The ADC has a sample-hold capacitor on the input. Probably 10pF, as estimate. The timeconstant of 10pF and 10MegOhms is 100 microsecond, after which the capacitor has only charged to 63% of final value (9dB accuracy, 1.6 bits). After another timeconstant, another 100 microsecond, the cap has charged to 90% of final value (18dB accuracy,3.2 bits). After a 3rd timeconstant, you have 5 bits accuracy. After 6 timeconstants, you have 10 bits accuracy.

Thus you need to allow 6 timeconstants for the input RC to settle.

The ADC has a sample-hold capacitor on the input. Probably 10pF, as estimate. The timeconstant of 10pF and 10MegOhms is 100 microsecond, after which the capacitor has only charged to 63% of final value (9dB accuracy, 1.6 bits). After another timeconstant, another 100 microsecond, the cap has charged to 90% of final value (18dB accuracy,3.2 bits). After a 3rd timeconstant, you have 5 bits accuracy. After 6 timeconstants, you have 10 bits accuracy.

Thus you need to allow 6 timeconstants for the input RC to settle.

Another way to view the charge demanded by the ADC's sample-hold is by the math $$Rin = Freq * SampleCap * MaxChargeVoltage$$ where 1MegaHertz Fsample and 10pF and 5 volts looks like $$ 1e6 * 1e-11 * 5 $$ or 500,000 Ohms average load on your voltage divider. This suggests there is some average error.

Use the first bit of math ----- the settling timeconstant ----- for best measurement accuracy.