Picture a drone on a flight-tracking screen. Every second it reports its east position x and its north position y: x=2t and y=3+t. At t=0 it is at (0,3), at t=1 at (2,4), at t=4 at (8,7). Each time gives a point, and joining the points in order traces the flight path, here a straight line. The parameter t does not appear on the graph; it is the clock that says which point comes when.
The same path can be traced in different ways. x=cost, y=sint for 0≤t≤2π runs once around the unit circle counterclockwise, starting at (1,0). x=cos2t, y=sin2t runs around it twice, at double speed, and x=sint, y=cost goes clockwise, starting at (0,1). All three satisfy x2+y2=1, and that shared Cartesian equation is exactly what eliminating t leaves you with: the starting point, direction and speed are gone.
Not every equation containing t is parametric. y=3t2 is an ordinary function that happens to call its input t; there is a single output. A parametric curve needs two coordinate functions driven by one shared parameter. Such a curve may fail the vertical-line test, looping back over the same x, because the rule is one point per moment, not one y per x.
Think of it like this. Parametric equations work like a drawing toy with two knobs, one moving the pen left and right and the other up and down, both turned as a clock ticks. The finished drawing shows the path; only the record of how the knobs turned over time shows when and how fast the pen got there.
Core idea. x=f(t), y=g(t) trace a path by time, capturing direction and speed, not just shape.