Do you have coating (e.g. hairspray, PVA glue etc.) on top of the glass? If so, then unfortunately the IR sensor is less than excellent with that surface. The variability in the coating affects the transparency, which unfortunately varies the contribution of the reflection from the back of the glass. Using black paper under the glass reduces the reflection from underneath the glass, but unfortunately there is still a significant reflection from the underside of the glass itself.
Aside from using a different type of Z probe (for example, the Smart Effector if your printer is a delta, or DjDemonD's piezo sensor), there are a couple of potential solutions:
1. Don't use a coating on the glass. If you are printing good quality PLA, you can get excellent adhesion to clean uncoated glass if you prepare it correctly. See my blog for details. This is how I do all the prints on my Cartesian printer.
2. Use tinted (preferably black) glass so that there is no reflection from the back.
