A modeling exercise in NLP means picking one person who does one thing noticeably well, finding out the sequence they run inside their head while they do it, and then running that sequence yourself until it either works or clearly does not. It is not admiration and it is not copying their mannerisms. The glossary entry on modeling in NLP explains where the method came from; this article is the exercise you can start on Monday.
What follows is a six-step version you can run in a week with one willing colleague or friend, the questions that actually bring a strategy to the surface, and the three ways this exercise usually fails.
| What you model | One specific skill in one specific context, not a whole person |
| What you collect | The order of their internal steps, plus what tells them to stop |
| Time needed | Two conversations of 30 minutes, then a week of testing |
| What you need from them | Permission, and patience with questions that sound strange |
| How you know it worked | You get their result in a situation where you previously did not |
What is modeling in NLP?
Modeling is the process of working out the structure behind someone’s ability so that a second person can reproduce the result. It is the method NLP was built out of, not one technique among many. Richard Bandler and John Grinder spent the early 1970s watching Virginia Satir, Fritz Perls and later Milton Erickson, and what they published in “The Structure of Magic” in 1975 was an attempt to write down what those three were doing that their own students could not explain.
The uncomfortable part is why modeling is needed at all. People who are excellent at something are usually poor witnesses to it. K. Anders Ericsson’s research on expert performance kept running into the same wall: experts give accounts of their own skill that do not match what they actually do, because the useful steps stopped being conscious years ago. So you do not ask “how do you do that” once and write the answer down. You ask in a way that recovers the sequence, and you check the answer against what you can observe.
What exactly do you model?
Four layers, in this order of usefulness. Most beginners stop at the first one, which is why their modeling produces imitation rather than ability.
- The strategy. What they do first, second, third, and what tells them they are finished. This is the part worth the most. The TOTE model gives you a frame for it: a test, an operation, a second test, an exit.
- The internal detail. Whether step one is a picture, a remembered voice, a pressure in the chest, and how big or loud or close that is. Two people can run the same steps and get different results because one of them sees it small and grey.
- The beliefs that make the strategy possible. Someone who negotiates calmly often holds “there will be another deal” without ever saying it. Hand their steps to a person who believes this is the last deal and the steps collapse.
- The physiology. Posture, breathing, pace. Easy to see and the least transferable on its own, which is why it comes last rather than first.
How do you run the modeling exercise?
Six steps. Keep the scope small enough to be embarrassing: not “how she leads”, but “how she stays level when a client raises their voice”.
- Name one outcome you want and one person who already gets it. Write it as an observable result. If you cannot describe what would be different in the room, the model has nothing to aim at.
- Ask permission and say what you are doing. “I want to learn how you do X. May I ask you some odd questions about a specific time you did it?” Modeling someone without telling them is how NLP earns its bad reputation.
- Anchor them to one concrete instance. Not “when you give feedback”, but “last Tuesday with Marta”. General questions produce theories. A single episode produces the sequence.
- Elicit the steps and write their words. Use the questions in the table below. Note the exact phrasing, not your summary of it. Your summary already contains your own strategy.
- Run it yourself, badly, within 48 hours. Take their first three steps into a real situation and notice where you fall out of the sequence. That fall-out point is the information you did not get in the interview, so go back and ask about exactly that.
- Subtract until it breaks. Drop one element and run it again. If the result holds, that element was decoration. What is left after three rounds of subtraction is the model.
Which questions pull out a strategy?
