for example, if they are using twitter via a script-controlled headless (or even non-headless) browser

That's exactly it, a bot using headless client or something would "view" the ad (trigger download of the ad, etc.) but not be a real person (no eyeballs on the ad). Hence, that instance of the ad display was fake/fraud/bot. Customers of twitter (i.e. people paying twitter to show their ads) are interested in this number. Also, investors are interested in that number. Twitter says <5% of its ad-showings are fraudulent - this lets investors say "Twitter's revenue is mostly genuine". That is why the metric is reported - it's for financial/investing purposes, not a claim about how full of bots their system is.

In concrete numbers, twitter is saying there are 1 billion twitter accounts, but most of those are bots or inactive accounts or API accounts or people posting through Tweetdeck (no ads) or joke/duplicate accounts or developer test accounts or... but 230 million of the users per day use a client that shows ads. And of that subset, less than 5% are generating those ad views fraudulently.

If twitter's been lying about that (i.e. their revenue) all along to the SEC that would be a problem for twitter, though not a big enough problem for Musk's claims in his Notice yesterday to work out in his favour.