Run the signal through an op amp voltage follower. This will preserve the voltage and lower the source impedance headed toward the ADC.
How well does this actually work in real life?
The output impedance of the op-amp will be non-zero, rising at higher frequencies, which is important as the ADC will draw a current spike when a sample is taken. I suspect the ADC will effectively see the op-amp's open loop output impedance, which will be much lower than the potential divider, but still not zero.
I suspect is works very well as the output impedance will be on the order of a few Ohms. There's no realistic way to get that low with divider resistors. Measuring 10V on a 5V ADC with a driving impedance of 10 Ohms takes 20 Ohm resistors for the dividers. 10V / 40 Ohms = 0.25 Amps and 2.5 Watts. Not a very realistic divider.
The objective is to measure DC voltages that are likely to be well filtered. If so, we're not concerned about the response up in the MHz range.
I know there are quad op amps in DIP 14 but that is pretty large. There may be something in a smaller package, I haven't looked. Of course you still have the divider resistors but they can be pretty small and values would probably be in the few kOhm range. No appreciable heating so very small SMDs can be used.
Still, the OP has already stated that op amps aren't viable so the question is how to get a low impedance divider and what 'low impedance' is in numbers. I would think the resistors would want to be in the high hundreds or low thousands of Ohms. Since the discussion is around DC volts, a storage capacitor on the input pin will be a big help but it would have to have fairly low (and constant) leakage as it is in parallel with the lower resistor of the divider.
Where the op amp solution really shines is when you want to expand the scale around some value. We may want to measure 12V but we're really only interested in, say, 11V to 13V - a 2V range. And we want to use a 3.3V ADC. The scale and offset approach can be done with a single supply op amp and 4 resistors. We needed 2 resistors anyway but if we can sneak 2 more onto the board, we can increase resolution considerably.
Chapter 4 of "Op Amps For Everyone" - free, check Google.
FWIW, some uC ADCs spec a 2k driving impedance and a lot of people have been caught on this. It's pretty low.
What is more interesting is that the datasheet says to put a 10 ufd Tantalum at the package along with the ever popular 0.1 ufd ceramic.