Meet the Agent

The Agent is CiteLadder’s in-app assistant. It can investigate saved project evidence, apply a skill and help you refine a plan, diagnosis, page edit or content draft.

Switch from Dashboard to Agent to open its workspace. You will find New chat, Actions, Skills, Context and your saved chat history.

Start with the work you need done

There are four useful entry points:

• New chat: describe a question and the outcome you want.
• Work on this: start from an Action, with its target and diagnosis attached.
• Ask agent: carry selected references from a supported evidence screen into a new conversation.
• Agent panel: on a Dashboard screen, open the Agent from the top bar to chat beside what you are viewing. On Search Intelligence, the rows you are viewing are attached for you; on a Site Health issue, the panel starts with a question about that issue. Choose Open in Agent to continue the same chat in the full workspace.

Before sending, inspect the composer’s reference chips. Remove anything that should not be part of the task. The Agent resolves those references against the saved data you can access.

Ask a focused question

A useful request names the target, the evidence and the deliverable.

Review this page’s confirmed Site Health findings and the attached demand signal. Prepare a short edit plan. For each recommendation, show the supporting evidence, say what is still unknown, and explain how we could check the result.

For a visibility investigation:

Explain the latest selected audit for this project. Keep brand mentions, owned citations and competitor appearances separate. Identify the saved answers that support the main findings.

You can choose a skill yourself or let the Agent select one suitable for the task.

Work beside the output

A chat has one deliverable. The conversation helps you refine it; the output pane keeps the current version available for review, editing, copying and Markdown export.

Follow-up messages revise the same deliverable. Start a new chat when you want a different kind of output or another project.

Long-form content starts with an outline. Review and explicitly approve that outline before the Agent writes the draft. See Outputs and revisions.

Know the boundary

The Agent reads saved CiteLadder evidence. It does not browse the live web, publish content, contact publishers, activate prompts, run crawls or purchase research datasets.

It also cannot declare an Action implemented on your behalf. You make the external change and then explicitly record implementation.

If a turn stops

Queued, running, cancelled, failed and stopped-at-limit are different outcomes. Use Stop while a turn is active if you no longer want it to continue.

A turn that reaches its runtime budget does not save a partial deliverable as a finished output. Inspect the status, narrow the request if needed, and start a follow-up only when you want another turn. Earlier saved revisions remain available.