it’s something specific to your school project, could end up being used to further train the model. Worse, there’s a chance it could be pulled into future conversations (not necessarily exactly, but in some form) when someone else asks a similar question. 

Here’s the kind of data that’s most vulnerable:

• Sensitive business or financial information (like statistics or internal strategies).
• Personal information, like your full name, address, or anything that identifies you.
• Academic research or thesis topics that haven’t been published yet.

What Can You Do About It?

I know, it’s easy to feel a bit paranoid after hearing all that. But here are some strategies you can follow to stay safe while still using ChatGPT for your academic and personal needs:

1. Be Mindful (And Demure) of What You Share: This might seem obvious, but it’s worth repeating: don’t share anything private. If you wouldn’t want it repeated publicly, then don’t enter it. 
2. Use Code Names: If you’re dealing with sensitive data (for example, working on a business case study), try using placeholder names or fake figures when asking questions. Instead of typing something like, “My company X made $10 million in Q3,” rephrase it as something more abstract like, “What are the common challenges companies face when they make $10 million in revenue?” 
3. Clear Your History And Turn Off The Model Improvement Mode: If you’re worried about sensitive data, get into the habit of clearing the chat logs where possible. While this may not completely prevent data from being used in training, it’s a small step that can add some extra privacy protection.

Summing Up

At the end of the day, AI tools like ChatGPT are not perfect. They can churn out decent text, of course, but they often fall into predictable patterns, repeat phrases, and lack our special human touch. Plus, as we’ve seen, there are real concerns about privacy, data leaks, and how much AI memorizes from what users input.

But we are not programmed. We know how to make content feel engaging and relatable. We know when to add humor or emotion to connect with human readers in ways AI simply can’t. 

So, if you’re looking for writing that’s fresh, creative, and fully personalized, opt for human writers. Let our professional custom essay writing service help you create something unique and meaningful!

FAQ