> ## 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.

# Actions

> The worklist of where to act so that the AI answers name this brand. The evidence behind each Action, and how its effect is measured.

An Action is one thing this brand can do to appear in more answers. It names a verb and a target: "Get listed on bestof.example", "Unblock OAI-SearchBot". The Actions page is the worklist, ranked, with the evidence behind each Action.

Actions come from the stored answers. No model gives advice here. A language model labels what kind of website a destination is, and nothing more.

## The answer at the top

The page asks whether the AI answers name this brand whenever they name a rival. It answers yes when no answer of the last 28 days names a rival without this brand. Otherwise it counts those losing answers and the open Actions.

## How the worklist is built

The worklist is built again after each daily collection. It reads the analysed answers of the last 28 days. Branded prompts stay out, as on every page.

1. **Losing answers.** A losing answer names at least one monitored competitor and does not name this brand.
2. **Destinations.** A destination is a website that losing answers cite, taken as one place to act. It is the site's domain. A subreddit, a LinkedIn page and a YouTube channel each count as a destination of their own. This brand's own site is never a destination.
3. **The gate.** A destination joins the list when losing answers cite it on at least 50% of the project's platforms. It also joins when they cite it on at least 25% of the days with answers, and on 2 days or more.

A destination that no longer passes the gate fades from the list. It comes back when it passes again.

## The kinds of Action

| Kind on the page    | What the target is                             | What to do                                            | Evidence |
| ------------------- | ---------------------------------------------- | ----------------------------------------------------- | -------- |
| Crawler             | A crawler that this brand's robots.txt blocks  | Unblock it                                            | Strong   |
| Register or ranking | A register, an association or a ranking        | Apply to it. Inclusion is by application              | Moderate |
| Best-of list        | A publisher of best-of lists                   | Pitch the list's editor                               | Moderate |
| Review site         | A marketplace or review site                   | Create a profile there                                | Moderate |
| Press               | A publication                                  | Pitch its editors                                     | Moderate |
| Publishing platform | A place such as LinkedIn, YouTube or Medium    | Publish there under this brand's name                 | Moderate |
| Rival's site        | A monitored competitor's own site              | Publish the page that its cited pages answer          | Moderate |
| Forum               | A forum or a thread, such as Reddit            | Take part, openly. A mention there is earned and slow | Weak     |
| Unclassified        | A site whose kind could not be read            | Open one page before you judge it                     | Untested |
| Opening             | An objection that answers raise against rivals | Make the case where this brand does better            | Moderate |

The evidence grade says how far the finding reaches. The [methodology](https://fullreach.ai/methodology#evidence) says what each grade rests on.

* **A Crawler Action** appears for each crawler that feeds a tracked platform and that this brand's robots.txt blocks. That is a search crawler, or a crawler that reads a page when a person asks. Its answers are the losing answers on the platforms it feeds. [Crawlability](/reference/crawlability) shows the rule.
* **The kind of a destination** comes from its host first. Forums, publishing platforms and known directories have fixed kinds. A domain that the project holds for a monitored competitor is a rival's site. For any other domain, a language model reads the titles and paths of its cited pages and labels it. A reference work, such as an encyclopedia, gives no Action. It shows as context in other Actions' evidence.

## Openings

An Opening is an objection that the [Perception](/reference/perception) answers raise against the monitored rivals. It needs at least 3 answers in the last 28 days. Only you know whether this brand does better there. So an Opening waits in **Openings to confirm** until you choose:

* **This brand does better here** moves it to the list as an Action.
* **Not for us** dismisses it.

An Opening's count is perception answers, not losing answers, so it sits apart at the end of the list. Openings need Perception, so they follow its plan gate.

## The order and the figures

The list puts Crawler Actions first. Destinations follow, most losing answers first. Confirmed Openings come last. The Open view shows the top 20, and **Show all** shows the rest. The sidebar badge counts the open Actions, up to the same number.

Each row shows:

* **The losing answers** it touches, and on how many platforms and days. An Opening shows perception answers instead.
* **The rivals** that those answers named.

The line under the title sizes the list. It says what share of the losing answers of the last 28 days cite the destinations listed. A large share means that the list explains most of what this brand loses. Answers can also name brands that nobody monitors. Those answers are no losing answers yet, and the page says how many there are.

## States

| State     | Meaning                                          | Moves                           |
| --------- | ------------------------------------------------ | ------------------------------- |
| Open      | On the worklist                                  | **Mark done** or **Dismiss**    |
| Done      | You finished it                                  | **Reopen**                      |
| Dismissed | You chose not to do it                           | **Restore**                     |
| Faded     | It no longer passes the gate. It keeps its state | It returns when it passes again |

The build never changes the state you chose. A restored Opening goes back to **Openings to confirm**.

