Is using ChatGPT cheating? How ChatGPT helped me write a blog post

In this post, I explain how I used ChatGPT to help me write another post about ChatGPT (Could ChatGPT end up raising the quality of writing and public discourse?) and discuss if this counts as “cheating”.

How I used ChatGPT to write a recent post

First, let me clarify that ChatGPT’s draft for the post in question required less editing than average. At first, I wasn’t sure if I should write this post at all, lest people think I rely on ChatGPT so heavily in all my writing. ChatGPT has also generated many drafts that I threw out and rewrote, largely from scratch. Sharing those wouldn’t be very interesting. There are other posts that I probably could have outsourced more to ChatGPT, but chose not to, because I wanted to develop my thinking through the writing process itself.

So I ended up going with this one, because it’s the best example I have (so far) of using ChatGPT in my writing, though it’s by no means the most represenative example.

My prompt

Here’s the prompt that I started with:

ChatGPT told me this was an “interesting and valid line of reasoning” (it went into some further detail that largely reiterated what I’d said more fluently). Then I asked ChatGPT to draft a blog post along the lines of what I’d said, and to make it interesting and thought-provoking. I also instructed it to use a casual tone and add in references and other examples as appropriate.

ChatGPT’s draft

Here’s what ChatGPT came up with:

Now this is pretty good. There was a fair bit of back and forth after this, which I won’t repeat in full here. I also followed up some of the references that ChatGPT produced, to check that they actually exist.

But as you can see, there are quite a few lines from ChatGPT’s very first draft that ended up verbatim or near-verbatim in my final blog post. Looking back, I’m rather surprised at just how many lines went unchanged.

Is writing with ChatGPT “cheating”?

In my opinion, no. The major exception is if you have represented in some way that the work is entirely yours, such as in an academic setting where there are strict rules on plagiarism and fair attribution. My comments in this post only concern the world outside academia.

Should writers at least specify when they have used AI in their posts?

Honestly, I expect we’ll quickly move to a world where most posts will be written with some degree of AI assistance, making such a disclaimer meaningless.

We don’t expect writers to tell us when they’ve used tools like spell-check or Grammarly, and it’s not reasonable to expect writers to disclose when they’ve used AI in a similar low-level way. Nor do I think anyone really cares if a writer uses AI very sparingly.

What if we make writers disclose how much AI they have used?

I can definitely see some merit in this idea. It seems desirable for readers to be able to know when they’re reading something written by an AI (this is partly the reason I’m writing this post at all).

Unfortunately, I think this will be very difficult to implement in practice.

Let’s put aside the idea of enforcement for now, and just try to think through a standard that ethical writers might voluntarily adhere to. How exactly does a writer disclose “how much” AI they used?

A subjective standard whereby the writer has to disclose if AI had “significant” or “more than minor” input would be incredibly hard to apply. People have very different ideas of what ambiguous terms mean, interpreting “real possibility” to mean anything from 20% to 80% chance!

An objective standard would be easier to apply — perhaps we could expect writers to disclose when they’ve used AI to write at least, say, 50% of their post?

However, I’m not sure this will help, either. Not every word, or every sentence, carries equal weight. Some words and sentences are relatively generic — they merely set the scene for what’s to come, or summarise what’s been said. Even without them, the post would still make sense. Other words and sentences are crucial. Without them, you’d have an entirely different post, or no post at all. Moreover, there’s generally a trade-off between:

• The amount of detail you put into a prompt; and
• The amount of editing you have to do afterwards to get it to reflect your vision.
  So if you give ChatGPT a very detailed prompt, you won’t have to do as much editing afterwards.

My prompt in the example above was reasonably detailed as it provided the core ideas and direction for the post. As a result, I didn’t have to edit too much. If I’d used a generic prompt like “draft a blog post about how ChatGPT will affect the quality of writing and public discourse”, I’d have had to make substantial changes to make the post reflect my views.

Ghostwriting is nothing new (in the business world)

To put this discussion about AI-generated writing in perspective, we must remember that ghostwriting is nothing new. In the business world, ghostwriting has been going on for ages.

Company executives, partners, and managers frequently sign off on memos and reports that their underlings prepared for them. Politicians give speeches written by their professional speechwriters. Even judges sometimes issue judgments their law clerks largely drafted. This is nothing new.

Ghostwriting seems less prevalent in the sphere of bloggers and influencers. I assume this is because are blogs are more personal, and are often hobbies or passion projects — there’s little point in paying someone else to do your hobby for you! However, successful bloggers and vloggers do hire writers to draft content and scripts, yet few would call this “cheating”. Seen this way, ChatGPT merely makes ghostwriting services accessible to many more people.

You’re always responsible for your posts

Now, the fact that ghostwriting has been happening for a long time doesn’t make it ethical. But here’s why I think ghostwriting is okay: When you sign off on something, you are responsible for it.

When I first started working, I felt slightly put off by how seniors seemed to “take credit” for my output. I’d think to myself, “I wrote 95% of this. They just changed a few words and signed it off!”

What I didn’t realise was that the real value lay in those few changed words and added signature. That’s what my seniors were being paid to do. They weren’t being paid to generate text. Generating text is easy. They were paid to pick up my mistakes and exercise judgment.

Contrary to what some may believe, signing off on something doesn’t mean you’re saying to the world, “I wrote this”. It means, “I agree with this piece of work enough to put my name on it”. If a partner signs off on a law opinion, they are vouching for the accuracy of the advice in it. They are legally liable for it and can be sued if the advice turns out to be wrong, They can’t turn around and say, “well, actually, my junior wrote that one”. Similarly, when a politician gives a speech, they’re responsible for every word they utter. If people criticise them for making a racist or sexist remark, they can’t just blame their speechwriter.

If anything, using a draft that ChatGPT generated is more ethical than signing off something that an underling drafted. After all, the underling is a sentient being and might feel upset if they don’t get enough credit. Whereas, as far as we know, ChatGPT does not have any feelings.

Bringing this back to blogging: even when ChatGPT helps me draft a post, the post must ultimately reflect my views.

Conclusion

To summarise:

• I’ve been using ChatGPT to help me write some of my recent blog posts;
• I don’t think that this is cheating, nor do I think writers need to make clear in every post how much ChatGPT helped with it. Ghostwriting is nothing new.
• Ultimately, the person signing off on the work has to take responsibility for what it says. For me, it means occasionally using ChatGPT generate some drafts — but still making sure that every post reflects my views.

Finally, in case you were wondering whether I used ChatGPT to write this post? No — just because this post flowed rather easily for me. I did ask ChatGPT to help me check it, but I didn’t choose to take on most of its suggestions.

Do you believe writers should give disclaimers when they’ve relied on AI? If so, what kind of disclaimers? Has your view of my earlier blog post changed now that you’ve seen how much ChatGPT drafted? Share your thoughts in the comments below!