Almost every aircraft lost over land is found. Almost every aircraft lost in shallow or coastal water is found. The wrecks that are never recovered share a specific combination of circumstances, and it is worth being precise about what they are.
Factor one: uncertainty in the entry point
Everything downstream depends on this. If the last known position is recent and accurate, the search box is small and the aircraft is usually found. If the last position is hours old, the box grows by the aircraft's range in every direction.
AF447 had a position and a time from its ACARS messages, and was found. MH370's position was derived from satellite handshakes into a 2,000-kilometre arc, and has not been.
Factor two: depth
Depth determines what equipment can be used, how long each dive takes, and how much the search costs per square kilometre. At coastal depths, divers and simple sonar suffice. Below a few thousand metres, only a small number of autonomous vehicles in the world can work, and each survey line takes many hours.
Factor three: time to define the area
The recorder beacons transmit for a limited period — historically 30 days, now 90. If the search area is defined while they are still active, acoustic detection can find the aircraft quickly. If the definition takes longer than the batteries last, the search reverts to sonar survey, which is orders of magnitude slower.
This is the single factor that most often decides the outcome, and it is why the tracking rules introduced after MH370 matter more than any improvement to the beacons.
Factor four: seabed terrain
A debris field on flat sediment produces clear sonar returns. The same field among volcanic ridges, canyons or steep slopes can be effectively invisible — geometric shapes are lost in the acoustic clutter, and shadows fall in the wrong places.
Much of the deep ocean has never been surveyed at useful resolution, so searchers frequently do not know what the terrain is until they map it.
The historical factor
Most famous unfound wrecks are old. Amelia Earhart in 1937, Flight 19 in 1945, Star Tiger in 1948, Flying Tiger 739 in 1962, Varig 967 in 1979 — none had recorders with beacons, most had no accurate final position, and all predate the deep-ocean survey technology that exists now.
Those cases are not unsolvable in principle. They are unsolved because nobody has spent the money to run a modern survey over an old and uncertain search box, and for most of them the box is too large to justify it.
What is different now
An aircraft lost today on an oceanic route reports position every 15 minutes and transmits autonomously once its state becomes abnormal. The search box is small enough that the beacons, now lasting three months, are likely to be heard. The combination has made permanent loss substantially less likely — though the ocean has not become any shallower.