## The evidence

A row opens its evidence:

* **Pages the answers cited**, the destination's pages behind the losing answers.
* **Appears in answers to**, the prompts.
* **Named there, where this brand was not**, the rivals.
* **Searched before answering**, the web searches the platforms ran first. Only some platforms report them. See [Searches](/reference/searches).
* **Also cited, as context**, reference pages the same answers cite.
* **Read one of them**, recent answers to open.

The evidence states what the answers did. "The answers citing this page named Rival A" is a claim about the answers. It is never a claim that the page mentions Rival A. Only a dated page read, below, says what a page printed.

## Page reads

FullReach AI reads the top 3 pages of a destination to see which names they list.

* The pages of an open destination are read once a week.
* After you mark a destination done, its pages are read every day for 14 days. After that, they are read once a week, until a read finds this brand.
* A project spends at most 100 reads a week. The pages of a destination marked done in the last days come first, then open destinations, then older done ones.
* A read obeys the site's robots.txt. It stores the names the page lists, the page's own dates and the link to its way in. It never stores the page's text.

For most kinds, the evidence shows what the page printed on the day of the read. The kinds are a register, a list, a review site, press, a forum and an unclassified site. It says how many names the page listed, and whether this brand was among them, at which place. When robots.txt refuses the read, the page says so.

## The evidence pack

For a register, a list, a review site and press, **Copy as text** gives an evidence pack to send to the publisher. It states how many answers cited the target while naming rivals and not this brand. It quotes some of the questions and gives the dated page read. It ends with this brand's Profile, when you saved one.

## The Profile

The Profile is what this brand states about itself: what it offers, who it is for, its prices with a link, and its proof. Proof is awards, rankings, numbers and clients. Edit the Profile under **Settings**, **Project**.

* Until you save it, the form shows a draft from the setup wizard's read of the website. The evidence pack carries only a saved Profile.
* [Perception](/reference/perception#which-answers-count) reads the offer, saved or drafted, to decide which company an answer describes.

## Live, and the outcome

A done Action goes **Live** when the product can see the change:

* **A destination** goes live on the first page read after "done" that finds this brand on the page.
* **A Crawler Action** goes live when the crawler audit no longer finds the block.

An Opening never goes live. The Overview's trend chart marks the day an Action was marked done and the day it went live.

### How the outcome is measured

For a destination that went live, the outcome compares two groups of prompts. It reads the 28 days before the day it went live and the 28 days after. That day itself counts in neither.

* **The prompts citing it** are the prompts whose answers cited the destination in the days before.
* **The other prompts** are every other prompt of the project.

For each group and each period, the share is the answers that name this brand, out of the answers. The outcome is the change of the first group minus the change of the second. So a rise that every prompt shares, such as a new model version on a platform, cancels out.

The outcome shows with a 95% interval. It needs 21 days with answers since the Action went live. Each of the four counts also needs at least 30 answers.

| The page says                        | Meaning                                         |
| ------------------------------------ | ----------------------------------------------- |
| Measuring                            | Not enough days with answers since it went live |
| Too few answers to measure yet       | One of the four counts is too small             |
| Not yet distinguishable from noise   | The interval holds zero                         |
| Rose or fell                         | The whole interval is above or below zero       |
| No prompt is left to compare against | Every prompt's answers cite the page            |

### An example

|                         | Before                    | After                     | Change     |
| ----------------------- | ------------------------- | ------------------------- | ---------- |
| Prompts citing the page | 60 of 300 answers, 20%    | 105 of 300 answers, 35%   | +15 points |
| The other prompts       | 400 of 2,000 answers, 20% | 440 of 2,000 answers, 22% | +2 points  |

The outcome is 15 − 2 = +13 points. Its 95% interval runs from about +5.5 to +20.5 points. The whole interval is above zero. So the page says that the share rose 13 points more on the prompts citing the page than on the other prompts.

## Where it appears

| Surface                   | Form                                                                     |
| ------------------------- | ------------------------------------------------------------------------ |
| The Actions page          | The worklist, its evidence and each outcome                              |
| Overview, the trend chart | A mark on the day an Action was marked done, and on the day it went live |
| The weekly Digest         | New Actions this week                                                    |
| MCP, `list_actions`       | The worklist with its evidence                                           |
| CSV export                | One line per Action                                                      |