Automations
Work that runs without you, from a plain alert to a limited site change. The permission ladder, the canvas, exact rules, delivery, and replay.
An automation is anything that runs without you. The safest kind only tells you what it saw. Give it more permission and the same automation can bring you a suggestion, prepare a draft, or make a limited change to your site. A notification is not a separate system: it sits on the bottom of that ladder.
Open it from Automations in the sidebar, or go to /automations. Automations need a paid plan, and client portal users do not see the page.
Some teams still see Workflows in the sidebar. That is the older page at /workflows, and it stays until the rollout reaches the brand. Nothing is missing from it: exact when-this-then-that rules live inside Automations too, next to the judgment calls the Agent runs, and saved /workflows links open the same automation once the rollout arrives.
The permission ladder
The ladder decides what may happen after a check. It starts at a report and ends at 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. |
| Prepare for approval | It prepares the change and waits for your approval before applying it. |
| Handle it | It makes changes up to your weekly limit and reports each one. |
The editor draws a line above Prepare for approval reading "everything below changes your site". Nothing crosses it because the Agent asked, found something serious, or read an instruction on a page. Only an editor grants a rung, and a new automation never starts above Suggest only.
Permission is separate for each kind of work, so one automation can prepare technical fixes while only suggesting content refreshes:
- Technical fixes
- Page improvements
- Content refreshes
- New content
- Getting listed
- Community replies
Three caps are enforced, not advised. Getting listed and Community replies never go past Prepare drafts. Prepare for approval needs a connected CMS that can undo a change. Handle it is earned: ten clean runs reach Prepare drafts, five approvals reach Prepare for approval, then ten more clean runs. The record shows that as a Trust bar.
A rung that writes to your site also carries a weekly cap (1 to 50, default 5) and a switch to revert automatically if the numbers fall. Every change is snapshotted first. Agent permissions has the wider safety contract.
Build one on the canvas
Every automation is one left-to-right flow of four steps.
| Step | What it sets |
|---|---|
| Watch | The scope: top cited pages, owned pages, tracked prompts, competitors, crawler activity, or subreddits, narrowed to a top N or a URL path. |
| Check | When it runs and the condition that counts: a schedule, or one event. |
| Act | Which of the six areas it may touch, an optional focus, and the rung. |
| Tell | Where the result goes. |
At /automations/new the empty canvas asks "What should Trakkr keep an eye on?". Describe the automation in a sentence and Trakkr drafts the flow, start from one of four patterns (Citation guard, Comparison watcher, Weekly digest, Crawler health), or build it block by block. Selecting a step opens a property sheet, and the numbers on each card are real counts from your brand.
You never pick an engine. Trakkr compiles the flow into an exact rule when the condition is precise, such as visibility falling 10% on ChatGPT, and into a judgment call for the Agent when the answer needs several signals weighed, such as which cited page to refresh.
Three automations seed with a new brand, all at Suggest only: Watch your money pages, Watch crawler health, and Find citation wins. Rename, retune, or pause any of them. Rows badged Trakkr are standing work Trakkr runs for every brand, and are read only.
What an exact rule can watch
One trigger per rule. Conditions live on the trigger itself, not in a separate step.
| Watch for | Fires when |
|---|---|
| Every day, Monday, or month | The schedule comes round, or only if the result changed |
| Visibility moves | Your score moves at all, by 5%, or by 10% |
| A rank moves | Your prompt position moves at all, by 3 places, or by 5 |
| Perception shifts | An attribute score moves past your threshold |
| The audit score drops | A site audit score falls at all, by 5, or by 10 |
| A citation appears | A new source is seen once, twice, or 5 times |
| A prompt has no answer from you | AI names 1, 2, or 3 rivals in your category and not you |
| Something worth doing turns up | A suggested action appears, optionally only at fit score 60 or 80 |
| A competitor moves | A tracked competitor's visibility passes your threshold |
| A competitor is added | A competitor enters tracking, by hand or from research |
| Research finishes | A research check completes, optionally only on a 5% or 10% move |
| A report is ready | A daily or weekly report finishes |
| An article is written | An article finishes generating |
| An article goes live | An article is marked published |
| A campaign finishes | Every article in a campaign is generated |
| Work is created | A new Action appears in your feed |
| Work is finished | Someone marks an Action done |
The legacy editor adds Brand Created, which fires once, before a brand has data to watch.
Crawler and AI-conversation triggers (spikes, first crawls, bot silence, error surges, robots.txt changes, blocked cited pages, AI referrals, page cited, citation lost) are no longer offered. Nothing in production emits those events, so a rule on one would wait forever. The picker hides them.
What an exact rule can do
A rule runs an ordered list of actions top to bottom in one check. There are no branches and no per-action conditions. For different actions in different situations, use two rules on one trigger.
| Action | What it does | Needs |
|---|---|---|
| Send Email | Emails the addresses you list, with an optional summary and digests | Nothing |
| Send to Slack | Posts to a channel as plain text or rich blocks | Slack connected |
| Send to Teams | Posts to a Microsoft Teams channel | Teams connected |
| Webhook | POSTs the event to any URL, with your own headers and payload | Nothing |
| Generate Summary | Writes a short or detailed briefing of what happened | Nothing |
| PDF Report | Builds a formatted PDF and can email it | Nothing |
| Generate Article | Writes an article aimed at a prompt from the event | Nothing |
| Create Campaign | Generates a batch of articles for the event's top prompts | Nothing |
| Draft Outreach | Writes a personalized pitch email for a citation target | Nothing |
| Add Prompt to Tracking | Pushes prompts from the event into your tracked list | Nothing |
| Publish to your site | Sends an article to a connected CMS | A connected site |
| Google Sheets | Appends a row to a connected sheet | Google connected |
| Create Work Item | Opens a task or issue in a work tracker | Provider connected |
| Notion Page | Creates a page in a Notion database | Notion connected |
Salesforce, HubSpot, and a generic CRM record appear in the underlying action list but are not available; a rule using one refuses to save and points you at Webhook or Zapier. Connect destinations first in Integrations: a rule cannot use what is not connected.
