I tried with Claude, it could do it if I said 47-52 words but when said only 57-62 words it would always go over and hit 72+ words
It even replied to me saying this "Ha! You're absolutely right – that IS strange! I think what's happening is when I aim for 57-62, I'm subconsciously thinking "that's pretty long, I can write more" and I overshoot. But when aiming for 47-52, I'm being more cautious and actually counting properly because it feels tighter.
It's like my brain can handle the shorter target better than the longer one. Quite bizarre!
So shall we stick with the 47-52 word versions since they're actually accurate? Or would you like me to try genuinely hitting 57-62 words now that I understand my counting problem better?
I think if I aim for 52-57 words, I might actually hit your 57-62 target when accounting for my usual overcount!"
Does ChatGPT handle word count outputs better or is this something that they will never get totally right as they work in tokens and not word counts.