It's probably the same as the Impressions count that was already available to the owner of the tweet.

From the existing Q&A out there, a lot of people think it's real but that it errs on the side of over counting (e.g. no deduplication, excluding owner views, etc.). But at Twitter's scale I'd imagine there's at least some extrapolation going on.

In the recording of Elon Musk on Spaces the other day, there is a bunch of conversation about views being computed off prem in Google Cloud unlike, say, the likes count which is computed in the main Twitter in prem database. Wasn’t quite clear if they’ve changed this since - I assume not.

My sense was it is pretty good but was more designed for AI use - for example the Tweet recommendation (which Musk seemed to think was really well done compared to the ad algorithm).