What they did
Zyppy crawled the raw title tag on 80,959 pages across 2,370 sites, then pulled the title Google actually showed for each using site: searches. They stored both in SQL and marked every case where the displayed title differed from the one in the code.
What they found
Google rewrote 61.6% of titles. The lowest rewrite rate came at 51 to 60 characters, about 39% to 42%, while titles over 70 characters were changed 99.9% of the time. Square brackets triggered a rewrite 77.6% of the time and Google deleted the bracketed text in 32.9% of cases, against 61.9% for round parentheses. Pipes were stripped 41.0% of the time versus 19.7% for dashes.
How much weight to give it
The strength is scale and a directly observable outcome: either Google changed the title or it did not, so the core rewrite rate is hard to dispute. It is one of the few public datasets big enough to talk about title behaviour with real numbers.
What it doesn't prove
It measures what Google does, not why, and cannot say whether a rewrite helped or hurt clicks. The sample is what Zyppy could crawl, not a random slice of the web, and displayed titles shift by query and over time. Treat the percentages as directional for 2022 conditions, not fixed rules.