This exercise involves walking towards something that the dog really likes, and not letting the dog reach the object unless it walks to it on a loose leash. This is probably the hardest exercise for the dog at first, but generally the exercise in which the fastest progress is made. This is because there is something tangible that the dog is being kept from when the leash is taut, as opposed to exercise #2 where the dog is just kept from going forward towards nothing in particular.
You have a choice of what object you choose. It could be a visible pile of treats (it's important that the dog can see them from a distance - you could have him watch you or someone else place them on the ground to help him know they are there) or a favourite toy or even a person sitting invitingly on the ground. Pick a starting point about fifty paces away from the object, start walking towards the object and wait for the dog to start pulling. Keep the leash a maximum of 2 - 3 feet long for this exercise. The instant the leash goes taut, turn and walk back to the starting point with the dog. You'll find yourself returning to the start quite often at first, but soon you'll also notice that the point at which the dog starts to pull is later and later. Good leash handling is important here. The closer you get to the object, the more 'on the ball' you have to be in terms of not letting the dog get to the object when it's pulling. The last thing you want is for the dog to get a mouthful of treats after pulling towards them! The last five or six feet are the hardest. Once the dog successfully reaches the object, if it's food, they'll eat it, and that's their reward. If you used a toy, let them play with it by themselves or with you. If you used a person, allow the dog to interact in a fun way with the person.
Once the dog has the hang of what you're asking, vary the object. If you started out using a toy, switch to food or a person. Each dog will have a 'hierarchy' of what things are most important to him. Start with the least important object and work up. Some dogs are food hounds, some can't get enough of toys, some will greet people before they will greet other dogs. You know your dog best.
There may be some of you who have worked hard at teaching your dog not to pick up food they find on the ground, so you may not want to use a pile of treats as the object and undo all your hard work in that respect. If you've taught your dog a "leave it" command, then you could use a pile treats and just don't tell the dog to "leave it", which in essence gives it permission to take it, or else give the dog a cue that means he's allowed to have what's on the ground.
The above method is from Morgan Spector, but it is an adaptation of a method from Lana Mitchell. The big difference between Morgan and Lana's methods are that Lana never lets the dog reach the desired object. She doesn't ever use a pile of treats by themself, but usually a person who has treats on them or right in front of them on the ground. This way the dog never gets the treats inadvertently after pulling if your leash handling wasn't the greatest (the person sitting can cover the treats up if the dog gets to them via pulling). With her method, she walks the dog towards the person and as above, takes the dog back to the starting line each time it begins to pull. But as soon as the dog is able to take more than a few steps without pulling, the dog is rewarded with food from you at that instant (before it starts to pull). The "game" ends at this point and they go back to the starting point. In this case, going back to the start no longer becomes a 'punishment' for the dog, but an opportunity to start the game again and earn more treats. She also advocates that even when the dog finally reaches the person, that they don't give the dog the treats they have. The treats always come from the handler (you).