Insert effects are isolated to the channel where the effect is inserted, while return effects can receive signal from multiple sources. So I tend to use return effects for groups of channels where I want the same effect to be applied with the exact same parameters.
If you're working with a physical mixer, the difference is very tangible -- insert effects are inserted at some point between a channel's pre-amplifier and fader. Multiple insert effects are daisy-chained, so the order matters too. If you have a compressor daisy chained with a chorus effect, then the compressed sound will be fed to the chorus effect. The combined result is then be fed the fader to be placed in the overall mix.
At some stage through the channel's pre-amp, EQ and insert point, a physical mixer will typically have a potentiometer that allows an amount of the channel's sound to be sent to a send & return effect. But the send is available to all the channels on the mixer, so you will essentially be mixing the channels not only on the faders but also on the sends. The return is then mixed mixed back just as if it was a regular channel. Bigger mixers might have multiple send & returns.
Most Digital Audio Workstation software (ProTools, Cubase, Logic etc.) basically mimic physical mixers in this regard, as does digital mixers. Typically, software mixers will allow many more send & return channels than you can find on a physical mixer.
So I would say there's no "wrong" way to go about this. It all depends on the effect you want on the final mix.
Using reverb on a send & return is a commonly used technique to achieve a "realistic" application of reverb. Our brain is finely tuned to detect differences in reverbs, so if each of your tracks had different reverb settings, it might sound too confusing or distracting to the user. But maybe that's exactly the effect you want in other cases.
I will argue it matters less on a software mixer. There are other constraints to consider though; with a hardware mixer, you might only have one reverb processor unit. Then you will obviously need to use a send & return if you wish to have (the same) reverb effect applied to multiple channels. With a software mixer, you might be limited by processor power. So if you find you've used up all CPU cycles, moving individual insert effects to shared send & return effects might be an option.