AI brand perception: how each AI engine describes your brand
Showing up in an answer isn’t enough; the words matter. AI Perception measures the tone, adjectives and traits that ChatGPT, Claude, Gemini, Meta AI and Google AI Overviews attach to your brand, and compares them with your competitors.
Short answer
To find out what ChatGPT says about your brand, asking it once isn’t enough: you need to ask the questions your customers ask, many times, and read the answers together. Refractia’s AI Perception module does that across five engines and returns a sentiment score from 0 to 100 (70 or higher is positive, 40 to 69 mixed, below 40 negative), the adjectives each engine uses, whether the traits you want to convey appear with real weight, and a trait-by-trait comparison with your competitors.
What AI Perception measures
The module reads each engine’s answers to the questions in your category and pulls out how you’re described, in seven parts:
| Part | What it is | How to read it |
|---|---|---|
| Sentiment | The tone each engine uses about your brand, from 0 to 100. | 70 or higher: positive. 40 to 69: mixed. Below 40: negative. |
| Descriptors | The adjectives and terms the engines use to describe you. | Start with the negative ones that repeat: that’s what a customer reads before choosing you. |
| Trait alignment | Whether the traits you want to convey show up attached to your brand, weighted by prominence. | A trait named once among hundreds of mentions doesn’t count as achieved. |
| Vocabulary diversity | How many distinct terms the engines use relative to the total number of mentions. | Low: AI repeats a few words about you. High: the description is scattered. |
| Positioning radar | Which positioning dimensions your brand’s description leans toward. | Compare the radar’s shape with the positioning you’re after. |
| Perception matrix | For each trait, whether your brand wins, ties or loses against your competitors. | Losses on traits your customers value go to the top of the work list. |
| Narrative | A three-part summary: Strengths, Vulnerabilities and Opportunities. | The starting point for anyone who won’t read the tables. |
Why it’s measured engine by engine
Each engine builds its answer differently, so the same brand can be described well in one and poorly in another. Refractia separates two types. The chat engines (ChatGPT, Claude, Gemini and Meta AI) answer from what the model knows. Google AI Overviews summarizes what ranks on the web for that search, queried from your market’s country.
The difference changes what you do next. If Google AI Overviews describes you poorly, look at the pages it cites: Influence Sources shows which domains it uses and whether they name or link to you. If a chat model uses adjectives that no longer represent you, you’re seeing what that model holds about your brand.
If an engine didn’t answer a question, that question stays out of the calculation. Not measured is not zero, and an engine’s silence is never recorded as a neutral description.
Trait alignment: whether AI tells the story you tell
The desired traits come from Identity & DNA, the profile Refractia builds by reading your website. If your brand says it is “reliable,” “innovative” and “approachable,” alignment measures whether the engines use those ideas when they describe you.
Weighting by prominence is what makes the number useful. A binary check (“did the word appear?”) returns 100% as soon as each trait shows up once, even if it’s buried among hundreds of other mentions. Refractia weights each trait by how often it appears relative to the rest of the vocabulary, so a trait mentioned in passing adds little.
Low alignment isn’t always AI’s doing. Sometimes your own site barely talks about that trait, and the fix starts at home: GEO Optimization checks whether your pages say what you want AI to repeat.
An example with a fictional brand
Illustrative example
Almar Mattress is a brand invented for this page. The data doesn’t come from any client; it shows how to read the module.
Almar sells mattresses in the United States and wants AI to associate it with “durability,” “better sleep” and “fair price.” The scan shows:
- Sentiment of 62 in ChatGPT and 74 in Google AI Overviews: mixed in chat, positive in search.
- Most repeated descriptors: “comfortable,” “firm,” “expensive” and “slow delivery.”
- “Better sleep” is well aligned; “durability” appears in very few answers, and “fair price” collides with “expensive.”
- In the perception matrix, Almar wins on comfort, ties on materials and loses on price against its two main rivals.
The narrative sums it up: the strength is sleep, the vulnerability is delivery and the opportunity is durability, a trait the brand claims and AI doesn’t repeat yet. That sets the work: content about the mattress’s warranty and lifespan, and tracing in Influence Sources where “slow delivery” comes from.
What decisions it supports
- Messaging. Which traits AI has already picked up and which need reinforcing, with evidence per engine.
- Content. What to publish so the weak traits have something to come from. The plan turns into tasks in Strategy and content.
- Reputation. Which negative descriptors repeat and in which engines, so you can trace where they come from.
- Competition. Which traits each rival beats you on, not just who shows up more.
- Tracking. How sentiment changes across scans, for your brand and each competitor, in Evolution.
For people (leaders, candidates, executives), the module switches to a reputation framing.
“Refractia delivered an extremely insightful report that gave me many ideas for improving our digital strategy. The way it analyzes how AI perceives a brand gave me great clarity on how we’re being seen in the AI ecosystem.”
Frequently asked questions
How do I find out what ChatGPT says about my brand?
Ask it the questions your customers ask, several times and in different phrasings, and read the answers together; a single question is an anecdote. Refractia does that work in ChatGPT and four other engines, with questions in your market’s language and country.
What does a sentiment score of 55 out of 100 mean?
That the tone is mixed: AI describes you with both praise and reservations. In Refractia, 70 or higher is positive, 40 to 69 is mixed and below 40 is negative.
Why does AI describe my brand differently in each engine?
Because each engine answers from different sources. Chat engines use what the model knows; Google AI Overviews summarizes what ranks on the web for that search in your country.
Can I change how AI describes my brand?
There’s no button for that. What you can work on is what AI reads: your own pages and the third-party sources it cites. Refractia shows you which traits are missing, which sources carry weight and in which engine each thing happens.
Can it measure how AI perceives a person?
Yes. For leaders, candidates and executives, the platform has a public figures mode that switches to a reputation framing.
Keep reading
Find out which words AI uses to describe you
Your customers already ask AI about your brand; today you don’t know what it tells them. We’ll show you what ChatGPT, Claude, Gemini, Google and Meta AI say about your brand and your competitors. Free, no commitment, and we reply within one business day.