Let's simplify the issue by considering polar coordinates.
Points in polar coordinates are represented in the VC packages as "vectors," sums of components times fake unit vectors in a rectangular version of the polar plane. Thus, the point where r=a, theta=b has a polar representation in terms of unbarred basis vectors e_r and e_theta. This allows the polar point to be represented in this rectangular polar plane as a "position vector" in that plane, a vector from the "origin" of that plane to the "point" (r,t). This is what is meant by the "vector" r*e_r+t*e_t, where these basis vectors are the unbarred basis vectors.
That's why the VC packages also have barred basis vectors as the moving, point-wise determined basis vectors needed for vector fields. When changing coordinates in a vector field, the basis vectors also have to change. But this is not the case for "vectors" that represent points and use the unbarred basis vectors.
MapToBasis(Vector(<r,t>,polar),cartesian[x,y]) produces the Cartesian vector (and hence, the Cartesian point (r cos(t), r sin(t))) expressed with the unbarred basis vectors e_x and e_y that can be interpreted as i and j.
Thus, the help page is correct when it says just components are converted because the argument to PositionVector is never a VectorField, just a Vector that represents the points along a curve or surface.