The best method is by observation for the proper filament squish while baby steps adjusting the z offset value during a test first layer print.

The paper method will not produce a correct result.

The objective of setting the z offset is to assure extrusions “squish” well together to bond more than tangentially.

The gcode z offset can not be set with the paper method. (Or with feeler gauges.) At best this will only asymptotically get you the thickness of the piece of paper, yet that's not going to be the correct value.

(work in progress.. in the mean time see this article part of Ellis' Tuning Guide )

Click an image to expand.

Slice and print a rectangle that’s about 55x85mm and (critically) sliced with solid infill at 0 degrees (so the infill lines print parallel to the x axis) and every 10mm or so of the print manually change the z offset by +/-0.020mm until you find the correct print height that neither buckles (too low) or doesn’t bond to the plate (too high). Interpolate for in between values or for 0.010. You’ll want to recheck that for each different filament you use as it will be slightly different.

You can also use this test print — http://danshoop-public.s3-website-us-east-1.amazonaws.com/z_offset-autotest-020offsets.gcode.txt — which will automatically increase the z offset by 0.020mm as it prints about every 15mm of its Y length (with tick marks between sections), see instructions in the gcode. It takes less than 8 minutes to print and you can visually select the best test height or interpolate between two printed heights in the test, or rerun and it will continue through the next 0.020mm increments. The latest version also even runs an adaptive bed mesh for the test to be certain you’ve got a good mesh.