The word you choose changes what you see
You watch a colleague present quarterly results. The numbers are down 12% from last quarter. As you listen, a word forms: "disappointing." Maybe "lazy." Maybe "unprepared." Within milliseconds, that single evaluative word has rewritten your perception of every data point in the presentation. The slides that showed incremental progress in a new market segment — you barely register them. The evaluation arrived before the observation was complete, and now you're not seeing the presentation anymore. You're seeing the label.
This is the most common failure in human cognition: evaluation outruns description. You apply a verdict before you've finished collecting the evidence. And the verdict doesn't just summarize what you saw — it actively distorts what you see next.
Descriptive language is the practice of stating what happened — what a camera would record, what a transcript would show — before attaching meaning, motive, or moral weight. It sounds simple. It is extraordinarily difficult. And mastering the sequence — describe first, evaluate second — changes how you think, how you communicate, and how accurately you perceive the world.
Observation and evaluation are different cognitive acts
Marshall Rosenberg, the psychologist who developed Nonviolent Communication (NVC), built an entire framework around a single insight: when people mix observation with evaluation, the listener hears criticism. Rosenberg argued that the first component of honest communication is pure observation — stating what you see or hear without adding interpretation. He used the analogy of a camera: a camera records facts but cannot record an evaluation, while evaluating is what your mind does after the image is captured.
The distinction sounds academic until you see concrete examples. "Doug procrastinates" is an evaluation. "Doug completed his Christmas shopping on December 24th" is an observation. "She's always late" is an evaluation. "She arrived after the scheduled start time at three of the last four meetings" is an observation. "You slammed the door" is an evaluation — "slammed" imports anger and intentionality. "When you closed the conference room door, the sound was loud enough to stop the conversation" is an observation.
The difference matters because evaluative language triggers defensiveness. When someone hears "you procrastinate" or "you were reckless," they don't process information — they defend their identity. The conversation stops being about what happened and becomes about who they are. Descriptive language keeps the channel open because there's nothing to defend against. Facts are facts. The evaluation can come later, after both parties agree on what actually occurred.
This pattern extends beyond interpersonal communication. When you evaluate your own experience prematurely — "that meeting was a waste of time," "my presentation was terrible," "this project is failing" — you close down inquiry in exactly the same way. The label replaces the data, and you stop examining what actually happened.
Your words reshape your perception
The idea that language shapes thought has deep empirical roots. In 2007, researchers Winawer, Witthoft, Frank, Wu, Wade, and Boroditsky published a study in the Proceedings of the National Academy of Sciences that demonstrated this effect with remarkable precision. Russian, unlike English, has two mandatory words for blue: "goluboy" for lighter blues and "siniy" for darker blues. English speakers use one word — "blue" — for both.
When tested on color discrimination tasks, Russian speakers were measurably faster at distinguishing two shades of blue when those shades fell across their linguistic boundary (one goluboy, one siniy) than when both shades were in the same category. English speakers, tested on identical stimuli, showed no such advantage. The effect was strongest when the colors were perceptually close — precisely the cases where language had the most work to do.
The mechanism is not metaphorical. Having a word for something literally changes how fast and how accurately you perceive it. This means every evaluative label you apply — "lazy," "brilliant," "careless," "inspiring" — is not a neutral summary of what you saw. It's an active perceptual filter that shapes what you notice next and what you fail to notice.
Alfred Korzybski, the founder of general semantics, articulated this principle decades earlier with his dictum: "The map is not the territory." Korzybski argued that humans routinely mistake their linguistic maps — their words, labels, and categories — for the territory of actual experience. When you say someone "is lazy," you've collapsed an enormously complex human being into a two-syllable map. And then you navigate by the map instead of by the territory. Korzybski's followers developed E-Prime, a variant of English that eliminates all forms of the verb "to be" used for identity ("she is lazy" becomes "she arrived late three times this week"), specifically to break the habit of confusing evaluation with reality.
The practical implication is stark: every time you reach for an evaluative word before you've finished describing what happened, you're installing a perceptual filter that will distort everything you observe afterward. Description first isn't just better communication. It's better cognition.
The engineering of blameless description
Software engineering has independently discovered the same principle through blameless postmortems — the practice of reviewing system failures without assigning personal blame. Google's Site Reliability Engineering team, which formalized this practice, enforces a specific linguistic discipline: describe what happened in the system, not what a person did wrong.
The difference in language is precise. "The on-call engineer ignored alerts" is evaluative — it assigns motive and fault. "Alert fatigue contributed to delayed response" is descriptive — it identifies a systemic condition. "Someone should have known the right settings" is evaluative. "Documentation gaps led to configuration error" is descriptive. The blameless postmortem doesn't pretend nothing went wrong. It sequences description before evaluation so that the team can understand the actual causal chain before anyone starts defending themselves.
The Center for Creative Leadership developed a parallel framework for management feedback called SBI: Situation-Behavior-Impact. SBI enforces description by breaking feedback into three discrete steps. First, describe the specific situation ("This morning at the 11 AM team meeting..."). Second, describe the observable behavior — not your interpretation of it, but the movements and words a camera would have recorded. Third, describe the impact the behavior had. The entire framework exists because decades of management research showed that evaluative feedback ("you were unprofessional," "that was sloppy work") produces defensiveness and disengagement, while descriptive feedback produces learning.
Blameless postmortems emphasize "how" questions instead of "why" questions — because "how" elicits description while "why" elicits justification. "How did the deployment happen without the review checklist?" produces a timeline of events. "Why did you skip the review checklist?" produces a defensive explanation. Same incident. Radically different information yield. The choice of one word — how versus why — determines whether you get description or evaluation in response.
