Can you (or anyone) supply an example where it doesn't?
I just tried a whole bunch of queries on seemingly generic-sounding sentences from old pages on niche blogs and it found every single one.
So I'd love to understand when it doesn't work and if there's some pattern -- like if certain punctuation trips it up, or if it's simply pages not indexed in the first place.
It didn't work for any query. A verbatim search was very similar to putting every word in quotes. That's very obviously not your experience, so it's either a personalization thing, or Google silently changed it. Anyway, there isn't a pattern to find, it works or it doesn't.
But then, I've seen it work too. And stop working again.
OK, so can you supply any query then? Any example will do.
This is just about verifying the behavior, that it's not finding a verbatim result that it should. And it has nothing to do with personalization -- it's easy to run these logged out.
Also, "A verbatim search was very similar to putting every word in quotes" is not an example of it not working. The question isn't whether they're similar -- your claim was that it "didn't work", which I take to mean it was returning incorrect results, that correct results were missing.
I think OP's point is that this isn't a deterministic system for a given query. The feature flags set for the particular user (are you in the verbatim or no-verbatim branch of today's AB test), and the timestamp are also inputs.
I just tried a whole bunch of queries on seemingly generic-sounding sentences from old pages on niche blogs and it found every single one.
So I'd love to understand when it doesn't work and if there's some pattern -- like if certain punctuation trips it up, or if it's simply pages not indexed in the first place.