That's entirely expected: A/D sampling on those MCUs connects a small capacitor to the input pin. This capacitor is usually discharged, and the switch+capacitor circuit are wideband and slew fast, in tens or at most hundreds of nanoseconds. They are nasty loads to deal with, and if you ignore the problem, you get nonlinearities and/or noise.
Whatever you use to drive this pin must be able to cope with such fast transients. If you don't have a fast op-amp driving the input, you'll need a fairly large decoupling capacitor, say 1uF at the input pin, and 50-100 Ohm isolation resistor at the output of the op-amp. If your load inherently has source impedance in the 50-100Ohm range, you won't need the isolation resistor. If the load's impedance is higher, you'll definitely need an op-amp voltage follower to isolate the ADC load from the source.
If, on the other hand, your op-amp is suitably fast, say >10MHz GBW and >40V/us slew rate, then you can use a much smaller decoupling capacitor - 0.5-2nF, and also a smaller isolation resistor - say 25-50 Ohm.
In my experience, rail-to-rail output op-amps running at small gains (<2.5), with GBW > 10MHz and slew rate >40V/us are typically suitable for "slow" ADCs that sample 100kS/s or slower, as long as they settle to your desired resolution well under 1us. The actual requirement is the aperture or "sampling window" of the A/D converter. You'll need even faster parts for faster sampling ADCs or higher resolution than 14-16 bits.
