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

# Citations, Sources, Sentiment, Scan history and Alerts

> The analytics screens behind the headline number — which pages the AI engines read and cite, how they talk about each brand, what each scan actually collected, and which changes raise an alert.

## Citations

An AI answer names brands and cites sources. **Citations** lists every answer that cited a source for your brand or a competitor, with the question, the engine, how the source was cited, your position in the answer when you were named, and the date.

The **citation matrix** crosses sources with brands: how many citations each source gives to each tracked brand, all time. The shade is that brand's share of the source's row; your column carries the numbers.

<Note>
  Three engines say, per source, whether the answer **quoted** the page or only **consulted** it behind the scenes: ChatGPT, Google AI Mode and Microsoft Copilot. The others report only what they quoted. Bluemoon keeps the three states apart and never folds "this engine does not say" into "not cited".
</Note>

## Sources

**Sources** ranks the domains and pages the AI engines pull from when answering your questions.

* **Domains** — every domain read in the window, with its type (brand, retailer, publisher, community…), its retrievals, and, where the engine reports it, how many were cited versus consulted. Open a domain to see the pages read, which questions cited it, which engines, and which brands appear when AI cites it.
* **URLs** — the same at page level. The first time an engine cites a page, Bluemoon stores its title and summary.
* **Gap analysis** — pages the AI reads that are neither yours nor a competitor's: the ground where being present would put you in the answer.

Sources are counted **per market**: each market's answers are read on their own.

The **Retrievals** figure counts every time an engine pulled a page from a domain. **Cited** and **consulted** count only the retrievals whose engine reports which of the two it was.

## Sentiment

**Sentiment** reads the tone of every sentence that names a tracked brand, extracted on each scan.

* **How the answers split** — positive, neutral, negative, over the window.
* **Per brand** — average tone of the answers naming each brand, yours against the competitor average, with the aspects (price, quality, support…) where you lead and where you trail.
* **Per engine** — average tone of the answers naming your brand on each engine.
* **The sentences** — every sentence about a brand, with the question that produced it.
* **Google search × AI sentiment** — with Google connected, your branded Google searches grouped by what people are looking for, next to the tone of the AI answers.

Tone is extracted by an AI model from the stored answer; when nothing was extracted in the window the panel says so rather than showing a neutral score.

## Intent map

The **Intent** page connects what buyers asked, what each engine did with the question, and whether your brand was named — and recommended. It has its own guide: [Read the Intent Map](/guides/intent-map).

## Scan history

**Settings → Scan history** lists every scan, with the engines that answered, the engines that failed and why, and where the answers came from. Open a scan for the per-engine audit. The reasons you may see:

* **No AI Overview shown for these queries** — Google did not render an AI Overview for the query. An observation, not an error.
* **Rate limit or quota reached**, **Too many parallel requests** — the provider throttled the request.
* **Missing or invalid credentials**, **Account credit exhausted** — a provider account problem, surfaced rather than hidden.
* **Source timed out** — retried on the next scan.
* **No fallback available for this platform** — the primary source failed and the engine has no second route.

A failed poll is excluded from every rate: it is neither a poll nor a miss.

## Alerts

Alerts are computed on every scan and listed on the **Alerts** page and under the bell in the header. Every alert is always recorded in the app; **Settings → Notifications** decides whether it also reaches your inbox, and from which severity.

| Rule                         | Fires when                                                                                                            |
| ---------------------------- | --------------------------------------------------------------------------------------------------------------------- |
| Visibility drop              | Your mention rate over 7 days falls more than 15% below the 7 days before, on a market with enough answers to measure |
| Recommendation drop          | Engines recommend you in a smaller share of the answers that name you than the week before                            |
| Platform dark                | An engine that named you last week names you in none of this week's answers                                           |
| New competitor mention       | A competitor you track is named in an answer for the first time                                                       |
| AI Overview impressions drop | AI Overview impressions on one of your tracked questions fall more than 30% week over week                            |
| AI Overview dominance        | Google shows an AI Overview far more often than a normal result on one of your tracked questions                      |
| Trending query               | A question linked to your prompts goes from no impressions to a real volume in 7 days                                 |

The last three read Google Search Console data and stay silent until [Google is connected](/integrations/google). Rules cannot be switched off one by one.

<Note>
  A vendor failing to deliver answers is not a customer alert. When a collection route stops answering, Bluemoon's own operators are paged and the scan falls back to another route; nothing about it appears on your Alerts page. "Platform dark" is a different rule: it is about an engine that stopped **naming** you.
</Note>