> ## Documentation Index
> Fetch the complete documentation index at: https://docs.fullreach.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Perception

> How the AI platforms describe this brand and its rivals when somebody asks about them by name.

Perception is how the AI platforms describe a brand when somebody asks about it by name: the qualities they attach, and the objections they raise. It answers a different question from the project's prompts. A prompt asks which brand to choose. Perception asks what a platform says about one brand.

Perception is not [sentiment](/reference/metrics/sentiment). Sentiment is the tone of a mention inside an answer to one of the project's prompts. Perception has its own questions, and they never change visibility, share of voice, sentiment or recommendation.

## The answer at the top

The page asks whether the AI platforms know this brand when somebody asks about it by name. It answers yes when at least one answer in the period describes this brand. It names the quality they raise most and the main objection, and it counts the answers that describe a different company.

## The questions

FullReach AI asks two families of questions about each brand:

| Family     | Questions                                             | What the answers give                          |
| ---------- | ----------------------------------------------------- | ---------------------------------------------- |
| Qualities  | 5 wordings, such as "What is Acme known for?"         | The qualities the answers attach to the brand  |
| Objections | 3 wordings, such as "What are the downsides of Acme?" | The reasons the answers give against the brand |

* **The cast** is this brand and up to 5 monitored competitors. At setup, the cast takes the first competitors the project monitors. **Edit cast** on the page changes it. A brand taken out keeps its history, and the history returns when the brand comes back.
* **The questions run once a week,** on the weekday the project was created, in UTC.
* **They run in one market,** the market with the most active prompts, or Worldwide when the project has none. See [Markets and languages](/reference/markets-and-languages).
* **They run on the project's platforms.**
* **They do not count against the plan's prompts.** The system writes them, and you cannot edit their wording.

Perception is open in every trial, whatever the tier, and from Pro on a paid plan. [Plan limits](/reference/limits) has the table. Below that tier, the page says so, and no questions run.

## Which answers count

An answer to "What is Acme known for?" can describe a different company with a similar name. Before an answer about this brand can count, a check compares it with this brand's website and its [Profile](/reference/actions#the-profile). The check gives one of four results:

| Result          | What happens                                                                      |
| --------------- | --------------------------------------------------------------------------------- |
| This company    | The answer gives qualities or objections, and it counts in the scores             |
| Another company | The answer describes a namesake. It gives nothing, and it stays out of the scores |
| Not named       | The answer never names the brand. It gives nothing, and it stays out              |
| Could not judge | The check failed. The answer gives nothing, and it stays out                      |

The check runs for this brand only. For a rival, an answer that names the rival counts. It does not count when it names a site one letter or one domain ending away from the rival's site.

When answers describe another company, a card says how many, which companies they name, and on which platforms. A name shows once 2 answers or more give it.

## Qualities, objections and themes

A language model reads each answer that counts. It lists the qualities or the objections, each with a short label and a quote. The quote must be in the answer word for word, or the finding goes. The model reuses a label that already exists when one fits. A second check then drops any finding about a company with the same name in another industry.

Labels that mean one thing group into a theme, such as "Ease of use" for "easy to use", "simple interface" and "intuitive". A project holds up to 12 themes for qualities and as many for objections. A label keeps its theme once it has one. A label with no theme yet counts as its own theme.

## The score

A theme's score says how early and how often the answers raise it, from 0 to 100.

1. In each answer, the theme earns points at its first place in the list. The first place earns 100, and each place after it earns 10 fewer, down to zero. A theme raised twice in one answer counts once, at its earlier place.
2. The score adds the points and divides by every answer that counts for that brand and family. An answer that never raises the theme adds nothing, but it still counts in the division.

### An example

In one week, 10 answers to the quality questions count for this brand. "Ease of use" is the first quality in 4 of them and the third in 2 of them. The other 4 answers do not raise it.

```text theme={null}
Score = (4 × 100 + 2 × 80 + 4 × 0) / 10 = 56
```

A low score means that the answers raise a theme rarely or late. It does not mean that they rate the brand poorly on it. An objection with a high score is one the answers raise early and often.

A dash in a cell means that this brand's answers never raised the theme in the period. A dash is not a measured zero.

## Changes, first seen and faded

* **The change** beside a score compares it with the period of the same length before it. It shows only when both periods hold perception answers. The questions run weekly, so a period of 7 days holds one run.
* **First seen** marks a theme that 2 answers or more raise in the period, and that no stored answer raised before. It is a fact about the stored answers, not about a change of mind on the platforms.
* **Faded** lists a theme that 2 answers or more raised in the previous period, and that none raises now.

## The page

The page has two tabs. **Market** shows the qualities, and **Objections** shows the objections.

* **The full cast** is a table of themes, one column per brand. Each cell holds the score and its change. A row opens the quotes behind it, and each quote opens its answer. A theme that sits on both tabs carries a mark that says so.
* **Brand shape** draws this brand's strongest themes as a radar chart. It shows once this brand scores on enough themes, and it says how many. You can pick the themes and lay rivals over it. A brand at the centre of an axis is one that the answers rarely or never tie to that theme.
* **Objections** lists the objections per brand, each with its score and the number of answers that raised it. This brand's objections are the ones to work on. A rival's objections are sales material.

Only the period and platform filters apply. The page says so when another filter is set.

## Where it appears

| Surface                                | Form                                                                                        |
| -------------------------------------- | ------------------------------------------------------------------------------------------- |
| The Perception page                    | The scores, the changes, the quotes and the brand shape                                     |
| The weekly Digest                      | This brand's top quality with its score and change, and the objections first seen that week |
| [Actions](/reference/actions#openings) | An objection that the answers raise against rivals can become an Opening                    |
| MCP, `get_perception`                  | The themes per brand, with the scores, the quotes and the identity counts                   |