> ## Documentation Index > Fetch the complete documentation index at: https://promptwatch.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Brand narrative > Brand narrative breaks AI answers into the individual claims they make about your brand and your competitors, groups them by theme, and scores how positive they are. ### What brand narrative answers [Sentiment tracking](https://promptwatch.com/docs/academy/sentiment-tracking.md) gives each response one score for its tone toward your brand. Brand narrative goes a level deeper and shows what the answers actually say. Every claim a model makes about a tracked brand, like "Acme is affordable for small teams" or "Acme's mobile app lacks features", is pulled out of the response, filed under a theme and scored. A sentiment score of 62 tells you the tone. The claims behind it tell you that AI praises your support and doubts your mobile app. Claims are collected for your competitors too, so you can see which parts of the story a competitor tells better than you do. You'll find the page under **Visibility** in the left rail, in the **Brand perception** group, then **Brand narrative**. It is included on every plan. ### How claims are collected Every response to one of your tracked prompts that mentions a tracked brand, yours or a competitor's, is read for claims about that brand. Each claim keeps the exact sentence it came from, so you can always check it against the response. A few rules keep the list clean: * **Quotes are verbatim**: a claim is kept only if its quote appears word for word in the response and is at least four words long. A claim the model can't back with a quote is dropped. * **The brand must be named**: a claim has to be about a brand you track. Claims about brands you don't track are left out. * **No prices or shopping cards**: claims that state a price, and fragments of shopping product cards, are dropped because they change too often to be a useful narrative. * **Repeated claims are grouped**: when different responses make the same point in different words, they are grouped into one claim. The **Claim** column shows the earliest wording, and the counts show how often AI repeats it and which models say it. ### Themes Each claim is filed under one of nine themes: * **Pricing**: price, cost, discounts, free tiers, value for money. * **Features**: a named feature, integration or capability. * **Quality**: reliability, speed, durability, how well it works. * **Usability**: ease of use, workflow, the interface. * **Range**: the size or breadth of the catalogue. * **Support**: customer service, support and delivery. * **Reputation**: ratings, reviews, awards, market standing and overall verdicts like "the best CRM for startups". * **Audience**: who the brand is for, or which use it suits best. * **Company**: facts about the business itself: founding, ownership, headcount, where it operates. A grouped claim shows the theme most of its quotes fall under. ### How sentiment is scored Every quote gets a score from 0 to 100. Below 40 counts as negative, 60 and up as positive, and anything in between as neutral. A grouped claim takes the average of its quotes, and its label is **Mixed** when its quotes fall into more than one of those bands. A theme's score is the average over every quote in that theme for one brand, so claims AI repeats often weigh more than claims it made once. 50 is neutral: a theme at 80 is mostly praise, a theme at 30 is mostly criticism. ### Reading the page The filter bar at the top (monitors, models, prompt types and date range) applies to everything on the page. **The radars** at the top compare brands at a glance. **Brand sentiment by theme** plots each brand's score per theme, and **Brand sentiment by model** plots each brand's score per AI model across all themes. The **Brands** list on the right picks which brands are drawn: by default your own brand and the three most mentioned competitors. Hover a brand to highlight it, and use **Reset** to go back to the default. The radar scale fits the brands you show, so small differences stand out; read the numbers in the tooltips rather than the size of the shape. A radar needs at least three themes, or claims from at least three models, before it can draw. ![The Promptwatch Brand narrative page with the theme and model radars and the Brands list.](https://mintcdn.com/promptwatch/bXNPJu4CsP2kwZvb/academy/images/brand-narrative-1.png?fit=max\&auto=format\&n=bXNPJu4CsP2kwZvb\&q=85\&s=4fc3929fc9f94ab88d584217ab2be623) **Brand sentiment by competitor** sits below the radars, with two views: * **Benchmark**: one row per theme your brand has claims on, with your best-scoring theme first. Each row places your brand and every competitor on a strip from negative to positive, with a dashed line at the neutral 50. **Rank** is where you place among them on that theme, and **Vs. best competitor** shows the top competitor and the gap between your score and theirs. * **Heatmap**: every tracked brand against every theme in one grid, colored from red to green. Hover a score for its positive, neutral and negative counts and the claim AI says most. Click a theme row in Benchmark, or a cell in the Heatmap, to filter the claims table to it. Click it again to clear the filter. ![The Brand sentiment by competitor benchmark, one row per theme, above the claims table.](https://mintcdn.com/promptwatch/bXNPJu4CsP2kwZvb/academy/images/brand-narrative-2.png?fit=max\&auto=format\&n=bXNPJu4CsP2kwZvb\&q=85\&s=42638741ffe6ec7504624d3ee40c1964) ### The claims table The table lists the claims behind the scores, one row per grouped claim, with the most negative claims first. The tabs split it by sentiment (**All**, **Negative**, **Mixed**, **Neutral**, **Positive**) with a count on each. Search matches the claim text, and the **Competitors** and theme filters narrow it to particular brands and themes. Each row shows the claim, its average sentiment, the models that said it, when it was last seen and its theme. Expand a row to read the individual quotes. Click a row to open the claim with four tabs: * **Claims**: every quote grouped into this claim. * **Responses** and **Prompts**: the answers that contain it and the prompts that produced them. * **Citations**: the pages the responses cited alongside the claim. A cited page is not necessarily where the claim came from, but it's the first place to look. ![A negative claim opened in its detail view, with the quotes grouped into it.](https://mintcdn.com/promptwatch/bXNPJu4CsP2kwZvb/academy/images/brand-narrative-3.png?fit=max\&auto=format\&n=bXNPJu4CsP2kwZvb\&q=85\&s=b91dbf07dcb6c148cf19b1a0590374be) ### Turning claims into action 1. **Start with the Negative tab**. The most negative claims are at the top. Open one to see how many responses repeat it before you act on it: a claim AI makes daily matters more than a harsh one it made once. Use the theme filter to see whether they cluster around one part of your offer. 2. **Find the themes where competitors win**. In Benchmark, any theme where you don't rank #1 shows who leads it and by how much. Click the theme to read what AI says about you there. 3. **Separate wrong from unwelcome**. A negative claim can be fair criticism or plain wrong. When it's wrong or out of date, mark it in [Fact check](https://promptwatch.com/docs/academy/fact-check.md) to find out whether your own site or the pages AI cites are the source. 4. **Fix the sources**. Open the **Citations** tab of a claim to see which pages feed it. Outdated pages on your own site are yours to update; third-party pages are outreach targets, as covered in [Offsite mentions](https://promptwatch.com/docs/academy/offsite-mentions.md). 5. **Come back after the fix**. Claims are collected from new responses as they come in, so set the date range to the weeks after a change and check whether the claim fades. The Agent Chat panel works on this page too. Ask questions like "Where do competitors get better sentiment?" and it answers from the radars, benchmark and claims on screen. See [Asking Agent Chat](https://promptwatch.com/docs/academy/agent-chat.md). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.