top of page
Writer's pictureBen Manuel

I ran 5 ChatGPT articles through a plagiarism checker. Here’s what happened:




Every so often, I’ll get asked:


“Do you use AI to write content? And if not, then why not?”


We don’t – we handwrite all our content. Generative artificial intelligence (AI) isn’t good enough to capture tone of voice, quote relevant sources, or be original.


It may be in the future, but not right now.


Generative AI might still be useful, depending on your content goals.


For example, although using AI to manipulate Google’s algorithm is against its spam policy, it’s not banned outright. This means it might not be terrible for purely SEO purposes.


Likewise, if you’re struggling to maintain an active social media profile, generative AI can be pretty useful to get things out there.


However, AI doesn’t help create relevant and compelling thought leadership pieces.


There’s also another problem:


Plagiarism

Recently, a group of authors, including George R. R. Martin and John Grisham, joined a lawsuit claiming that ChatGPT infringed on their intellectual property.


Most of the lawsuit is focused on potentially illegal use of the writers’ text to train the language model, not the outputs themselves.


Still, how close are the outputs to anything already written – human or otherwise?

Generally speaking, anything less than 10% on a plagiarism checker is considered original – so how does AI match up?


Article #1

For the first article, I used a generic prompt:


“Write me an article for my marketing business”


Result: 22%


Anything above 15% is enough to get a student’s paper disqualified. The margin’s a bit stricter for SEO purposes, with anything above 10% being penalised in most search algorithms.


22% is a problem – but the prompt was quite generic.


Furthermore, most of the articles picked up by the plagiarism checker linked back to articles published throughout 2022 and 2023.


When I ran these referenced articles through a “GPT detector”, most came up positive.

It seems as though these articles were ‘original’ at one point. However, people used this prompt (or something similar) enough times for that to no longer be the case.


Article #2

This time, I got more specific:


“Write me a blog that I can publish to my marketing company's website explaining how plumbers can use AI to get more business in.”


Result: 9%


We’re probably okay to submit this as a student paper - but we’re pushing the margin for SEO purposes.


Once again, most plagiarism citations linked to articles published throughout 2022 and 2023 and were picked up on a GPT detector.


Article #3

This time, I got really specific with the prompt:


“Write me an article that I can publish on my recruitment firm's website. My company values are professionalism, expertise, and diligence. I am based in London. Make the subject of the article about what leaders in FinTech companies need to know when headhunting in 2024. Analyse contemporary trends. Keep the tone of voice business-casual.”


Result: 4%


Now we’re getting somewhere. It wouldn’t disqualify a student, and it’s certainly safe for SEO on the plagiarism front.


It still gets picked up on a GPT detector, so it might still cause issues with search engines.


Article #4: What about Bard?

I ran the prompt from Article #3 into Bard and put the output text into a plagiarism checker.


Result: 18%


Back to square one - not suitable for academia or SEO. Ironically, it did slightly better using the first ‘generic’ prompt


Result: 14%


Still not ideal for publication.


Article #5

For the final article, I ran the following system.


“Come up with some titles for articles I can publish on my marketing company website”.

(This system had been recommended to me by someone too lazy to ideate subject matter, let alone write the article.)


I then asked ChatGPT to turn two of the articles into blogs.


So how did it fare?

1st result: 35%

2nd result: 22%


Two of the ten outputs were the same as articles which had already been published.

Using a title which had already been published, the result was:


Using a title which hadn’t:


Both are dire results.


Conclusion

Leaving aside the quality of the content, it’s pretty clear that GPT has potential problems with plagiarism.


You could mitigate it by using complex, long, and original prompts.


However, suppose you’re using it regularly for your content marketing: In that case, you run a relatively significant risk of plagiarising someone else’s material – whether AI content published before yours or human-written content.


It’s unclear how this will play out in the long term.


Will businesses face legal consequences for stealing material, even if they didn’t realise they were doing it?


What will happen to these generative AI companies?


Plagiarism aside, each article's subject matter and content was repetitive, shallow, and fundamentally unoriginal.


If you plan on using generative AI for your business’s content marketing, running every output through a plagiarism checker is probably best practice.


We don’t use AI to write content, but you might. 







bottom of page