Draw y=x2 and pick the point (2,4). Lines through that point can cut steeply across the curve or lean lazily under it, but exactly one leans the same way the curve does at that spot. Its slope is the derivative, f′(2)=4. Point-slope form gives y−4=4(x−2), which is y=4x−4. Zoom in on (2,4) and the parabola and this line become hard to tell apart.
That closeness is what makes the line useful. At x=2.1 the tangent gives 4(2.1)−4=4.4, while the true value is 2.12=4.41, an error of only 0.01. At x=3 the line gives 8 against the true 9, so the approximation is only good near the point of contact. Because x2−(4x−4)=(x−2)2≥0, the parabola stays above this tangent everywhere, so its estimates come out slightly low.
A secant line goes through two points on the curve, and its slope is an average rate. The line through (1,1) and (3,9) on y=x2 also has slope 4, but it is y=4x−3, one unit above the tangent, and it crosses the curve twice. Matching the slope is not enough: a tangent must also pass through the point of contact. Touching only once is not the test either: the tangent to y=x3 at x=1, which is y=3x−2, meets the curve again at x=−2.
Think of it like this. A tangent line is like the beam of a car's headlights on a winding road: at each instant it points exactly the way the car is moving at that spot. Follow the beam a short way and you stay close to the road; follow it far and the road curves away. Further on, the beam may even cross the road again.
Core idea. A tangent line touches a curve at a point with the same value and the same slope, giving the best linear approximation there.