Home

 › 

Articles

 › 

Why You Should Never Use an AI Writer: Line-by-Line Analysis of AI Writing from ChatGPT, Chatsonic, Google Bard AI, and Bing Chat

Opera introduces ChatGPT Integration; Deep Learning

Why You Should Never Use an AI Writer: Line-by-Line Analysis of AI Writing from ChatGPT, Chatsonic, Google Bard AI, and Bing Chat

Key Points

  • AI writers lack creativity and produce content that sounds the same and lacks originality.
  • AI writers rely on training materials that may be obtained unethically, potentially violating copyright laws.
  • AI writers have poorer credibility than human writers, often making up information and producing inaccurate content.
  • The analyzed AI writers showed word-for-word similarities and lacked attention to detail in understanding the subject matter.

We know how tempting it can be to outsource your work to AI. After all, who doesn’t want to do less work and get paid as if they were doing the same job?

However, there are many reasons why you should avoid using an AI writer. These reasons are compelling as both why individual writers should avoid using AI as a “shortcut” to their work and why clients should avoid trying to cut corners by outsourcing their writers’ work to AI writers.

In this spicy opinion piece, we’ll analyze passages written by AI writers line-by-line to see if they hold up in the writing world. We’ll rate and critique these pieces on their credibility, fluency, and how well they mirror human writers’ passion and research in every article they produce. We’ll also give you some reasons why we think that human writers are irreplaceable in the journalism world.

Let’s break it down below.

3 Reasons Why You Should Avoid Using AI Writers

There are several reasons why AI writing should be avoided. To truly understand why some of these issues arise when using AI writers, you must first understand how AIs are developed and trained. AIs must be trained on training materials. These training materials must be scraped from other sources and are typically not produced by the AI company

This sourcing can create problems for the company since the training materials may only be ethically sourced if funding to produce appropriate materials is provided. Inappropriate or unethical sourcing can create many problems for the company regarding the material produced from that training material.

Let’s examine why avoiding an AI writer is in your company’s and individual’s best interest.

AI Writers Lack Creativity

One of the primary reasons that you should avoid an AI writer is that AI writers lack the creativity that naturally occurs when a human writer produces a piece. AI writers are trained on already written articles that they analyze and learn from. Unfortunately, this learning process means that the AI will naturally emulate the pieces it learned from… down to the word.

Since AI writers are trained on a limited amount of training material, an AI writer will likely suffer from “same article” syndrome, where the articles it produces all sound the same and use the same language. Even when we simply put in a prompt to create pieces for us to analyze for this article, different AI writers produced articles that suffered from the “same article” syndrome. As you’ll see below, sections of separate pieces by different AIs are word-for-word the same because the AIs were trained on similar materials.

This lack of creativity produced by AI writers significantly affects the content you create for your website. When the content on your site is too similar across several articles, it can be exhausting for the mind to read. This exhaustion will cause your site to bleed readers. Rather than becoming fans of your site and loyal readers, your readers will tap out after a few articles and probably won’t return. This completely hurts your credibility.

Secondly, it can cause your articles to appear too similar to other articles on the web, especially other articles produced about the same topics by the same AI writers. This same article syndrome can lead to accusations of plagiarism from other sites and the original writers whose writing trained the AI to write in the first place, leading us to our second point.

AI Writers Are Just Softcore Plagiarism

The heading here may not entirely be true. Some people would rightfully argue that AI writers are not “softcore” plagiarism but hardcore plagiarism. However, to not completely alienate the people who need to hear this message, we’ll be gentle.

AI writers need to be trained to write. The way we teach AIs is to input information that we want them to replicate. This training method means feeding written pieces into the AI so it can learn what the user wants it to reproduce. Unfortunately, these written pieces fed into the AI must come from somewhere.

If AI companies were ethical about training AIs, they would pay writers to write articles specifically for teaching their AIs. However, as most writers will tell you, companies want to avoid paying for art. Finding companies that don’t pay starvation wages to their writers is like finding a needle in a haystack.

Thus, we can safely assume that most of the writing fed into AIs to train them wasn’t obtained in a way that most people would consider “ethical.” One might even call it “plagiarism.” (That’s what it’s called. It’s plagiarism. We’re just trying to be nice.)

