I tried using Chat GPT to proofread a story

Short answer: It didn’t go very well.

Over the past couple of months, my brother has been writing a fairly lengthy story. It’s quite an enjoyable read, but there are also many grammatical errors throughout. Now I could do another thorough read through of every sentence, activating full grammar nazi mode, and tell my brother the errors I found and how to fix them. But then I’d be sitting here for the next 20 years. Instead, I decided to try and see if the AI sensation of the year, Chat GPT, could do it for me.

I gave Chat GPT the following prompt: “I am going to show you a story with multiple chapters. Would you be able to proofread it and correct grammatical errors?” Of course, Chat GPT obliged, and I gave it the text of Chapter 1. It then spit back the presumably proofread text of Chapter 1.

I immediately realized a significant problem in doing it this way. None of the grammatical errors Chat GPT found were highlighted in any way. I’d imagine that’s probably a little beyond the capabilities of Chat GPT currently. Unfortunately, that still meant I had to scan through the text that it outputted and compare it with the original text to determine which errors it found. I didn’t look through everything, but already I noticed that Chat GPT had left some of the most glaring errors completely untouched. Run on sentences, odd dialogue spacing, weird punctuation, and some strange word choice ultimately survived the “purge.”

But this chapter had a lot of grammar mistakes. Maybe that wasn’t fair to Chat GPT. So, I randomly picked a later chapter and fed that into Chat GPT with the same prompts as before. For whatever reason, this time it inputted a list of all the errors that it found, rather than simply returning the corrected text back. This was a lot more convenient to read, but it ultimately introduced some new problems. The biggest issue was that the explanations were… weird. This was one of the fixes it suggested:

““It’s not my fault!” I exclaimed” — Add a space after the quotation mark and “I” for proper punctuation: ““It’s not my fault!” I exclaimed.”

Looking back through the original text, this sentence was in fact missing a space. But Chat GPT forgot to show this original version and displayed the fixed version both times. Meanwhile, some explanations simply didn’t make sense, like the one below (this passage had the same spacing issue as the one above):

““Done!” Nina interjected” — Add a comma after “Done” for proper punctuation: ““Done!” Nina interjected.”

In all fairness to Chat GPT, it did suggest some actually good fixes, like cleaning up some oddly worded sentences or adjusting word choice here and there. I tried adjusting the prompt to explicitly say to list out the errors like it did here, and repeat the process for Chapter 1, but it just listed out every paragraph instead.

I suppose the main thing to learn here is that Chat GPT might be good for proofreading short passages that don’t already have many grammatical errors, but longer passages like the ones I tried are better proofread with some combination of Grammarly and human overview. I don’t use Grammarly too much, but it seems to be much better at picking up these various types of grammatical errors.

That’s pretty much all I have to say for this post. I’ll probably go back to proofreading the story now.