Perception

Perception at /perception answers a different question from the rest of Trakkr. Not whether AI mentions you, but how it describes you when it does. Once a week Trakkr asks four models to describe your brand and to score it against five competitors, then keeps both the numbers and the sentences behind them.

Open it when you need to know what AI is actually saying, when a claim you do not recognize starts showing up in sales calls, or when you are deciding what a content or PR campaign should try to change. Every claim on the page carries a quote from the model that made it, so you can read the sentence that produced the read rather than trusting a label.

Perception needs a paid plan.

What a run does

A run is one batch: your brand plus the five competitors you picked, sent to ChatGPT, Claude, Gemini, and Perplexity. Each model sees the whole set at once, which is what makes the numbers comparable.

Two passes come back, and they do different jobs.

The scoring pass asks each model to rate every brand from 0 to 100 on twenty fixed dimensions. Those ratings are averaged across the four models into your dimension scores, your five category scores, and the single Perception Score at the top of the page. Because the models are told to score you relative to the others in the batch, your competitor list materially affects your numbers.

The description pass asks the same four models to describe your brand in prose, with no numbers involved. A second pass reads those descriptions back and pulls out the specific things they said, each with a short quote taken from the description it came from. That is where the Claims tab comes from.

Keep the split in mind when you read the page. The scores are the models' own ratings on a fixed scale. The claims are the models' own words. Trakkr used to print a one-word brand tone label worked out from the score thresholds; that was retired in July 2026 because no model ever said it.

Runs happen once a week, and every historical run is kept, so the trend chart and the moved-this-week ledger go back as far as the brand does. Refresh in the header pulls the latest stored run rather than starting a new one.

The twenty dimensions

Twenty metrics in five categories of four. Each is a 0 to 100 score with its own rank and trend.

Open a category card to see its four metrics. Trust splits into overall trust, reliability, transparency, and safety. Innovation splits into forward thinking, adaptability, likability, and confidence inspiring. The set is fixed for every brand so that scores stay comparable.

The headline Perception Score is the mean of all twenty. It moves slowly. A five point swing in a week is worth looking at; a one point swing is not.

The four tabs

Overview

The default view. A Perception Score ring, your rank among the tracked brands, the seven-day change, and how far you sit above or below the competitor average. Under it, one sentence of headline: the models' read on you in plain language.

Then three strips. How AI Describes You splits the claims into praised and questioned, with a right rail for what moved this week; click any claim to open its evidence. Category Scores carries the five cards, each expanding to its four metrics. At the bottom, a trend chart sits next to AI Model Breakdown, which shows how many of the four models recognized you, each model's one-line take, and a marker on any model that read live web sources rather than answering from training data.

Claims

The full board, and the tab to spend time in. It opens with the lean (mostly positive, mixed, or mostly critical), how many models know you and at what confidence, the counts of praised, questioned, and new claims, and the date of the last run.

Below that, two uncapped columns of Praised and Questioned claims, ordered by how many models said each one. Click any claim to open the evidence drawer: the actual quotes, which model said what, the live sources shaping it, and a suggested fix you can push straight to Actions.

Needs work switches the board to a worklist, one row per questioned claim: the claim, what AI actually said, the suggested move, and how often it came up. Filters narrow by sentiment, model, trend, and source type, and the CSV export carries the quotes with the rows, so a claim stays defensible away from the app.

Moved This Week sits under the board when there is movement to report: new claims, sentiment flips, and changes in whether a model recognizes you at all.

Competitors

Five sub-views over the same batch: an overview, Positioning (a perceptual map plus which brand owns each attribute), a Score Matrix of all twenty dimensions against every tracked brand, Trends, and a Head-to-Head picker for one rival at a time.

This is not the Competitors page, which counts how often each brand gets mentioned. This measures how well each brand gets described when it is mentioned. A brand can win one and lose the other.

Tracked

Goals and watched themes in one list. A theme is a subject you want followed over time. Leave the target off and it is a watch. Set a target and it becomes a goal.