AI writers are not the only type of AI that requires a constant drip feed of other people’s work to learn from, either. AI image generators and AI voice generators need input to create an output, and these generators are primarily being fed other people’s work without those people’s permission.

Several voice actors and artists have spoken out about their work and, in the case of AI voice generators, their likenesses being used without their consent. (Remember that using someone’s likeness without their permission is not protected by the First Amendment or Fair Use laws.)

The works used to train the AIs to generate content may also have been copyrighted. Thus, the AI’s very training would have violated copyright law unless the artists, writers, and voice actors whose work was used to train the AI were appropriately compensated. The permission of the copyright holder (who, in the case of writing, may not even be the writer) would also have to be appropriately requested and granted through a legal contract.

While no verdict has been handed down at the time of this article’s writing, several artists and the media company, Getty Images, are suing several AI image generator companies for using copyrighted images to train their AIs. The AI companies have asked the courts to dismiss the case, but should the courts decide to proceed with the cases, these cases could end up providing a legal precedent for writers and media companies that employ writers to sue for the misuse of copyrighted written pieces to train AIs.

AI Writers Have Poorer Credibility Than Human Writers

Human writers have to research everything they write about. This research process differs from writer to writer, but every writer has at least some due diligence when writing. AI writers are well-known for simply making things up when they run into a knowledge gap, such as in the case of this lawyer who was sanctioned for citing cases that never happened because ChatGPT made them up.

With the amount of stuff that AIs have been found to make up on the spot, you can barely even trust an AI to function as an aid for research, let alone trust it to write your whole piece for you. At some point, you will run into a situation where the AI makes up some unfounded stuff in one of your articles, and you’ll get called out for not doing your research due diligence.

ChatGPT, one of the most popular AI writing tools, even has a warning at the bottom of each chat stating that ChatGPT may produce inaccurate information about “people, places, or facts.” Google Bard AI comes with a warning stating that the AI may produce “inaccurate or offensive information that does not represent Google’s views.” Chatsonic/Writesonic and Bing Chat don’t have warnings, but they probably should because AIs are pretty fond of making stuff up.

AI Writers: A Line-By-Line Analysis

Now, we will analyze some samples written by different AI services. For this exercise, we selected Google Bard AI, ChatGPT, Chatsonic/Writesonic, and Bing Chat as our guinea pigs. We had them write some samples about mechanical keyboards, a topic we are intimately familiar with and quite knowledgeable about.

Google Bard AI 

“Mechanical keyboards use individual mechanical switches under each key, which provide a tactile and audible click when pressed. This makes them more satisfying to type on than membrane keyboards, and they also offer a wider range of customization options. Some popular switch types include Cherry MX Red, Blue, Brown, and Black. Mechanical keyboards are a popular choice for gamers, typists, and anyone who wants a more satisfying typing experience.”

For starters, Google Bard AI is incredibly long-winded. I asked it four times to write a one-paragraph passage about mechanical keyboards, and it spat out four paragraphs every time. Ultimately, I took the first paragraph of what it wrote and used that since the first paragraph was fairly comprehensive and easily comparable to the other AIs passages.

That being said, let’s take a look at what was written. First of all, this passage is a little bit wordy. “Which provide” could easily be shortened to “providing,” the “they are” in “when they are pressed” is superfluous, and “a popular choice” can be shortened to “popular.”

Additionally, the “this” in “this makes them more satisfying” is an unclear antecedent, meaning it’s unclear what the word “this” refers to, grammatically speaking. The passage also gets a bit repetitive, with the terms “satisfying” and “popular” being used multiple times in the same paragraph.

There is also the fact that the passage implies that all mechanical switches provide a “tactile and audible click” when pressed, which is untrue. Tactile and linear switches do not provide an audible click; this is a unique feature of clicky mechanical switches.

Now, let’s move on to ChatGPT.

google bard ai technology artificial intelligence
Google Bard’s AI-written passage was very wordy and included factual errors and repetitive language.

