A better criteria is: what fraction of people does it stop from visiting the source (because they got the information they were looking for from the summary). Yes, yes that system has it's own problems, but it is more holistic that string.length().
Determining that fraction would require the search engine to run a series of A/B tests with varying preview lengths, and keep detailed records of test results. That seems like an unreasonable burden.
A better criteria is: what fraction of people does it stop from visiting the source (because they got the information they were looking for from the summary). Yes, yes that system has it's own problems, but it is more holistic that string.length().