Actions and measurement

An Action groups work around a target: an owned page, planned page, earned-source page, product, category, search query or visibility prompt. Multiple findings can support the same Action.

This gives your team one place to understand the diagnosis, start Agent work and track implementation.

Choose what to work on

Open Agent → Actions. The list groups your open work:

Each group shows how many Actions it holds and its top five by priority. Choose a group’s heading to fold it, or Show all to list every Action in that group. Choose All filters to narrow the list by status, group, action type, priority, target type or source; the filters stay in the page address, so you can share the view. Choose a row to open the Action’s detail.

Every page that the Sources URL table shows as a Gap becomes a “Get mentioned elsewhere” Action after its next refresh, whatever kind of page it is.

Read the diagnosis, contributing evidence, recommended approach and what to avoid. The priority comes from CiteLadder’s deterministic rules. An Agent explanation does not replace that ranking or create new measured facts.

Choose Work on this when the Action is suitable for your team. Linked chats remain visible on the Action.

Analyze one of your pages

Choose Analyze a page above the list to work on a content page that no finding has raised yet. Pick one of the articles, guides or other content pages from your latest crawl, a topic, and the topic’s prompts the page should answer; only topics with prompts are listed, and all their prompts start selected. Generate action creates the page’s Action, or reuses it if the page already has one, and opens a new chat that reads the live page first. Nothing is charged until the page has been read.

Work with the agent on an Action for one of your content pages opens the same kind of chat.

Work on the page

Once the page is read, the chat opens as a workspace: the page on the left, and Suggestions, Outline and SEO on the right with the chat under them. On a narrow screen each of these is a tab.

• Suggestions lists the agent’s proposed changes, grouped by the passage they belong to. Text highlighted in blue has a change ready: choose Apply to make it, or Dismiss. Text highlighted in amber needs a fact from you: Add to context puts the question in the chat for you to answer. Apply all makes every ready change that doesn’t overlap another; overlapping ones are marked Choose one for you. Applied and dismissed suggestions stay listed, and Revert puts one back.
• Choose highlighted text to open its suggestion, or a suggestion to find its text on the page.
• Select text and choose Ask AI to quote it in the chat. Nothing is sent until you send it.
• You can edit the page directly; Save changes keeps your edit as a new revision.
• Outline lists the page’s headings; choose one to go to it.
• SEO shows the page’s title tag, meta description and structured data, and the checks Site Health ran when the page was read. Editing the page does not rerun those checks.
• Export copies the page as Markdown, HTML or plain text, or downloads it as a .md or .html file.

When the page is live, choose Mark implemented in the workspace to declare the revision on screen.

Understand status

A generated draft does not mean the work has shipped. Dismissal is not implementation. A later conflicting observation can move previously verified work back into measuring.

You can dismiss or reopen eligible Actions. Declared work follows its measurement lifecycle; you cannot set it to another status at will.

Declare implementation

First make the change in your website, publishing workflow or external process. Then use Mark implemented for the relevant Action and choose the day it went live: today or any day in the last 30 days. Readings from before that day are not counted.

When work came from the Agent, confirm the output revision that matches what you shipped. Work done outside CiteLadder can be declared without an Agent output.

CiteLadder uses the Action’s current findings to determine the expected checks, and the dialog shows what each finding will be measured by before you declare. Some findings, such as a product or theme with no specific prompt, have no reading that isolates your change; they are recorded but not measured automatically. An Action without a current finding cannot be declared merely to create a measurement plan.

Read the measurement result

The declaration keeps the target, baseline and expected checks tied to that implementation event. Each check is read by the evidence that can answer it:

• a Site Health check by the next crawl that reads the page;
• a prompt by the next visibility run;
• Search Console clicks per day on the page or query by each synced window that starts after the go-live day;
• a keyword gap by whether your site now appears for that search, in Search Console or in a later Keyword Research dataset.

Each check keeps its own latest reading, so a crawl and a Search Console window together can complete an Action that neither completes alone.

Review each check and what its reading is waiting for. A page check that is still waiting offers Run crawl now. A page the crawl did not read, or an incompatible window, is a reason for limited measurement, not an observed failure. New evidence is read for 30 days after the go-live day, or 90 days for a keyword check, because a new page takes longer to rank than an edited one takes to recrawl. After that, the last reading stands.

An earned Action (a cited page that lists competitors but not you) is measured on the tracked prompts whose answers cited that page. A later improvement on those prompts is evidence to review, not proof that the listing caused it.

Report progress honestly

Use “implemented” for the work your team shipped and “verified” for the checks supported by later evidence. Neither status proves that the change caused every movement in traffic or AI visibility.

For the actual draft history, see Outputs and revisions.