Goals map your chosen words to the real dimensions behind them, so "innovative" tracks forward thinking and "premium" tracks overall quality and professional image. Each goal reads as Setting baseline, Behind, At risk, On track, or Achieved, from where the current number sits against the target. Watches carry their own score, a per-model breakdown, and a status of Active, Critical, Watching, or Resolved, and Trakkr escalates one on its own when the score falls hard or sits low.

Add a corrective message to a watch, with the phrases you want AI to repeat, and Trakkr reports what share of models now reflect it. Use it to see whether the message appears in the measured model descriptions. It does not prove the campaign caused a change. Open full view on any row goes to the theme's own page.

Setting it up and changing it

The first time you open Perception you get a three-step setup: Competitors, Goals, Confirm. Pick exactly five competitors, from the ones Trakkr already discovered or by typing your own, then one to five goal words (twelve presets including Trusted, Innovative, Premium, and Simple, or write your own). Confirming triggers the first analysis.

Both choices are editable later from Configuration in the header. Changing the competitor list changes what future scores mean, so change it deliberately rather than to make a number look better.

Acting on it

A category score drops more than five points in a week. Go to Claims and look at what is new. Compare the new claims and sentiment with the prior run, and inspect the evidence drawer. If one model moved, investigate that model first. The pattern alone does not establish the cause.

The claims read like a different company. You are "scrappy" and "startup-friendly" a year into repositioning for enterprise. That is a content problem, not a scoring problem: the models are averaging what they can find, and what they can find is out of date. Publish the new evidence, then watch for the new words to appear as new claims.

A competitor owns a dimension you need. Open Head-to-Head, find the gap, then compare claims for both brands. The word they own is usually the one you have to either reclaim or deliberately differentiate from. That is the dimension to give a target in Tracked.

Plan access

Perception is included with every paid plan. Watched themes need Scale access; the standard allowance is 10 active themes per brand. Follow the allowance shown for your account, since custom entitlements can differ. Stopping a watch frees an active slot; resuming one needs an available slot. Goals work on any paid plan.

Client portal users get Overview and Claims read-only, plus Competitors when their agency allows it. Tracked and Configuration are hidden from them.

Common questions

Why is my perception score different from my visibility score?

They answer different questions. Visibility measures whether AI names you on the category prompts you track. Perception measures how AI describes you when it is asked about your brand directly. High visibility with low perception (mentioned often, described badly) and the reverse are both common.

What exactly is quoted, and what is scored?

The twenty dimension scores, the five category scores, and the Perception Score are ratings the models gave on a fixed 0 to 100 scale. Every claim on the Claims tab is a thing a model said, carrying a short verbatim quote from its own description. If a claim shows no quote, the extraction pass could not attach one, and the row says so rather than inventing one.

Why do the four models disagree so much?

They are trained on different snapshots of the web and reason differently, and Perplexity reads live sources on top of that. A ten to fifteen point spread on one dimension is normal. A much larger spread is itself a signal, and the evidence behind the claim usually names the source driving it.

How often does perception refresh?

Once a week. Each run becomes the current snapshot, and every earlier run is kept, so the trend chart and the moved ledger show drift over months rather than only the latest read.

Does changing my competitors change my scores?

Yes, materially. The models are told to score every brand relative to the others in the batch. Drop a strong competitor and your relative position can rise without anything changing in the real world. Track the brands you actually compete with.

Why is our Value for Money score low when we are not expensive?

Because the models are averaging public discussion, not reading your pricing page. Complaint threads, review bullets that lead with "pricey", and competitor copy framing you as premium all get folded in. Claims will usually name the specific concern, with the sentence behind it.

Can I add my own dimensions?

Not to the twenty. They are fixed across all brands so scores and ranks stay comparable. For a subject of your own, add it in Tracked instead; that is the custom lane, and it is where a launch, a controversy, or a corrective message belongs.