Agent Chat and Unified Actions for AI Search Visibility Workflows

Dashboards get ignored. A chat that can query the live prompt log does not, if someone still owns the tickets. On client programs we use Promptwatch Agent Chat as the analyst and Unified Actions as the to-do list. We still decide what ships, and we still write the slide. The chat finds the row. The row is the evidence.

Paid plans refresh daily. There are no instant alerts. The weekly habit is the workflow, and the weekly habit is what keeps the account honest. A daily refresh with no human looking at it is the same as no refresh at all.

Agent Chat in a review, not as a mascot

Agent Chat is a conversational analyst over the project's live data. It can walk prompts, citations, competitors, and traffic, run multi-step research, and draw charts in the thread. The skills library has ready jobs: weekly briefings, competitor gap analysis, citation research, content sprints. Those skills are prompts someone already wrote, not magic. They work because the data underneath them is the project's actual data.

We open Chat with a narrow question. "Which of last week's absent prompts now cite a competitor domain we already know?" is useful. "How do we win AI search?" is not. The agent is only as good as the prompt list and the Brand Book underneath it, so the first job on any account is to freeze the prompt list and the aliases before anyone opens Chat. A vague question against a messy prompt list produces a confident answer that is wrong.

We do not paste Chat output into a client deck without opening the underlying rows. The chat is a faster way to find the row. The row is the evidence, and the row is what we quote in the deck. A chart that Chat drew in the thread is a draft of a chart we still have to verify.

MCP and the REST API exist if the account already lives in Cursor or a notebook. Slack can map a channel to a project. Those are the same data, read through a different surface. Chat in the UI is what we use on most retainers because the account lead is already in the project and does not need a second tool to ask the question.

Unified Actions is the ticket list

Unified Actions is an AI-generated GEO to-do list from the data available on the project. Essential can supply visibility and citation data. On Professional, Business, or an agency plan, Actions can also use Agent Analytics crawler data. Sentiment Actions and the weekly digest email sit next to it. Some action items include social post drafts.

We treat Actions like a backlog, not like a sprint the product already prioritized for us. A crawler 403 outranks a social draft. A missing comparison passage outranks a tone tweak. We reject items that ask us to publish a page we already have, or to "create visibility" without a URL. The point of the list is that a human ranks it, and a human accepts or declines each row. If we accepted everything the list would not be a list, it would be a feed.

Content Agents can draft into Webflow or Framer from a gap. WordPress is listed as coming soon. The review inbox stays on. Autopublish stays off. Agent Chat can suggest the brief, and a human still accepts or declines. The boundary we keep is simple: the product can draft, the product can suggest, the product does not publish.

Explore (free, 10 ChatGPT prompts) is enough to show a stakeholder the chat. Essential at $95/mo is the first real prompt and citation program (50 prompts, one project), but it has no listed crawler-log allowance. Professional is $245/mo when a brand account needs Actions tied to crawler errors. Kick-off at $199/mo is how we hold several brands with crawler logs, unlimited prompts, and 10 seats. Business is $579/mo. G2: 4.7/5.

Otterly at $29 is a mention checker with Gemini as an add-on and lag. Peec at $95 will not give you this action list tied to crawler errors. Profound Starter at $99 annual is ChatGPT only. We will use a client's existing Peec export if they insist. When the brief needs Actions that can point at a crawl failure, we run Professional or an agency plan.

What we will not automate

We will not let the weekly digest become the report. The digest is a nudge, not a deliverable. We will not file every Sentiment Action as a PR crisis, because most sentiment movement is noise. We will not run a content sprint from Chat because the skill exists, because a sprint needs a brief and a brief needs a measured gap.

Engines on paid plans include ChatGPT, Gemini, Claude, Perplexity, Grok, Llama, DeepSeek, Mistral, Copilot, AI Overviews, and AI Mode. The chat can see them if the project tracks them. We still start with the four assistants the buyer named, and add the rest when the buyer agrees the channel is real.

FAQ

Does Agent Chat replace the monthly slide?

No. Chat finds the three URLs. We write the slide. Scope, dates, and prompt changes still get said in prose a client can forward.

Can the client talk to the agent without us?

Yes. We prefer they own the login. Kick-off includes 10 seats. We take one. If they ask Chat a question we would not, we see it in the next review and we correct the prompt list, not the chat history.

How we run it

Freeze prompts and aliases before anyone opens Agent Chat.
Monday: run the weekly briefing skill. Export the three actions that name a URL we control.
On Professional, Business, or an agency plan, check crawler errors before accepting a content action. On Essential, check the CDN outside Promptwatch.
Put accepted items in the CMS or the ticket tracker the client already uses. Leave social drafts in draft.
hello@1001seomedia.com if you want us to run the backlog. Bring the current Promptwatch project, or we will start one.