Skip to main content

What is the Action Board?

The Action Board is a kanban board of concrete GEO tasks, generated from your project’s data. Promptwatch monitors how AI represents your brand across responses, citations, content gap scores, sentiment, Reddit threads, offsite mentions, and AI crawler activity, and a GEO action agent cross-references those signals to surface the highest-impact things to do. Instead of digging through raw data, you work a short, prioritized list. Open it by clicking Actions (the lightning-bolt icon) in the left icon rail. The page is titled Actions, and the help icon next to the title opens the How do Actions work? explainer if you want the short version of this article inside the app. The Promptwatch Actions page showing the board with Suggested, To Do, In Progress, and Done columns, plus the Severity, Assignee, and Action type filters and the Create action button.

The five columns

The board view is the default. Its columns run left to right in the order an action moves:
  • Suggested: recommendations detected from recent monitoring data. This is the agent’s inbox, so everything here is waiting for your Accept or Dismiss.
  • To Do: accepted actions ready to start.
  • In Progress: actions currently being worked on.
  • Done: completed actions. The column shows the last 14 days of completions, and Show N older reveals the rest, so a long-running project doesn’t end with a column of ancient wins.
  • Dismissed: suggestions dismissed in the last two weeks. It is collapsed by default, so it stays out of the way, but nothing is lost: expand it and press Restore on a card to bring it back.
Each column header carries a count badge and a chevron that collapses the column to a thin strip. Collapse state is remembered in your browser, so if you never look at Dismissed it stays folded. An empty column reads No actions. The Board | List toggle in the page header switches to a list view: the same actions as rows (severity, title, type, assignee, date) grouped under collapsible status headers, with Suggested and Dismissed folded by default. The list is quicker to scan when a project has many actions; the board is better for moving them. Both views are covered in Working an action to done.

Filtering the board

Three multi-select filters sit in the page header, next to the view toggle. They appear once the project has at least one action, and they apply to every column at once:
  • Severity: Critical, High, Medium, Low, or All severities. Actions without a severity always stay visible, so a custom action you created without one never disappears from a filtered view.
  • Assignee: Unassigned plus the project’s members, searchable with Search members…. Viewers are not listed, because they cannot be assigned.
  • Action type: Performance, Site Health, Content Gap, Sentiment, Reddit, Setup, Untracked, Untapped, Outreach, or Custom, or All types.
Filters and the chosen view are stored in the URL, so a filtered link opens the same view for a teammate. They reset when you switch project.

Where actions come from

Three sources feed the board, and knowing which one produced an action tells you how much scrutiny it earned:
  • The discovery agent: an agent investigates your data on a weekly schedule, one focus per day: content gaps, page signals (Untracked and Untapped pages), Reddit, offsite mentions, and sentiment. It works evidence-first: it pulls your current board state, gathers fresh data with tools, drafts candidates, deduplicates against what is already on the board, and only saves actions the data justifies, capped at 5 new suggestions per run so the board stays workable. These land in Suggested for you to accept or dismiss. The agent only runs for projects with a website configured and an active or trialing subscription. Content Gap suggestions are not created while the Content Agent’s automation is on, because that pipeline already turns gaps into drafts; see Content Agent overview. The schedule and the types are in How actions are prioritized.
  • Setup actions: when a core integration is missing, a check creates a setup action programmatically, no AI involved: Upload your sitemap, Set up Crawler Logs, Set up Visitor Logs, and Connect Google Search Console. They are created when a project is set up and again at the start of each scheduled agent run, but never duplicated while one is still open. They first land directly in To Do at High severity, because everything else the agent does works better with those data sources connected. See Sitemap or GSC connection, Log ingestion, Visitor analytics, and Google Search Console.
  • You: the Create action button opens the Create custom action dialog with a Title, a Description, an optional Severity (Critical, High, Medium, or Low), and an optional Assignee with an Assign to me shortcut. There is no status field: a custom action always lands in To Do, with the type Custom. Title and description stay editable in place on custom actions only, using the Edit title and Edit description pencils in the detail dialog; severity, assignee, and status are editable on every action. The button is hidden for viewers, see Team and roles. Agent Chat and the MCP server can also create actions when you ask, either typed like the agent’s own or a free-form custom action when nothing fits, and both land in To Do. See Asking Agent Chat.

What’s on a card

Every card carries what you need to triage without opening it:
  • Type and severity: the type icon and label, plus a severity badge (Critical, High, Medium, or Low). A custom action created without a severity has no badge.
  • Title and description: the concrete fix and why it matters. The description is clamped to two lines on the card; agent-written descriptions run three paragraphs (what to do, why it matters, the GEO impact), so open the card for the full argument.
  • Created date: the month and day the action was added.
  • Buttons that depend on the column: Suggested cards carry Accept and Dismiss, Dismissed cards carry Restore. To Do, In Progress, and Done cards show the assignee’s avatar and a call-to-action button whose label depends on the type: Run content agent on a Content Gap action, View Reddit post on a Reddit action, Generate prompts on an Untracked action, View more on a custom action, and so on. Some of these open a dialog in place rather than navigate, so you can act without leaving the board.
Click the card body to open the detail dialog. The left side holds the title, the AI Recommendation (Description on a custom action), a collapsible Details section with type-specific context such as the target prompt or the cited page, and Activity with the comment box and timeline. The right column holds Copy share link, Copy recommendation, and Dismiss action icons, then Assignee, Type, Severity, Status, Added, Agent Reasoning (the evidence trail; hidden when there is none, as on setup actions and actions you create yourself), and Actions with the type’s call-to-action buttons. Close it with the X, Esc, or a click outside. A Promptwatch action detail dialog for a Content Gap action, with the AI Recommendation, Details, and Activity on the left and the Assignee, Severity, Status and Agent Reasoning on the right.

The board keeps itself honest

The board is not an append-only backlog. On each run, the agent reviews Suggested and To Do items of that day’s types that were created 7 or more days ago. When fresh data shows an action is outdated or already handled, it refreshes the title, description, severity, and reasoning in place, or dismisses it, in which case the card moves to the Dismissed column exactly as if you had dismissed it. It also avoids flooding: it checks how many open items each action type already has before adding more. New actions trigger an in-app notification for every non-viewer member, bundled per day, with a message like “3 new setup actions for Acme”. Organization owners can also receive a weekly digest email on Mondays with the new Suggested actions of the past week and the top Critical and High priorities, controlled by the Weekly action digest toggle in User Settings. The email is skipped in a week without new suggestions. See Notifications.

How to read it

Work left to right: accept or dismiss the Suggested column first, then pull the highest-severity To Do items. Every column is sorted by severity (Critical, High, Medium, Low, then no severity), and within a level newest created first, so the top card is always the agent’s best current answer to “what should I do next?” For the mechanics of moving an action through the flow, see Working an action to done.