transistors / area is not necessarily better, because not all transistors are equal and the other major components which are wiring and manufacturing rules can be as big a factor as transistors.
It's not uncommon for chips to have < 80% utilization because they can't be wired more densely. Cache vs logic can be very different, and even within those things the circuit design and cell libraries and design rules etc can all make huge differences. One transistor could have 50% the switching time of another, or 2x the leakage, etc. And it's not only switching and leakage but a myriad of variables involved in devices - NFET vs PFET profiles, other non-transistor devices like inductors and capacitors, voltage limits, soft error susceptibility, electromigration resilience, etc. A given process may support 9T and 7T libraries (and more). Area scaling is not all equivalent either, broadly, SRAM, logic, and analog/IO may scale at different rates.
Different performance metrics also improve at different rates, are inter-related, non-linear, etc. A new process might be able to produce a CPU that is 20% faster or use 50% less power with a 20% area reduction, or using 20% less power might allow a 50% area reduction.
Process naming is not a big conspiracy to confuse people. Yes there's a bit of marketing and investor relations involved (probably internal politics too), but if you could only use a single number to describe a process, transistors/area is not any better than a made-up "goodness" ratio vs previous processes that we have now.
# of transistors / area is the closest metric to compare different processes against, which this should be a step up from the status quo.