> ## Documentation Index
> Fetch the complete documentation index at: https://docs.get3rd.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Sentiment

> Sentiment measures how AI describes your brand when it mentions you — positively, neutrally, or negatively — and surfaces the phrases behind the score.

**Sentiment** captures the tone AI assistants use about your brand. Being mentioned a lot is only good if the framing helps you — sentiment tells you whether the answer is selling for you or against you.

## How it's determined

Every response that mentions your brand is classified as **positive**, **neutral**, or **negative**, and the dashboard shows the split across those three. The analysis works at two levels:

1. **Response level.** AI analysis evaluates the overall tone toward your brand in each response.
2. **Sentence level.** Individual brand-mentioning sentences are analyzed with a vocabulary-based system — positive words ("excellent", "leading"), negative words ("terrible", "unreliable"), negation handling ("not good" → negative), and intensifiers ("very reliable" → strongly positive). Works in English, Danish, Swedish, Norwegian, and German.

<Info>
  Sentiment is **only assessed on responses where your brand is actually mentioned** — non-mentions never dilute the picture.
</Info>

## How to read it

Look at the *distribution*. A half-positive/half-negative split is a very different story from uniformly neutral coverage — the first means AI is actively arguing about you; the second means it has nothing strong to say.

Drill into negative mentions in [Conversations](/results/conversations) to find the recurring reasons: price, support, a missing feature, an outdated claim.

## How to improve it

* **Correct outdated or inaccurate claims** at their source — your own pages and the review sites AI cites.
* **Strengthen the third-party sources** AI leans on for your category. See [Authority & Outreach](/drivers/authority-and-outreach).
* **Address recurring negative themes** in your own content, so the model has a positive, current alternative to cite.

<Tip>
  Sort conversations by negative sentiment on **Must-Win prompts** first — that's where bad framing costs you the most revenue.
</Tip>

## Related

* [Understanding conversations](/results/conversations) — read the answers behind the score
* [Authority & Outreach](/drivers/authority-and-outreach) — influence the sources that shape your framing