I recently wrote a post about my experiments with Chat-GPT with a view to using it to generate sports match reports that could be SEO optimised and used to attract new users to the platform.
Whilst I thought that my initial effort was decent, I felt that it lacked personality and didn't really sound like me, even though it being "trained" in my writing style with some previous articles.
Despite the lack of personality, I was impressed with the output.
Note: I won't share the exact methodology that I used throughout this article because (in the unlikely event that a scammer reads somebody else's content), I don't want to give the scammers any bright ideas. @disconnect commented that this first article was identified as a 17% probability of being AI generated.
2nd Attempt: ⚽ Arsenal v. Manchester City
👆 Source + Photoshop
So I tried again, this time creating a new account (@match-report) which would help with SEO when people search for "Blah Blah Blah Match Report". This time, for the Arsenal v. Manchester City match.
On this occasion, I asked Chat-GPT to include some emojis to give it a bit of personality and I think that proved to be successful. I did encounter a number of problems generating this though, in particular with Chat-GPT's database only going up to September 2021. One specific detail being that at that time, Gabriel Jesus played for Manchester City and not Arsenal which it got confused by.
The outcome in this case was impressive once again although it did take a lot of effort to reach this point.
In this instance, the article was identified as 37% AI generated.
3rd Attempt: 🏏 India v. Afghanistan
With no new football on due to the International Break, I decided to see how we would get on with a cricket match - this time for an event that I hadn't watched.
India v. Afghanistan Match Report
This is without doubt my favourite match report of them all but it's fair to say the effort to produce it wasn't the 15 minutes that I'd hoped... it probably required more effort that it'll take me to write this post!
Whilst the emojis of the 2nd Attempt added some character, some of the feedback from my initial test was that it lacked my customary "sarcasm" which I took to mean "immense wit and humour" that embody my content 😅
So this time, I asked Chat-GPT (amongst other things) to include some "amusing metaphors" which I think that it probably went over the top with.
As I mentioned above though, this was hard work - much harder to get an accurate output despite it being fed accurate data. It would frequently get run totals wrong and obvious errors - like the chronology of events seemed to be a huge problem. With a lot of requests and corrections, I got there in the end.
The AI checker though thought that this content was more authentic, assessing that it was 100% human written (hence my reluctance to share my methodology).
😠 Pure Frustration: The End of my Experiment
Despite what I consider to be impressive output, the process to get there is infuriating to say the least. Chat-GPT makes schoolboy errors which despite requests to correct, it fails at. I've been told that Chat-GPT works better if you're nice to it but much like asking a young child to sit at the table to eat their food, patience is limited.
As you'll see below, even after being asked to correct its contradiction and it thinking that it had done so, it had simply highlighted the exact words that I had an issue with.
These kinds of errors are consistent and it would appear persistent. To correct them all using Chat-GPT would be more onerous and time consuming than simply watching the match highlights and writing my own match report, somewhat defeating the objective of my experiment and bringing an end to my use of Chat-GPT to generate match reports.
In addition, at the time of writing, none of the match reports have been indexed by Google 🤷♂️
