tedster - 9:51 pm on Dec 27, 2010 (gmt 0)
I do appreciate this note from Matt Cutts - but Twitter might not have been the ideal place for it. Maybe he'll write a blog post.
Any algorithmic treatment of a complex area is bound to have some false positives out of the box. So in some sense, it always requires using the general population as a beta test bed.
I think one of the worst algorithm efforts in recent times for generating false positives was the heuristic Google tried for automatically identifying link buyers and sellers. I hope that cloaking should be an easier goal, and the pitfalls should be easier to catch during R&D.
One that concerns me is the practice of adding a welcome message to a user based on the referrer - "welcome Google user", "welcome Twitter user" and the like. An even heartier example would be the way WebmasterWorld captures the referring keyword on Google and adds highlighting to the page as a visitor aid.
I'm going over any site I work with that somehow customizes the content - by referrer, by visitor, and especially by user-agent or IP address - to make sure that googlebot and googlebot-mobile get straightforward handling, the same as the visitor with a common browser.