Under Advanced Settings each rule takes a Cooldown Period: none, 15 or 30 minutes, or 1, 2, 6, 12, or 24 hours. Set one whenever the trigger is broad. The pattern that catches people out is a noisy trigger with no cooldown and an email with no digest.
The register
/automations lists exact rules and Agent automations together. Above them a status line reads like 3 running · 1 draft · 12 found this month.
| Column | What it shows |
|---|---|
| (status dot) | Green while running, grey while paused, red when it needs attention |
| Name | Your name for it, plus badges: Draft, Needs attention, Trakkr, and the latest outcome |
| Flow | The shape of its flow in miniature |
| Found | Findings this month, or a dash |
| Last | The date it last checked or fired |
| Autonomy | The highest rung still allowed after every safety cap |
The outcome badge is not a self-reported success claim. It says what came of the last check: Nothing found, Found something, Prepared work, Performed work, Delivered, Partly delivered, Latest failed, or Waiting on setup when the automation is fine but its data is not connected.
Right-click a row, or use the "···" menu, for Open, Rename, Duplicate, Rehearse last 30 days, Pause or Resume, and Delete. Pausing keeps the flow, settings, history, and delivery. A duplicate lands as a paused draft, and a copy of a site-changing automation steps back down the ladder rather than inheriting its authority.
The record
Opening an automation shows the same canvas in one of three states.
- Draft. Build or retune the flow. Review and turn on moves you forward, and nothing is armed until you do.
- Review. A panel headed Before it goes live replays your last 30 days: how many times it would have reached you (or how many scheduled checks it would have run, which are checks, not assumed findings), a per-day strip, real past moments with what it would have done, and a preview of the message it would have sent.
- Live. The Service record, drawn from real runs, with a Trust bar toward the next rung.
Read the replay before turning anything on. Ten matches in a month might be right for a Slack alert. Three hundred means the condition is too broad or wants a digest. Nothing matching can be fine too. If the replay fails, retry it: it is your only preview of the volume ahead. Rehearse last 30 days does the same for a live automation, and sends nothing.
One execution is one check, and quiet checks are recorded as real results. A system that only reports problems makes silence ambiguous: you cannot tell whether it looked and found nothing, failed, or never started. The most recent check is always shown, and the quiet ones are counted out loud ("14 quiet checks. Nothing else needed you.").
Exact rules open at /automations/rules/:id as a read-only canvas, with Open legacy editor for the full form: detailed conditions, multi-step actions, destination pickers, and two test modes. Preview Mode says whether the trigger would fire and why, and prints one line per action without sending. Live Mode really runs them, and needs editor access.
Delivery and digests
Delivery is separate from permission. The rung controls what may happen; delivery controls where the result appears. Sending a result to Slack does not let an automation post a community reply.
In-app delivery is always on. The canvas adds Slack, email, and a webhook (webhooks ride plain alerts only, since they carry machine payloads). Teams, a Notion database, a spreadsheet tab, or a work-tracker project need the legacy editor, which validates the destination before you save.
On an email action, Batch into digest emails groups matching events into one message every 15 minutes, hourly, or daily. Leave it off when the signal is urgent.
Plan limits
Saved automations are counted per brand, and rules and Agent automations share the allowance.
| Plan | Saved automations per brand |
|---|---|
| Growth trial | 1 |
| Growth | 3 |
| Scale | Unlimited |
A paused automation holds its slot, because its settings and history are kept. A deleted one does not. Agent automations carry runtime caps on top: each runs at most 60 checks a calendar month by default, and a brand's Agent automations share a monthly spend safeguard, so they pause rather than run up an open-ended bill.
Common questions
Does an automation cost anything?
There is no separate charge for switching one on, though automations need a paid plan. Work that consumes a paid resource still consumes it, such as generating an article or a report. Judgment calls also carry check and spend caps, so they stop rather than run on.
Why did my automation not fire?
Check its state first, because a paused automation does nothing. Then check the condition, the cooldown, and whether the destination is still connected. Event rules fire on a new matching event, not because the current value already sits past your threshold. For a judgment call, a quiet check means it looked and found nothing.
What happens when I lower a permission?
New checks use the lower rung immediately. The automation can still inspect the same area and bring you a suggestion unless you pause it too. A check already under way may finish under the permissions it read at the start, so pause it when you need it stopped now. Work it prepared stays in the record.
Can two automations fight each other?
They can overlap. Two alerts on one signal send two messages, and two broad site automations can inspect the same page. Trakkr snapshots a page before an allowed change and rechecks its state before applying it, but distinct scopes are still better. Search the register before covering an area twice.
Does a live test really send?
Yes, for messages. It posts to the real Slack channel with a [TEST] prefix and calls a real webhook. Test emails go to your own address, not the configured recipients, and actions that write or publish content return a preview. Use Preview Mode for the logic, then Live Mode once to prove the connection.