Actions

Actions at /actions is where you decide what to do about your AI visibility and then do it. Pages tells you which URL has a problem, Actions is where the work on it gets chosen, run and shipped, and Results says what the change earned.

It keeps two facts apart. Did the work ship? is one question. Did it earn anything? is a different one, answered weeks later by measurement. Work that shipped and moved nothing is a real, visible outcome.

One list, one page per piece of work

Actions shows your work in a short list, grouped by what it waits on. Open a row to read that work's own page. Two doors under the list open Ideas and Results. There is no tab bar and no stats strip.

A row shows its title, one line about it, and where it stands: With you, Not started, or the day of its measuring window.

A work page opens with the verdict and the first step, then the evidence behind it, the change or draft if there is one, What to do as numbered steps, and Activity. What you see depends on where the work stands: a change waiting for approval, a piece of your work in progress, an idea, work being measured, or a result.

The decision area carries at most two verbs. The secondary is usually Decline or Later; the primary is chosen for that work, so it says what the click actually does rather than a generic "Start". Measuring work shows a sentence instead, because there is nothing to decide.

The list has a keyboard: j and k move between rows, Enter opens one, x dismisses (with a reason), s snoozes for a week, d marks complete, and ? shows the full map.

The three groups

The In progress heading shows this week's completed and planned work. This week's plan covers Monday through Sunday.

Ideas and Results are doors under the list, not groups. Open Ideas for everything found and ready for a decision, best first, including work you snoozed or dismissed. Open Results to read measured outcomes, because a measured outcome is not waiting on you.

Old links still resolve: ?view=proof and ?view=measuring both open Results, and /proof redirects to /actions?view=results.

What lands in Ready for you

Four things, and none of them applies itself.

• A staged change the Agent prepared, shown as a before and after diff with its passing gates. Approve or Decline.
• A draft it wrote. Nothing sends itself; you send it or it stays a draft.
• A question that changes the plan, with the options it can see. Answer hands it to the Agent, or take Use the default.
• A measured drop, where a change coincided with a fall large enough to be worth a decision. Revert or Keep.

The nine kinds of work

The kind of work says what sort of change this is. It is not a status, and it decides what measurement will watch.

Those windows are the defaults. The frozen plan shown before you start has the final word, and each kind has a permitted range around its default rather than one fixed number.

Starting freezes the plan

Starting is a commitment to a plan, not a claim that the change is done. Everything except Setup opens What we'll measure first: the primary measure, the subject it applies to, anything else being watched, the window in days, and one line stating the bar. You also choose whether it joins this week's plan or just queues.

Confirm, and Trakkr stamps the plan and freezes it. Nobody swaps the measure later because a different number looks better. Declining asks why in one click (Not relevant, Wrong target, Bad evidence and five more), and that answer tunes what gets suggested next.

Once the work ships it leaves the active plan and appears under Measuring, then moves to Results when the window closes. The Agent never marks its own work Earned.

Doing the work

The work page shows exactly one way to do the work, picked for that work, so there is never a tab bar hiding three you do not need.

Surface	When it appears	What it does
The change	A fix is proposed for a page you own	Shows the diff and applies it through your connected site
Draft	The kind of work is draftable	Writes the article, FAQ, outreach note or reply for you to send
Hand to your Agent	You have a live MCP assistant connected	Builds a self-contained work order to paste into Claude, Cursor or ChatGPT
Run it	Trakkr can carry out this kind of work itself	Runs it in one click

When a better path is one connection away, the surface offers the connection rather than hiding it. A handoff follows a plain division of labor: Trakkr diagnoses, your AI acts, Trakkr verifies. Mark as handed off starts the action and writes a note into Activity.

How much the Agent can do

Permission belongs to a standing automation, not to you as a person or to a kind of work in general. Each automation sits on one rung, and each area of work inside it can sit lower.

Outreach drafts and reply drafts are capped at Prepare drafts permanently. There is no rung that sends a pitch or posts a reply on your behalf.

The rung is a ceiling, not a promise. Before any autonomous change Trakkr checks that the target is a host you own, the platform can reverse that exact change, the original value is saved, the weekly limit has room, auto-rollback is armed, and the quality record is clean. Any failure stages the work in Ready for you instead. Agent permissions has the full rules.

Where suggested work comes from

Suggestions are built from evidence already attached to the brand, not from a content calendar: Search Console gaps, crawler access and fetch patterns, site audit findings, outreach sources worth being cited by, discussion threads, content gaps, competitor movement, perception and diagnosis findings, and answer accuracy checks.

The same suggestion appears wherever you meet the problem. Optimize, Accuracy and the crawler dashboard each embed the matching lane, and those rows are the same rows as the queue's, not a second list. Duplicate signals update one suggestion instead of filling the queue with copies, and a suggestion can retire when the fact behind it stops being true.

If the same work should happen every time a known condition appears, that belongs in Automations.

Common questions