The points are collinear if the line through 1 / z<sub>1</sub> and 1 / z<sub>2</sub> is parallel to the line through 1 / z<sub>1</sub> and 1 / z<sub>3</sub>. Thus, we need to show that arg[(1 / z<sub>1</sub>) - (1 / z<sub>2</sub>)] = arg[(1 / z<sub>1</sub>) - (1 / z<sub>3</sub>)]
Start by taking this statement, and showing it is the same as arg(z<sub>2</sub> - z<sub>1</sub>) - arg z<sub>2</sub> = arg(z<sub>3</sub> - z<sub>1</sub>) - arg z<sub>3</sub>.
Now, draw a diagram of a circle, passing through O and z<sub>1</sub> and z<sub>2</sub> and z<sub>3</sub>, and use it to prove (geometrically) that the above statement is true.
Edit: Note - the above proof requires the points O, z<sub>1</sub>, z<sub>2</sub> and z<sub>3</sub> to be placed in this order around the circle. This may be done without loss of generality. Alternately, if not placed in this order, then the proof needs to be extended to state that either arg[(1 / z<sub>1</sub>) - (1 / z<sub>2</sub>)] = arg[(1 / z<sub>1</sub>) - (1 / z<sub>3</sub>)] or
arg[(1 / z<sub>1</sub>) - (1 / z<sub>2</sub>)] = pi - arg[(1 / z<sub>1</sub>) - (1 / z<sub>3</sub>)]. Depending on the relative positions of O, z<sub>1</sub>, z<sub>2</sub> and z<sub>3</sub>, exactly one of these statements is true (provable geometrically), and either will result in 1 / z<sub>1</sub>, 1 / z<sub>2</sub> and 1 / z<sub>3</sub> being collinear.
Further note - can you construct a geometric proof that the line on which 1 / z<sub>1</sub>, 1 / z<sub>2</sub> and 1 / z<sub>3</sub> lie does not pass through the origin?