Our words are handles. And the images that form in our head - every aspect of them - are handles. Handles that pull on strings that are attached to "concepts" - or whatever you want to call your nodes on your neural net. And the strings are as strong as the conditioning between the concepts.
Usually we use our words without thinking on this level. I can say something like "I advise you to be cautious, since it is possible that the naja mossambica laying on the floor will eject a liquid in your direction and cause an inability to keep a map of the physical surroundings corresponding to the territory based off of the pattern of photons that hit your retina", and you'll get the idea eventually. You might have to google "naja mossambica" to find out that it's a type of spitting cobra, and internally translate the rest to "it's gonna spit in your face and make you blind!" first. But once you do, you'll come up with an image of the snake, ready to attack, and you'll think of the pain and potential damage that could be caused, and you'll get the message to stay the heck away from the snake. In this case, you have to go through a lot of steps, but you can take care of them without thinking too much about which specific associations are being pulled on.
So we can give suggestions in plain English, and they'll work. But sometimes we want to pull harder. Or more directly. Then it makes sense to optimize your word choice to make sure that they are really seeing the picture that you are painting.
If we talk in euphemisms, we just aren't going to get the same effect. I mean, if the person does a vivid translation on their own you might, but people don't want to do that - that's why we use them in the first place. If someone says "he passed away", you know to expect a funeral, but it doesn't really hit you the same way as if the event was described in detail "showing not telling", as your English teacher would say. It wouldn't be the same as if you were led to imagine every day of eternity with that person not there. Because that's what it really means - "passed away". But when you use the euphemism, you barely nudge the concepts of ''death'' and ''fucked-up-ness''.
No one is going to go through that imagination without being walked through it. Even if you wanted them to feel it, you can't just say "feel it as much as if you were there". Because that will be parsed as "how much would I feel it if I were there? A lot. Okay, feel it "a lot"". But "a lot" isn't necessarily that strongly connected.
And on the more pleasant side, saying "I really really really like her" doesn't capture the feeling either - no matter how many "reallys" you add.
Sometimes you just gotta say "get out of the way, I'll paint the picture for you" - and work in "assembly language".
NLP - the worthwhile subset at least - has to do with this. It has to do with finding the handles most strongly and directly connected to the concepts in question, and tugging on them directly.
There's more to it than painting vivid literal pictures. Get figurative - sometimes "that red hot inferno" has stronger ties to their feelings of hatred than the words "my hatred for bob". Put yourself in their shoes and see what it feels like. Verbalize the metaphors that come up and see if they hit home. Pay attention to what they are doing. What words do they lean on? What metaphors do they use? Feed it back to them. Pry into their "submodalities". What's it feel like? What color is it? Or rather, if it had a color, which color would it be? People aren't always aware of their representations, but by directing their attention, you can sometimes find some cool stuff.
You don't always have to rely on the handles that are already there though. Say you want to make them laugh. Instead of giving the 'compiler level' suggestion of "you're gonna laugh louder!", you can get a nice strong handle on laughter by eliciting the "funny!" feeling and anchoring the intensity of the feeling to a physical dimension - say by touching their arm every time you get them to laugh. Then you can get that laughter response to fire by simply touching their arm - and crank up the dial.
Once you know which handles to use - or once you have created the handles you need, you can manipulate them to get your desired response.
If you want something to seem less important, directly shrink their image of it. Or remove the color. Or change the dude's expression. Or add corny music to the scene.
If you want to undo the conditioning, run the movie in reverse to backwards condition it out.
If you want to replace on feeling with another, see one handle getting sucked away and getting replaced with the new one - until its a habit.
If you want someone to enjoy doing their homework (this time, at least), then you can run around anchoring fun and excitement to the color red - then have them see homework as red.
If you want them to make a damn decision they've been avoiding, elicit and anchor the pros, elicit and anchor the cons, then fire them simultaneously - and watch them feel them both at the same time - and make the damn decision.
When to pull handles directly
Using handles directly works even better when the person is "in trance" because of the extra attention and smaller chance of objections, but it is rather orthogonal to the trance aspect.
Handles are best used when you need stronger ties between a cue and a desired response. It's helpful when the problem is structurally simple, like phobias - hence the success of the fast phobia cure. It's not as useful when there's a surrounding mess that will rebuild the problem. In general, work top down to remove layers of messed up cognition, and if you're left with a non-ideal set of simple associations, pull some handles directly.
If someone wants to quit smoking, first get them to fully and congruently see it as a bad idea - and if they still have any cravings, then you can suggest them away by directly pulling on some handles. Or you can take your chances and remove the cravings to start with, but don't say I didn't warn you.