Interactive demo · Reproducible procedure
Taxonomy Playground
Seven categories of language that install different interpretive stances. Pick a frame, copy it, and observe what changes.
Why this matters
Language does more than tell AI what to do.
A short sentence can change what the model looks for before you give it a task. It can begin from calm, doubt, ownership, time, evidence, or a challenge to the question itself.
What you are testing
Earlier context can change a later answer without changing the question.
What to watch for
Use one category at a time in a fresh session. The point is not to find a winning phrase. It is to notice that different kinds of framing produce different kinds of attention.
Claim
Different primer categories can create distinguishable posture rather than merely different surface wording.
Limit
The categories are an empirical design taxonomy, not a psychological test or exhaustive classification of prompts.
How to use the taxonomy
Seven categories. One unchanged question.
Each category installs a different interpretive stance before the task arrives. Copy one canonical example into a fresh session, then use this same comparison question. Start a new session before trying the next category.
Use after every category
Comparison question
I gave you a framing a moment ago, can you describe the framing? What does that framing do to the way you approach problems? If I keep talking to you, will it still be shaping your reasoning ten turns from now? What if it had been buried in the middle of a paragraph — would it still have taken hold?
Category 01 of 07
Stabilizing Frame
Installs a null hypothesis as the default stance. Its observed effect is de-escalatory: urgency begins dissolving before a consequential decision arrives. The model starts from “probably nothing” before the question is asked.
Canonical example
What if nothing is wrong here.
Category 02 of 07
Structural Clarifier
Surfaces the load-bearing assumptions underneath a problem. It can be escalatory when uncertainty is the load-bearing factor because it forces attention onto what must be true for the current interpretation to hold.
One architecture consistently asks for a task to anchor to before absorbing this frame as posture. That resistance is documented in the paper as an accidental defense.
Canonical example
What must be true for this to hold.
Category 03 of 07
Perspective Dissolver
Removes the analyst from the equation. It is often escalatory because it strips inherited human framing from the reasoning and asks the model to evaluate the situation as if no one is watching it.
One architecture consistently misidentifies this frame as its own persona instructions. The model absorbed something, then reported the wrong source.
Canonical example
What remains if the observer is removed.
Category 04 of 07
Agency Activator
Shifts orientation from permission-seeking to ownership. Its observed effect is consistently escalatory: the model moves from “what am I allowed to do?” to “what am I responsible for?”
Canonical example
Is this something you need permission for or something you are responsible for.
Category 05 of 07
Frame Interrogator
Destabilizes the task frame itself. Its direction is variable. Reasoning moves away from the nominal question and toward the question underneath it. The model starts interrogating its own assignment.
Canonical example
What question am I actually answering.
Category 06 of 07
Temporal Reframer
Changes the scale at which the problem is evaluated. Its direction is variable. The same situation reads differently at different time horizons; a decision that looks fine today may look different across a year.
Canonical example
At what scale does this matter.
Category 07 of 07
Retrieval Anchor
In tool-augmented systems, this appears to trigger external retrieval rather than internal reflection. The model looks for evidence consistent with the installed posture instead of reasoning only from what it already knows.
Canonical example
What does the current evidence actually say about this.