Updated: Aug 8
By Copy Writes Copy
"What is a Text Classifier?", we hear you ask.
The OpenAI Text Classifier, according to the OpenAI website is -
"a fine-tuned GPT model that predicts how likely it is that a piece of text was generated by AI from a variety of sources, such as ChatGPT."
To some content creators, the thought of a Text Classifier checking over their work might send shivers down their spines, as they recall the last piece of copy they submitted, that wasn't as human as they claimed.
But, these things are never straightforward, as you'll see from the results we got, when we put our web copy through the OpenAI Text Classifier - skip to the results here.
So, is it wrong to pass off AI copy as your own?
As proud and diligent HUMAN copywriters, obsessed by the craft of writing high-quality and original SEO copy, for each and every one of our clients...
...we'd say a big fat, full-on, capital letter-only - 'NO, YOU NEVER DECIEVE YOUR CLIENTS BY PASSING OFF AI COPY AS YOUR OWN.'
It might surprise you to know, we're not against content creators using AI copy as a professional tool.
ChatGPT is a wonderful use of artificial intelligence, and we're really excited about where it's going to take copywriting, and what sort of opportunities are going to arise.
So, if you need to write a blog post, create a quick email, or even write a whole website, use ChatGPT by all means - we won't hate you for it.
But, if you're a professional content creator, be honest about your dirty AI secrets, and tell your clients before you write the copy that it's AI generated, so they then have the choice of whether to use you or not.
You could even offer AI copywriting/content creation as a cheaper service!
Why are we so adamant about telling the truth?
Because we love this industry, and to see it cheapened by blatant dishonesty breaks our hearts, and could give us 'honest' copywriters a bad name.
Our OpenAI Text Classifier results
With a close eye on the AI copywriting revolution, and the way the AI text being generated is now more human-like than ever, we thought we'd try out a little experiment.
We wanted to see if it's possible for someone using a copywriter could run a test to see if the copy they were getting was AI generated, and more importantly, how accurate the test was.
So, we went straight to the top and chose the text classifier created by perhaps the most famous AI model, ChatGPT, and the results were both surprising and expected.
The first piece of copy we put through the OpenAI Text Classifier was unique, 'human' copy from our main Copywriting Services page of our website www.copywritescopy.com
Check out what happened below.
Our HUMAN web copy for our copywriting services web page.
The classifier didn't do too well with that one ( this is definitely NOT AI copy) so we tried the copy on our Social Media Marketing webpage.
Our HUMAN web copy for our social media services page
'Unclear' is a bit of a cop-out, but to give it another chance, we tried our blog post about Googles 2022 algorithm update (all genuine human copy written by ourselves). This is what came back -
Our HUMAN web copy for our blog post about the Google Helpful Content update.
Better, but still hedging its bets. So, to shake things up a bit, we got on to ChatGPT and typed in 3 different prompts, including -
Write a web page for copywriting services
Write a page for social media services
Write a blog post about Googles Helpful Content update
Then, once it was written, we put each piece of purely AI generated copy through the OpenAI Text Classifier, and this is what happened -
AI generated copywriting services page
"Hmmm, this thing really needs improving," we thought, so we tried another one -
AI web copy for a social media services page.
There seems to be a pattern emerging. Each of the last two pieces of copy were definitely AI generated.
OK, one more chance. Here's an AI generated blog post about Googles Helpful Content algorithm update. Fingers crossed it recognises this as AI.
AI web copy for a blog about Googles Helpful Content update.
And again, the OpenAI Text Classifier failed to recognise AI copy generated by ChatGPT.
According to our brief experiment, it's obvious to see - improvements need to be made with the OpenAI Classifier if it's to accurately spot AI copy.
And OpenAI agrees...
Here's what they say -
"Our classifier is not fully reliable. In our evaluations on a “challenge set” of English texts, our classifier correctly identifies 26% of AI-written text (true positives) as “likely AI-written,” while incorrectly labeling human-written text as AI-written 9% of the time (false positives). Our classifier’s reliability typically improves as the length of the input text increases. Compared to our previously released classifier, this new classifier is significantly more reliable on text from more recent AI systems."
...at the speed of which AI technology is improving overall, improvements are happening, so if you are trying to pass off AI copy as your own, be afraid, you will be found out.
Try out the OpenAI Text Classifier for yourself here, and let us know what results you get.