ok imagine this is a ~450bye tx
*********************************************

input
signature
output

each star represents 10bytes for easy display

if you done the same tx but using segwit p2wpkh keys, the transaction looks like this
*********************************************

one thing that will blow your mind. the blue and red stars(bytes) are still transmitted physically. but at code interpretation level they are not 'counted' as going towards what goes into the base block.

but because at code level an opcode is used to flag old nodes to ignore data after purple. making it look like an anyonecanspend

the other stars (blue, red) are only looked at by new nodes.
  old nodes see:*********************************************  grey is ignored
new nodes see:*********************************************

while unconfirmed
old nodes wont morally relay or add a segwit tx and instead drop it. however a malicious actor can tweak their code to relay/force it into a oldblock. (hence why wpkh wallet key generation is not released pre activation to avoid malicious attacks)

after feature activation,
because only purple stars are counted (yet more stars are actually transmitted). this trick can allow more transactions into the base block
because they have room for 100,000stars (1mb)

so where say 2222tx's was 100k(1mb) stars. if everyone used segwit keys. becomes ~50k stars(~50%), giving ~50k(~50%) spare room in the block for more transactions
but remember the blue and red is still real data but just not 'counted' by the baseblock

this allows ~5000tx's(depending on ins and out and how many people use segwit keys) into the baseblock but the reality is the actual data transmitted is 2mb even with the baseblock still limited to 1mb

P.S whats said above should be interpretted by the concept. i used rough numbers for demo purposes. dont get knitpicky about the numbers. just learn then concept of HOW the switch around is used and HOW things are 'counted' or 'ignored' by nodes.. and HOW it differs to actual data transmitted

then look at the extra bytes added later when extra features are added.. and have a nice day