"Proven" to have x% of efficacy confusing a malicous bot at time t? It could be done, if you find it let me know. But nothing's 100%, and can only become less useful...

Any method of conveying information that can be understood by a person can be programmed to be understood by a bot (and many a person can't even understand without technological assistance.) Popularity of an evasive method will determine the efficacy of using a particular counter-method, so a (currently) novel method is the best (current) approach. Even the image generating, mangled-text methods are only a stopgap. Running is the only way to keep one step ahead of the spider.

This is the same essential problem with creating an effective DRM- if people can perceive information, they can create ways of harvesting and manipulating that information in new, useful, harmful, or generally unintended ways.

Context seems to me to the best method to seperate the people from the bots. But it's hard to appear official and obscure something using context... ewe (opposite of yes) Watt (<O><O> - s) (starts with joined 'nn', rhymes with green)? Any rule can be learned, avoid rules.