Perception is the way to know what AI assistants tell buyers about your brand: the sentences they repeat, the claims your own facts contradict, and the places they aren’t sure. You catch the wrong description before a customer reads it.
Four readings tell you whether the description is working for you or against you. The example workspace: net sentiment +44, the most positive of five brands; engines unsure 15%, hedging in 4 of 27 answers; 1 unsourced claim; and no missed categories, with 5 of 8 discussed.
Every sentence the engines hold about you, sorted into claims, hedges and the themes they keep coming back to. A claim is checked against the facts you approved: contradicted if your record disproves it, unsupported if your record is silent. Open any sentence to read the answers it came from.
| What the screen shows | What it is | What it’s checked against |
|---|---|---|
| Claim | A statement of fact about you: a price, a feature, a person | Your approved facts. Contradicted if they disprove it; unsupported if they don’t cover it |
| Unsure | A hedge: the engine said may, appears to, is reported to | Counted as the engines being unsure; a hedge is not filed as a claim |
| Theme | A description the engines keep using: a simpler alternative to HubSpot | How often it appeared this time against last time, and whether that changed |
Under the list, the sites the descriptions come from. One engine saying it once is an engine talking; several engines saying it is what the web says about you.
Sentiment is easy to headline and hard to act on. One real workspace read 293 positive mentions and 0 negative, and learned nothing it could do. A wrong price repeated in four answers tells you exactly what to fix. Perception leads with the sentences and keeps sentiment as one reading beside them.
The kinds of questions buyers ask about your category where you are never named, beside how many were discussed at all.
Claims are read against the facts you approved. When your record doesn’t hold the fact, the claim reads not checked, never clean.
Every reading and every theme shows its change since the last measurement, and whether it is real or within noise.
Perception reads the language in a sample of AI answers. Here is what it doesn’t do.
| It does not | What it does instead |
|---|---|
| Measure public mood | Reads the answers the engines gave to your tracked questions; sentiment is a reading of that language, not of a market |
| Headline a sentiment score | Leads on the sentences the engines repeat and the claims that contradict your facts; sentiment is one reading |
| Make an engine retract a sentence | Traces the sentence to the page that feeds it, drafts the correction, and measures whether the next answers still carry it |
| Pass a judged reading as human-validated | Publishes the judge’s agreement with hand-labelled answers on the methodology page, and marks unlabelled readings as not human-validated |
Measure your brand once, and Perception lists every sentence, checked against your facts.