technically, you could supply it a specific player, a model like 'Stockfish' or 'super-grandmaster'[0]

[0] https://saintlouischessclub.org/blog/how-face-super-grandmas...

It should be compared to a simpler prompt with less instruction ("what move should be played in this chess game: ..." or something). I've found that additional constraints sometimes aren't applied very well and actually result in worse output.