"I just can't talk to him." "Everything's ruined." "You always do this." "It's impossible now." Hear these sentences the way a practitioner is trained to hear them: not as facts, not as complaints, but as containers with hatches. Every one of them has deleted something, bent something, or promoted something. The hatch is where the sentence stops being a wall and opens into a door. The Meta Model is the set of questions that find it.
The three leaks
The model, built by Bandler and Grinder from watching Virginia Satir and Fritz Perls work, sorts the leaks in human language into three families:
Deletions: what got dropped. "I can't talk to him" deleted how, about what, since when, and the subject of the whole history. The world is too rich to represent fully, so the mind compresses; the complaint is usually mostly compression.
Distortions: what got bent. Mind reading ("he obviously doesn't respect me"), cause-effect ("she makes me furious"), complex equivalence ("he's late, so the promotion is off"). The bent sentence asserts connections that were never shown.
Generalizations: what got promoted to law. "You always", "nobody ever", "everything", "nothing ever works". One case, or three, stamped as the universe.
The questions are almost embarrassingly simple. Which how, exactly? Who says, according to whom? Always? Every time, including yesterday at lunch? How, specifically, does her tone cause your fury? The simplicity is the point: the operator was never hidden deeply. It was hidden in plain sight, behind the fluency of the sentence.
Why the question works when reassurance fails
Reassuring someone who says "everything's ruined" argues with the generalization, and arguing with a generalization makes it load-bearing: now it must be defended. Asking "everything? What specifically is ruined, right now, today?" does something reassurance cannot. It forces the map to be re-rendered at a scale where action is possible. "Everything" has no handle. "The client email" has a handle.
This is also why the questions feel, to the person receiving them, less like being corrected and more like being finally heard. The question retrieves what they actually meant and never got to say. The best Meta Model delivery does not sound like an interrogation; it sounds like genuine interest in the specifics, which is exactly what it is.
A worked example, honest size
"The team doesn't respect me." Run the model. Who specifically? "Mostly Rohan." Rohan doesn't respect you when, doing what? "In stand-ups, when I present timelines." What does Rohan do, exactly, that you call not respecting? "He questions the estimates." Every time? "Twice, maybe three times." So: one colleague, one meeting type, two or three specific incidents of questioning estimates. That is not a curse. That is a work problem, and work problems have next steps. The wall was never real. The compression was real.
The operating manual and the warning label
Two rules of use. One: this is a precision tool, not a personality. Run at everyone, all the time, and you become the person nobody talks to, the human red pen. The model has a sibling, the Milton Model, which does the opposite; knowing when to open and when to close is the actual skill. Two: the ethics from the consent piece apply with full force. These questions open people. Opening someone to hand them your conclusion instead of theirs is not coaching, it is installation.
Three questions, this week, whenever a heavy sentence shows up: specifically what, specifically how, specifically since when. Watch the wall crack along the exact line of the deletion. We train this until it is reflexive in the Spirit of NLP programme, because it is the single highest-frequency tool in the kit.
Related: words that open people, the origin of the model, the founder's companion post on vagueness.
Every stuck sentence is holding a door shut. The question is just the handle.