I haven't implemented the best strategy at _all_ times, but this blog post[0] proposes a strategy (pick the word that minimizes the expected-size of the resultant set of possible answers). I made a User Script[1] that tells you how many words remain after each guess - it shouldn't be too hard to expand that to tell you a) what the expected answer-set-size was with your answer, and b) how that compares with the "best" answer.