ChatGPT

“Mechanical keyboards are a popular choice among computer enthusiasts and gamers alike due to their exceptional performance and unique tactile experience. Unlike traditional membrane keyboards, mechanical keyboards utilize individual mechanical switches beneath each keycap. These switches come in various types, such as Cherry MX, Razer, or Logitech Romer-G, each offering distinct actuation forces and feedback levels. This diversity allows users to customize their typing or gaming experience to match their preferences and requirements. The satisfying click-clack sound produced by the switches adds to the overall appeal of mechanical keyboards, enhancing the typing experience and creating a sense of nostalgia for those accustomed to the classic typewriters. Beyond their distinctive feel and sound, mechanical keyboards are also more durable and reliable, with each key capable of withstanding millions of keystrokes. Whether for work, gaming, or pure typing pleasure, mechanical keyboards stand as a top choice for users seeking exceptional performance and a satisfying tactile response.”

ChatGPT’s passage is also wordy by most SEO standards. Typically, SEO looks for paragraphs no more than four lines long; this paragraph is a staggering eleven lines long (at least that’s how long it is in Google Docs in Times New Roman 11-point font). If submitted to most SEO-compliant companies, this paragraph would probably have to be split manually by either the writer or the editor.

Digging into the writing of the paragraph, we have a few syntax and grammatical issues. Once again, “a popular choice” can be shortened to just “popular,” the “mechanical” in “individual mechanical switches” is redundant and repetitive since the phrase is preceded by “mechanical keyboards.”

This passage also implies that all mechanical keyboards provide a tactile click, which is untrue. This passage goes quite in-depth with that untruth describing the “satisfying click-clack sound” that is only offered by clicky switches and won’t be present if you buy a keyboard with tactile or linear switches.

Additionally, “mechanical keyboards are a popular choice” is a word-for-word similarity between the passages from ChatGPT and Google Bard AI. That’s not super important since there aren’t any other word-for-word similarities, and ChatGPT and Google Bard AI produced larger samples, but the lack of originality is worth noting.

Let’s move on to Chatsonic.

openAI chatGPT
ChatGPT’s passage was also wordy, plus it contained factual inaccuracies, verbatim phrasing, and grammatical and syntax-based errors.

Chatsonic/Writesonic

“Mechanical keyboards are a popular choice among computer enthusiasts and typists alike. Unlike traditional membrane keyboards, mechanical keyboards use individual switches for each key. This design provides several advantages, such as tactile feedback, increased durability, and improved typing speed. The unique feel and sound of each keypress make typing on a mechanical keyboard a satisfying experience. Furthermore, the ability to customize keycaps and switch types allows users to tailor the keyboard to their preferences. Whether for gaming, coding, or everyday use, mechanical keyboards offer a superior typing experience that is hard to match.”

Chatsonic/Writesonic created a much more reasonable single paragraph. It’s still a bit long for SEO purposes, but not by a blinding amount like ChatGPT. Unlike Google Bard AI, it’s also not four paragraphs, so it followed the instructions.

Regarding syntax and grammar, you’d typically want to change “the unique feel and sound of each keypress” to “each keypress’ sound.” Other than that, Chatsonic/Writesonic is pretty solid with basic grammar structures.

However, when we dig deep into the writing, we have a startling lack of originality. Several phrases are word-for-word in the other AIs’ writing, particularly ChatGPT. “Mechanical keyboards are a popular choice” was found in both Google Bard AI and ChatGPT’s passages, and “Unlike traditional membrane keyboards” was found in ChatGPT’s passage. Additionally, this passage also used the word “satisfying” to describe the experience of typing on a mechanical keyboard, though the section wasn’t word-for-word the same.

We also have the same lack of attention to detail in understanding that not all mechanical keyboard switches provide a tactile experience. Linear switches are specifically designed not to provide any tactile feedback. Thus, implying that the tactile experience is universal to mechanical keyboards is a factual inaccuracy.

Next, we’ll analyze Bing Chat’s paragraph.

Chatsonic vs. Google Bard
Chatsonic returned an AI-written passage that was still long, unoriginal, and repetitive with grammar and syntax issues.

Bing Chat

“Mechanical keyboards are built with high-quality, typically spring-activated key switches. These key switches vary based on the keyboard’s application or user preference. Mechanical keyboards are known for their tactile feedback and audible click sound when typing. They are also more durable than membrane keyboards and can last longer. Mechanical keyboards are popular among gamers and typists who want a more responsive and tactile typing experience.”

Bing Chat was much more reasonable with the paragraph it produced. It clocked in at just six lines which is a little long for our tastes, SEO-wise, but not by too much. I would still submit a paragraph of this length. There are paragraphs longer than this in this very article.

