Spectral Brand Theory · measurement layer

Read the perception
you don't control

The Brand Spectrometer reads how different audiences actually perceive a brand — audience by audience, across eight dimensions, from public material — and tells you which of those differences are real and which are just noise.

Now reading cohorts

The problem

One number hides a population that disagrees

The instrument's scale

Eight dimensions of a perception

A brand is one signal made of these eight. Two audiences can hold the same overall opinion while reading those dimensions completely differently — the same score hiding two different brands. Toggle the audiences below to see it.

Each dimension carries two readings. One is how strongly it registers — the score, 0–10 (the line's height). The other is which way it leans — its valence:

−1 — hostile / negative
0 — neutral / mixed
+1 — favourable / positive

Strength is not approval: an audience can read a brand as intensely economic and feel negative about it. On the spectrum strip the dot slides from negative (left) to positive (right) along each line — a line sitting high with its dot on the left is the dangerous one: an audience that cares a lot, and not in your favour.

A real read

Five cohorts, one brand

A spectrum strip: eight wavelength lanes, one per dimension. Each audience is a line at its score (height = how strongly the dimension registers), the dot on the line is its valence (left −1 negative … right +1 positive), and the band is the spread across audiences. Valence is floored: where the lean sits below the instrument's noise floor the dot turns hollow and grey at centre — the read is sub-resolution, so the tool abstains rather than fake a direction. No radar — a spider chart's silhouette is an artifact of axis order, not the brand. Turn the detail dial up to add sentiment, uncertainty and the noise floor; each step says what it still hides. How to read this →

Detail

Who agrees · who doesn't · what's real

A map of perceptual distance

Hover any cell to see the pair, the dimensions driving their distance, and whether the gap clears its own noise floor. Flip the switch to grey out every sub-resolution pair — what remains is real.

Noise-floor gate

Hover a pair

Distance is 1 − cosine similarity between two cohorts' eight-dimension vectors. The noise floor is the wobble from swapping AI operators alone. Above the floor = real; below = we can't tell.

In this worked example, the owners ↔ online-debaters split is the largest and is resolved; several press-to-press gaps sit below the floor and are honestly reported as sub-resolution, not findings.

The read, in numbers

Every cohort, every dimension

Hover a row to highlight that cohort in the chart above. These are the raw vectors the whole read is built from.

The managerial reading

What you actually do with it

It's an instrument, not a slideshow

Try it on your own brand

Bring your own atlas — the whole read re-renders for it. Nothing is uploaded: every atlas is processed entirely in your browser.

Currently loaded —

Make one in three steps with your own AI

  1. Copy the prompt. Full guide
  2. Paste it into your AI (Claude, ChatGPT, Gemini…) together with your material — public reviews, press and posts, or your own survey, poll, or panel results. The prompt carries everything the AI needs and pulls the format itself; you don't attach anything. It hands back a ready file — or a note on what's missing.
  3. Bring it back here — paste it in, drop the file, or paste a link.

Bring the result back — pick how:

Drop your file here, or choose a file · max 2 MB

Verification equipment — the PRISM instrument family, three published measurements about instruments of this kind: what a single blended score destroys (PRISM-M); whether stated readings predict agentic choice (PRISM-C — a companion instrument; choice is not claimed by this meter); whether readings survive model-version changes (PRISM-T — the version floor readings are epoch-stamped against).

Code open-source under MIT; theory published under CC BY 4.0.

Measurement instrument for how a brand is perceived — not a verdict on what it "is." Cohort metameric variance is the measurement; there is no presumed ground truth. No brand assets reproduced (nominative use).