I quickly realized that this was not a very good idea since sensor output (and consequently the ADC output) depends on the supply voltage. Although supply voltage fluctuations aren't big, they are still big enough to make sensor readings useless.
If the output rises and falls ratiometrically with the power rail then you don't need to do anything because the ADC reference voltage and input change proportionally and, this results in the correct (and constant) digital value despite changes in Vcc. It's called a ratiometric measurement and is widely used for devices like pressure sensors and strain gauges etc..
As an example consider a sensor that nominally produces +2.5 volts out when the supply rail is +5 volts. If the ADC reference is also connected to 5 volts then the digital output will be half full-scale. If the supply rose to +5.5 volts then the ADC will still produce a digital output that is half full-scale. This is what ratiometric means.
Guide to Ratiometric Sensor Applications: -
A ratiometric measurement provides a low-cost method to interface a microcontroller to a sensor. The key to an effective ratiometric measurement is to use a common excitation voltage for both the sensor and the microcontroller’s Analog to Digital Converter (ADC). As this excitation voltage naturally drifts, it impacts both sensor and ADC. However, this is mitigated by the ratiometric properties of each device. As an example, consider an excitation voltage drift from 4.9 VDC to 4.2 VDC, the ratio for both the sensor and ADC is equal. Consequently, the binary result of the ADC remains unchanged.
This led me to the idea of using some kind of voltage reference IC.
If you can use ratiometric measurements then it's unlikely to be needed.