Grammatically, Bing Chat did a good job. Now, that could be attributed to how short the paragraph is. Put simply: the AI has fewer chances to mess it up if there’s less information. However, one notable thing I would have personally changed is that I would not have included the phrase “and can last longer.” This phrase is a bit redundant following the statement “also more durable” as typically, items that are more durable last longer; the “can last longer” is already implied by “more durable.” 

Bing Chat also had no word-for-word phrases found in other AIs’ writing. Some phrases like “Mechanical keyboards are popular” are similar to “Mechanical keyboards are a popular choice.” However, there are no phrases that are outright the same.

Still, Bing Chat fell into the same factual hole as the other three AIs in that it implied that all mechanical keyboards produce tactile feedback and audible click noise. This issue was consistent amongst all four AIs and is likely due to the prevalence of clicky switches and how popular these switches are with mechanical keyboard users.

bing chat AI
Bing Chat’s passage was redundant and not entirely correct with similar phrasing to the other AI-written passages shown above.

How Did the AI Writers Do Overall?

Of course, I will be biased here since I am a writer and have a vested interest in my job not being taken by an AI. However, I actually found that the AIs did much better than we thought they would. My personal experience with AIs has been less in AI writing — as a writer, I have no reason to use an AI writer –– and more in AI companionship. Thus, I see more AI-generated content with staggering grammatical errors and confusing syntax than is present with AI writing.

However, the lack of originality in the AI’s writing was very apparent. All of the passages were very similar in content, and several of the passages had word-for-word consistencies. If these passages were used in actual articles, they would feel and sound far too similar and suffer from same-article syndrome. Many readers would notice this immediately if researching and reading several articles on the same topic.

The grammar and syntax errors were somewhat minor, which is much better than assumed, but it’s still a major issue in readability and credibility.

It’s important to remember that Google Bard AI, ChatGPT, Chatsonic, and Bing Chat are primarily conversation models. They’re not meant to do research or write professionally, as shown in the passages we showed above.

Overall? The AIs did a not-terrible job. The stilted writing style and lack of passion is noticeable, but you may not notice at first (or second) glance that they’re AI-written. It may just seem like a not-so-skilled writer wrote it. However, they don’t hold up to deep scrutiny and won’t pass for human writing when put through the wringer.

Final Thoughts

Artificial intelligence is a huge market and has sparked massive debates about the creative process. These conversations are healthy and important as they allow creatives to better showcase what sets them apart and makes them so valuable.

AI writing might be tempting whether you’re a writer or a client, but understand that as they are, these programs cannot withstand the deep scrutiny that a worthwhile editor will put them under. 

Summary Table

Reasons to Avoid AI WritersExplanation
AI Writers Lack CreativityAI writers are trained on already written articles, which they analyze and learn from. This learning process means that the AI will naturally emulate the pieces it learned from, leading to a lack of originality and creativity in the content produced.
AI Writers Are Just Softcore PlagiarismAI writers need to be trained to write, which involves feeding written pieces into the AI so it can learn what the user wants it to reproduce. This training method often involves using other people’s work without their permission, leading to ethical and legal issues.
AI Writers Have Poorer Credibility Than Human WritersAI writers are known for making things up when they run into a knowledge gap, which can lead to credibility issues. Human writers, on the other hand, have to research everything they write about, ensuring accuracy and credibility in their work.

Frequently Asked Questions

What is an AI language model?

AI language models distribute probability over words. In practice, these AIs read and generate text, simulating a conversation.

What is an example of an AI language model?

ChatGPT is probably the most well-known AI language model at the time of this article’s writing. Other examples of AI language models are Google Bard AI, Chatsonic, and Bing Chat.

Can an AI language model write?

AI language models are capable of writing prose, poetry, and even generating song lyrics. The question we have is not whether they’re capable of writing but how well they’re capable of writing.

How are AI language models trained to write?

AI language models are trained to write by feeding written pieces into the AI and allowing AI to analyze and learn from the written pieces.

Is feeding written pieces into AI plagiarism?

At the time of this article’s writing, there is no legal precedent declaring the use of someone else’s work -– even without their permission -– plagiarism. However, there were several lawsuits in progress at the time of this article’s writing that may influence that legal precedent one way or the other.

To top