# Sentiment and stance

> In what tone AI answers talk about your brand, and whether they recommend it or just list it.

The **Sentiment** page answers two different questions: **in what tone** answers describe you (sentiment) and
**what role they give you** — they recommend you, list you among the options, caution against you (stance).

## Sentiment score

**A score from 0 to 100: how favorably answers talk about your brand.**

Each mention of your brand gets a tone: positive, neutral, mixed or negative. The score weighs them like this:

```
score = (positive × 100 + (neutral + mixed) × 50 + negative × 0) ÷ mentions with a tone
```

### Example

20 mentions: 12 positive, 6 neutral, 2 negative.

```
(12 × 100 + 6 × 50) ÷ 20 = 75
```

### What counts and what doesn't

- Only mentions of **your brand** in the answer text count. Competitors do not get a sentiment score.
- An appearance only in the Google map (the local pack) has no tone, so it does not go into the score.
- With no mention with a tone in the period, the score is "—".

The score **over time** is the same calculation for each day, week or month. The change next to the big number is the
difference from the period before, in score points.

## Stance

For each tracked brand in an answer, we read what role the answer gives it. There are six values:

| Stance | What it means |
|---|---|
| **Recommended** | The answer explicitly prefers it over the alternatives. |
| **Conditional** | It recommends it explicitly only for one case, need or type of customer. |
| **Listed** | It appears among the options, without an explicit recommendation. A factual list (services, address, contact) is "listed" for every brand in it. |
| **Cautioned** | The answer cautions against it or points out important drawbacks. |
| **Comparison** | It is named mainly as a reference point, to describe another brand. |
| **Context** | It is mentioned in passing: background, source, example. |

**Recommended + Conditional** make up the "recommendations" in the numbers below.

### Most often

The stance you get in the most answers. On a tie, the most favorable one wins. Below the number:
in how many answers, out of how many with a stance read.

### Standings

Your brand and the direct competitors named most often (up to 8, you always included), ordered by **recommendation
rate**:

```
recommendation rate = answers in which the brand is recommended or conditional ÷ answers with a stance read
```

The **Seen as** column is each brand's most frequent stance.

### Strongest rival

The direct competitor that answers recommend **most consistently**. Only competitors with a stance read in at least
3 answers and at least one recommendation qualify. The order is not by the plain rate, but by the lower bound of a
95% confidence interval (Wilson): 4 recommendations out of 5 answers beat 1 out of 1, because they are more certain.

### Weakest model

The AI model with the lowest sentiment score for your brand, compared with the best one. It shows only on "All models"
and only when at least two models have mentions with a tone in the period.

## How to read them

- Good sentiment with a low recommendation rate means you are described nicely, but only put on the list. For
  recommendations, what matters is content that explicitly sets you apart.
- "Cautioned" is worth reading answer by answer: the **Prompts and who gets recommended** table leads to the answers
  behind each row.

## Where it appears

- **Sentiment** — all the numbers above, the chart over time, the ranking and the table by prompt (where the score is
  colored: green from 66, yellow from 40).