I like the comment about the 'grid clamping effect'. Translated that means it presents a diode across the audio and bias is achieved by the charge created on the input capacitor. Unfortunately, that results in maximum anode current when there is no input and you have to be careful what voltage you present to the audio in socket.
In general, you can control the current in a tetrode or pentode stage by varying the screen grid voltage but try to keep it from going too low or the oscillator may stop running.
There is a compromise between modulating an early stage and modulating the PA. The earlier you modulate the signal the less modulating power you need, the big 'but' is that all the stages following it have to be linear which tends to make them less efficient. If your transceiver has SSB capability, the core sideband signal is probably generated by a balanced mixer and filter circuit which inherently produces low level output. That means the amplifying stages have to be linear and so it makes sense to generate low level AM and use the same amplifiers rather than use another high power audio stage to modulate the PA.
Brian.