What? reCAPTCHA is very predictable. There is one word that always looks the same. It's the word, and then the word blurred and slightly skewed and placed on top of itself. The other is a scanned word (or somethings symbols or mathematical equation). That's the bit where they're crowd sourcing book scanning.
You don't have to get it right, and in fact you can type in gibberish if you like, as long as you get the known one right, and the known one is very easy for a human to read.
I know, I know. I like the idea of reCAPTCHA to crowdsource a very hard problem for the common good. I know how it works in and out.
It is awesome.
The only problem: things sound nice in theory do not always work well in practice. I ran into multiple occasions where reCAPTCHA presented me impossible to recognize scans, one extreme example is a printed texture pattern with not a single letter on it.
Your average, non-tech-savvy customers will never know how the system works. They shouldn't either, otherwise the whole idea of reCAPTCHA would have been defeated.
Judging by the volume of complaints about it on various support forums, it is really a bad idea to deploy it in production when you care even slightly about usability, user experience, or simply conversion rate.
You don't have to get it right, and in fact you can type in gibberish if you like, as long as you get the known one right, and the known one is very easy for a human to read.