Automations

:::summarybox learn The one model behind notifications, exact rules, and Agent work How to edit an automation as a sentence The five permission rungs and six separate work areas How checks, outcomes, delivery, and backtesting work What happens to exact rules you already use

note

An automation is anything that runs without you. The safest version only tells you what it saw. Give that same automation more permission and it can suggest an Action, prepare a draft, or make a limited change to your site.

That is the whole model. A notification is not a separate system. It is an automation on the first rung of one permission ladder.

Start with the ladder

The ladder tells you what can happen after a check. It starts with a report and ends with limited, reversible work.

Rung	What it means
Tells you	It tells you what it saw. Nothing else happens.
Suggest only	It brings you findings. Nothing changes without you.
Prepare drafts	It prepares the work. You approve before anything ships.
Do it, then tell me	It makes the change, tells you right away, and you can revert.
Handle it	It makes changes up to your weekly limit and reports each one.

New Agent automations start at Suggest only. Exact rules that only send a message appear as Tells you. Nothing moves up the ladder because the Agent asked, found a serious issue, or read instructions on a page. An editor must grant the higher rung.

Permission is also separate for each area of work:

Technical fixes
Page improvements
Content refreshes
New content
Getting listed
Community replies

For example, you can let an automation prepare technical fixes while it only suggests content refreshes. There is no global switch that quietly grants the highest rung to every kind of work. See Agent permissions for the full safety contract.

Edit a sentence, not a form

When you create an automation, the editor reads as one sentence. Each highlighted phrase is a choice you can change.

For example:

Every Monday, look at all owned pages and suggest what to do, telling you in-app.

Or:

When a citation is lost, look at your top cited pages and just tell me, telling you in-app + Slack.

The four parts answer four plain questions:

Part	Question it answers	Examples
When	When should Trakkr check?	Every day, every Monday, when visibility moves, when a citation is lost
Look at	What is in scope?	Owned pages, tracked prompts, competitors, crawler activity
How far	What may happen after the check?	Just tell me, suggest what to do
Delivery	Where should the result go?	In-app, email, Slack, Teams, webhook

The short editor covers schedules and the events most teams use: visibility, ranks, perception, citations, competitors, crawler health, research, reports, articles, campaigns, and Actions. If an exact rule needs detailed conditions or a destination picker, open Edit rule details. The same automation opens in the full exact-rule editor.

The sentence tells you the outcome, not the machinery. Trakkr decides whether an exact rule can handle it or whether it needs the Agent to make a judgement call.

One check, one result

A single execution of an automation is a check. The checks list is useful even when nothing needs your attention.

This matters. A system that reports only problems makes silence ambiguous. You cannot tell whether it checked and found nothing, failed before it checked, or never started. Trakkr records quiet checks as real results, such as:

Checked 61 pages. Nothing changed.

On an Agent automation, the result may be:

Nothing new, when the check completed and found no useful change
Suggested, when it found a specific Action for you
Draft prepared, when you allowed it to prepare the work
Changed, when the granted rung and safety checks allowed a reversible site change
Failed, when the check could not finish

Exact rules are different. They react only when their trigger matches, so their history shows sends rather than quiet checks. A citation-loss rule with no matching event has nothing to send. The backtest and live test tools help you check its conditions.

The Automations register

Open /automations to see exact rules and Agent automations in one list. The row does not expose which database or service owns it. You only need to know what it watches, what it is allowed to do, and what came of the last checks.

The default columns are:

Column	What it means
Automation	The name you gave it. The second line shows when it checks and what it watches.
Allowed to	The highest useful permission summary after all safety caps have been applied.
State	Running or Paused, plus a switch that changes that automation only.
Last check	When it last checked or, for an exact rule, last fired. A blank mark means no check yet.
Outcome	What came of it. When there is work to review, the cell links to the relevant Action or editor.

You can also show Activity, Delivery, and Type from the column control. Activity gives a 30-day strip of quiet and productive checks. Delivery shows where results go. Type says Exact rule or Judgment call.

Use the switch in a row to pause one automation without deleting it. Select several rows to pause or resume them together. Pausing keeps the sentence, settings, history, and delivery choices.

The Outcome column is not a self-reported success claim. It links to the work that came of the check. Later measurement appears with the Action, using neutral before and after evidence. Trakkr uses coincided with when it reports a measured change; it does not take credit for that change.

The automation record

Open an Agent automation at /automations/agent/:id. Its record has four sections:

Overview shows its state, sentence, schedule, scope, current permission summary, delivery, and latest check.
Permissions lets an editor set a separate rung and weekly limit for each work area.
Checks lists when each check happened, what it did, and what it found. Open a row to see its evidence and steps.
Delivery controls where findings go. In-app is always on; email, Slack, and Teams can be added when available.

