There is a little debate on that. You can do both, the most accurate is in the enclosure on the bottom, but it is potentially unsafe as well. The probe MUST NOT MOVE (no tape inside either hot melt glue works well for this, big blobs ) if you are doing a interior mount. The other issue is pee/poo on the probe can cause a temporary over heating. If you go this route I'd usually suggest a back up, a second t-stat on/off with one plugged into the other and the second's probe on the heat source set to hotter than is required to reach 90 interior temps, if in the event of the above the second will shut down the first on before critical temps are reached.

The probe between the heat and glass will vary based on the room temps and glass temps. It is hard to get balance and often needs a bit more adjustment especially if the room varies in temperature. If it changes a lot and large swings the temps can change a lot too and again the same condition can occur IE in a 60º room the pad is set to 110ºF to get 90 interior, if the sun comes in during the day and the room shoots up to say 75º the glass of the aquarium warms up and the next thing you know the interior is reaching 105º. Again a second t-stat will stop this.

I am sure you will get lots of do this or do that, the answer is what kind of conditions you have and which one will produce safe consistent temps. If your room varies a lot I'd go inside, fixed really well on the bottom and place something over it to prevent faeces on the probe like a bit of tile. If will cause a bit of a buffer in that spot but it is localized and not where the hide is. The other consideration in this type of set up is the probe accuracy the on/off ones typically have a potential 2ºF error up or down due to this I'd never set it higher than 91 and likely stick to 90º

You will have to figure out what will work the best in your situation. No one can tell you what will work the best for you as they are not there and do not have enough information to have an informed decision.