While I'm really not a fan of modelling communication as messages passing between sender and receiver, there's something to be said for it just for giving us some handy words.
(From memory now, and there's much more to be said about this...)
You can think about a message as encoded information arriving at the receiver through a channel. The channel has some noise associated with it.
Meaning, or knowledge, is extracted from the message by the receiver, but how much depends on things like knowledge the receiver already has, how much the receiver can presume the sender knows, how garbled the message has got in the channel, and so on.
With a long email, you (as the receiver) can pretty much assume the sender meant to send that mail. It's a pretty strong indication they were thinking of you.
But if the email was just a little thing that said "I am thinking of you", and you knew it was either sent by them OR there was a 50% chance it had been send randomly by a robot -- what could you assume then? You wouldn't be able to say with full certainly that the sender was actually thinking of you.
This turns into an example of implicature.