The check history keeps the ten most recent checks on the first view and tells you when more history exists. Repeated quiet checks are grouped so the useful findings stay easy to scan.

Exact rules open at /automations/rules/:id. The full editor handles detailed conditions, retries, multi-step delivery, and destination settings.

Delivery and digests

Delivery is separate from permission. Tells you controls what may happen; delivery controls where the result appears.

In-app delivery is always present for Agent automations. Depending on the automation and the integrations connected to the brand, results can also go to:

Email
Slack
Microsoft Teams
A webhook
A work tracker
Google Sheets
Notion

Exact rules can also format a result as a summary or PDF. Detailed destinations such as a specific sheet tab, Notion database, or work tracker project stay in Edit rule details, where Trakkr can validate the destination before saving.

Email delivery on exact rules supports instant, 15-minute, hourly, and daily digests. A digest groups several matching events into one message. Use it when the signal is useful in aggregate but noisy one event at a time. For an urgent crawler outage, use instant delivery. For routine new-citation notices, an hourly or daily digest is usually calmer.

Delivery never lifts a permission. Sending a result to Slack does not let the automation post a community reply or send outreach on your behalf.

Exact rules versus judgement calls

Some automations can be stated as a precise condition:

When visibility falls by at least 10% on ChatGPT, send a Slack message.

That is an Exact rule. It evaluates the same condition the same way every time. It is quick, predictable, and does not need a language model to decide whether the condition matched.

Other automations need context:

Every Monday, look at our most cited pages and suggest which one needs a refresh.

That is a Judgment call. The Agent must compare evidence, respect the stated scope, and decide whether anything is worth bringing to you.

Use an exact rule when you can write the trigger as a number, event, schedule, or filter. Use a judgement call when the useful answer depends on several signals or on the content of a Page. You do not pick an engine. You pick the sentence and permission, then Trakkr chooses the suitable path.

Exact rules remain fully reachable. Open Edit rule details when you need the trigger catalog, action catalog, conditions, cooldowns, retries, testing modes, templates, or plan limits.

What happens to existing exact rules?

Nothing happens to their data or schedule.

They appear in Automations without changing how they run:

Existing exact rules appear in the Automations register.
Active rules stay active; paused rules stay paused.
Their trigger, conditions, actions, cooldown, retries, and destinations stay unchanged.
Old links continue to open the rule.
Edit rule details opens the existing full editor.

The register is a combined view over the two existing systems. It does not copy or move exact-rule records.

Backtest before turning it on

Every creation flow shows what the sentence would have matched in the last 30 days.

For an event-based exact rule, the backtest uses real past events and shows up to five examples. For a schedule, it estimates how many scheduled checks would have occurred. For a judgement call, it shows the moments the Agent would have inspected. It does not pretend the Agent made a past decision or spend money simulating one.

Read the result before you click Turn it on. Ten matches in a month might be right for a Slack alert. Three hundred probably means the condition is too broad or needs a digest. If nothing matched, that can also be valid: the automation will wait for the next matching event.

If the backtest fails, retry it before relying on the automation. A failure to read history does not prove the live automation is broken, but it removes your best preview of its likely volume.

Common questions

Does an automation cost anything?

There is no separate charge for switching one on. Exact rules require a paid Trakkr plan. Work that consumes an existing paid resource still consumes that resource, such as generating an article or report. Judgment calls also have built-in check and spend caps, so they pause rather than create an open-ended model bill.

Can two automations fight each other?

They can overlap. Two alert rules may send duplicate messages, and two broad site automations may inspect the same Page. Trakkr snapshots a site value before an allowed change and checks the latest state again, but it is still better to give automations distinct scopes. Search the register for the Page or signal before creating another one. If you see repeated overlap, pause one and narrow its sentence.

What happens when I revoke a permission?

New checks use the lower permission. The automation can still inspect the area and bring you a Suggested action unless you also pause it. A check that has already started may finish under the permissions it read at the start, so pause the automation first when you need an immediate stop. Already prepared drafts and completed changes remain in history.

Why did my automation not fire?

Check its state first. A paused automation does nothing. Then check the exact condition, cooldown, delivery connection, and recent history. Event rules only fire for new matching events; they do not fire because the current value already meets a threshold. For Agent automations, a quiet check may mean it looked and found nothing worth suggesting. A failed connection pauses delivery without deleting the automation.

Do automations count against my plan limit?

Exact rules count against the automation limit: paid trials allow 1 per brand, Growth allows 3, and Scale allows unlimited rules. Paused exact rules still occupy a slot; deleted rules do not. Agent automations use a separate allowance. By default, each has up to 60 checks in a calendar month, and all Agent automations for a brand share a monthly spend safeguard.