Questions about sequence and about the moment of decision, asked in the present tense so the person goes back into the experience instead of describing it from outside.
| Ask this | What it surfaces | What a useful answer sounds like |
|---|---|---|
| “What is the very first thing you are aware of?” | The trigger, and which channel it arrives in | “His voice goes flat. I notice that before anything else.” |
| “And then what happens?” | The next step, repeated until they run out | “Then I check whether I am still breathing out.” |
| “How do you know it is time to speak?” | The test that ends one step | “When the room stops feeling tight.” |
| “When you see that, where is it and how big?” | The internal detail that makes the step work | “Small, off to the left, like a note I already wrote.” |
| “What would have to be true for this not to work?” | The belief holding the whole thing up | “If I thought I had to win this one, I would push.” |
| “What do you do that you think anyone would do?” | The step they never mention because it feels obvious | “I write down the one sentence I will not say.” |
That last question earns its place. The step a person considers too obvious to mention is often the one doing the work, and it never appears in the answer to “how do you do it”. The same listening skill is at work when you pick up meta programs in ordinary conversation, where the shape of the answer tells you more than its content.
The first time I modelled properly, I spent an hour collecting a colleague’s steps for opening difficult conversations and got nothing I could use. The step I had skipped was what she did in the two minutes beforehand, which she had not mentioned because it happened before the conversation started and therefore, to her, did not count. Ask what happens before the beginning. Carolin Mallmann, Licensed Trainer of The Society of NLP, Lisbon
Where does modeling go wrong?
In three predictable places, and all three are avoidable once you know their shape.
- Modeling a person instead of a skill. The scope quietly widens from “how she handles an angry client” to “how she is”, and then nothing can be tested. Keep it to one context, one result.
- Taking the self-report at face value. If your notes read like a philosophy of leadership, you collected opinions. Steps have an order and a stopping point. Opinions do not.
- Skipping the ecology check. A strategy that works for them may cost you something they are not paying. Someone who stays calm because they genuinely do not mind losing the deal has a strategy that will not fit a person whose income depends on it. Ask what the model would cost you before you install it.
There is also a limit worth saying plainly. Modeling transfers procedures well and transfers the state behind them less reliably. You can learn the sequence of someone’s calm in an afternoon and still need months before it holds under real pressure, which is the part no written exercise shortens. It is also the main argument for practising this with a live partner who can see what you are doing and tell you what you left out. In the seven days we spend on practitioner skills in Lisbon, modeling is run in pairs for exactly that reason, in groups small enough that nobody practises unobserved.
FAQ
How long does a modeling exercise take?
Two interviews of about 30 minutes each, split by a week of testing, is enough for one narrow skill. The interview is the short part. Most of the time goes into running the sequence in real situations and going back with the questions you only think of after it fails.
Can you model someone you cannot interview?
Partly. From recordings, interviews and writing you can often recover the order of steps and the beliefs, which is how people model public figures and historical teachers. What you cannot get is the internal detail, so you end up filling that in with guesses. Treat it as a draft model and test it harder than usual.
What is the difference between modeling and copying?
Copying reproduces the visible behaviour. Modeling reproduces the sequence that produces the behaviour, then strips out whatever turns out to be unnecessary. The test is simple: a copy looks like the other person, a model gets the other person’s result in your own style.
Do you need their permission to model someone?
For anything that involves asking them questions, yes, and it also produces better data because people cooperate with a stated purpose. Observing publicly available behaviour needs no permission. Covertly interviewing someone about their inner process while presenting it as small talk is the version worth refusing.
Is modeling in NLP scientifically validated?
Not as a formal method. There is no body of controlled studies showing that NLP modeling reliably transfers skill, and the published literature on NLP as a whole is thin and mixed. What it has in its favour is overlap with better-studied work on expert knowledge elicitation, and the fact that each model you build is testable by you, on you, within a week.
Can you model yourself?
Yes, and it is the most practical use of the exercise. Take something you do well on a good day, recover the sequence from one specific occasion, and you have a procedure for the days when it does not happen by itself. Self-modeling is harder in one respect: you are the expert who cannot see their own steps, so write answers down before you judge them.
Keep exploring
- NLP coaching questions, for the wider set of questions that recover a sequence instead of an opinion.
- Meta programs examples, the most recent piece here, if you want to hear filters in a conversation before you try to model one.
Carolin Mallmann is a Licensed Trainer of The Society of NLP, trained by Richard Bandler. She has trained more than 300 people and runs English-language NLP trainings in Lisbon, in groups of no more than ten.