Description as neurological regulation
There is a neurological reason why description before evaluation is not just better communication but better self-regulation. Matthew Lieberman's 2007 fMRI study at UCLA, published in Psychological Science, demonstrated that when participants attached descriptive emotional labels to images of angry or fearful faces, their amygdala activity decreased significantly. The act of putting an experience into descriptive words — labeling what you observe — activated the right ventrolateral prefrontal cortex, which in turn dampened the amygdala's threat response.
This is the neural mechanism behind what Lieberman called "affect labeling." When you describe what you're experiencing — "I notice my chest is tight," "I observe that my voice is getting louder," "the thought 'this isn't fair' just arose" — you shift processing from the amygdala (reactive, emotional, fast) to the prefrontal cortex (deliberate, analytical, accurate). You are literally changing which brain region drives your next response.
Crucially, the study found that generic labeling didn't produce this effect. Participants who labeled an angry face with a random name ("Harry") showed no amygdala reduction. Only emotional description — accurately naming the observed phenomenon — produced the regulatory effect. This means the precision of your description matters. Vague labels don't help. Accurate, specific description of what you actually observe is what shifts the neural pathway from reaction to regulation.
James Pennebaker's four decades of research on expressive writing converges on the same point. Across hundreds of studies, he found that people who benefit most from writing about difficult experiences are those who use cognitive processing words — "realize," "understand," "because" — rather than raw emotional venting. The mechanism is the same: externalized description transforms the experience from something that happens to you into something you can observe, name, and work with.
Using AI to practice the description-first sequence
AI provides an unusually effective practice environment for this skill because language models can mechanically separate description from evaluation in ways that make the distinction visible.
Take a statement you made today that contains an evaluation: "The client meeting was a disaster." Ask an AI to decompose it into pure observation: "The client asked seven questions that weren't addressed in the slide deck. The meeting ran 22 minutes over the scheduled time. Two action items from last month were listed as incomplete." Now you can see exactly what your evaluation compressed — and what it may have distorted. Maybe the meeting wasn't a "disaster." Maybe it was a meeting where preparation gaps became visible. Those are different situations that call for different responses.
You can also use AI to flag evaluative language you don't recognize as evaluative. Words like "interrupted," "ignored," "refused," and "dismissed" feel factual but carry strong evaluative connotations. Ask an AI to rewrite your description of a conflict using only camera-observable behavior, and you'll discover how much evaluation is embedded in language you thought was neutral.
The goal is not to outsource your observation skills to AI permanently. The goal is to use AI as a calibration tool — a mirror that shows you the gap between what you think you're describing and what you're actually evaluating. Over time, the two-second pause between observation and evaluation becomes automatic, and the descriptive version arrives first without external help.
Protocol: the description-first sequence
Step 1 — Capture the raw data. When you notice a reaction — frustration, admiration, confusion, anger — pause. Before forming a sentence about what happened, ask: what would a camera and microphone have recorded? Write that down. Only sensory data. Only observable behavior. Only verifiable facts.
Step 2 — Check for smuggled evaluation. Read your description back. Look for words that import motive, character, or judgment: "ignored," "reckless," "lazy," "brilliant," "careless," "slammed." Replace each one with the behavior it's summarizing. "He ignored my email" becomes "he has not replied to the email I sent Tuesday." "She was reckless" becomes "she deployed without completing the review checklist."
Step 3 — Sequence the evaluation explicitly. Now that you have a clean description, you may evaluate. But do it consciously and label it as evaluation: "My interpretation is..." or "The story I'm telling myself is..." This keeps the evaluation available for examination rather than embedded invisibly in the description.
Step 4 — Notice what changes. When you separate description from evaluation, you'll often find that your evaluation changes. The description reveals details that the premature label had obscured. "The meeting was a disaster" might become "the meeting surfaced three issues we hadn't identified." Same meeting. Different evaluation. The difference is that the second evaluation is grounded in described reality rather than in a reflexive label.
This sequence is not about eliminating evaluation. Evaluation is essential — it's how you make decisions, set priorities, and navigate the world. The practice is about sequencing: description first, evaluation second. Because evaluation that's grounded in thorough description is evaluation you can trust. Evaluation that outruns description is just a reflex wearing the mask of judgment.
In the next lesson, you'll examine why even your most careful descriptions are never truly neutral — because the filters that select what you notice are always active, operating before conscious observation begins.
Sources
- Rosenberg, M.B. (2003). Nonviolent Communication: A Language of Life. PuddleDancer Press.
- Winawer, J., Witthoft, N., Frank, M.C., Wu, L., Wade, A.R., & Boroditsky, L. (2007). Russian blues reveal effects of language on color discrimination. Proceedings of the National Academy of Sciences, 104(19), 7780-7785.
- Lieberman, M.D., Eisenberger, N.I., Crockett, M.J., Tom, S.M., Pfeifer, J.H., & Way, B.M. (2007). Putting feelings into words: Affect labeling disrupts amygdala activity in response to affective stimuli. Psychological Science, 18(5), 421-428.
- Korzybski, A. (1933). Science and Sanity: An Introduction to Non-Aristotelian Systems and General Semantics. International Non-Aristotelian Library.
- Pennebaker, J.W. & Chung, C.K. (2011). Expressive writing: Connections to physical and mental health. In H.S. Friedman (Ed.), The Oxford Handbook of Health Psychology. Oxford University Press.
- Beyer, B. et al. (2016). Site Reliability Engineering: How Google Runs Production Systems. O'Reilly Media.
- Center for Creative Leadership. (2019). Use Situation-Behavior-Impact (SBI) to understand intent. CCL Research and White Papers.