An ADC has resolution, but neither accuracy nor precision without considerable externally applied "help." Internal references are convenient but difficult to "help" with their job. External references are easier to "help" but sometimes more susceptible to external influences. Vcc and a voltage divider is seldom going to produce a reference voltage capable of providing precison beyond 5 or, if you're lucky, 6 bits. An external reference, well filtered and isolated as well as possible, with a solid, low-impedance ground reference, is capable of providing the entire specified accuracy and precision. However, on its way to the converter, it can become corrupted. Beware! Internal references are less susceptible to external influences, and more susceptible to conditions within the MCU, e.g. what code is operating at the time what switching is taking place, etc. Often it's wise to enter a "quiet" loop to allow the internals to settle before taking a measurement if you're really fussy about precision. That isn't always convenient, but you have to decide what's important.

If your system, stimulated with a highly precise input level, reads within a bit or so of the know value it is fed, then it is reasonably precise. If it does that repeatedly, over time, temperature, humitidy, etc, then it is accurate.

If it reads the correct voltage +/- 20% all the time, then it is accurate, only to within 20%, but is probably reasonably precise.

If on some occasions, it reads the correct value + 25% -15% scattered 20% about a different-than-anticipated value, then there's a precision problem. Accuracy is still about 20%.

If it reads the correct voltage, sometimes to within 1 ppm and sometimes within 10000 ppm, then you don't really know whether the issue is with precision or with accuracy.

If one one day, you read the anticipated value -3% and scattered 15% on either side of that value, and on another, you read the anticipated value +10% and scattered only 5% about that value, then you have both precision and accuracy issues, (probably due to noise) and need to dig for the cause.

You can often compensate for errors in precision by calibration, in fact, self-calibration, sometimes. You can often compensate for errors in accuracy by averaging. If you have noise problems, both will probably be affected, and little can be done by either averaging or calibration until the root of the noise problem is discovered and mitigated.

RE