Loops
A loop is an automation you build once: a trigger, then steps that find your ads, check their numbers and act on them.
Create a loop
Open Loops in the sidebar and start one of three ways:
- Describe a loop. Type what should happen, such as "Pause ads that spend without results". A new draft opens in the builder and Agent builds it while you watch. It never turns a loop on unless you ask.
- Start from a template. A few templates show on the page; All templates opens the rest.
- New loop. An empty draft opens. Choose a trigger, then add steps.

A draft saves as you edit, so you can close the tab and come back. A new draft you leave without changing anything is discarded.
Templates
Each template is a complete loop you can change before you turn it on. A card is listed for the platforms in your
workspace where its loop saves as built, the same check every loop goes through. Reply to comments runs on Meta. A
card opens on the platform you'd use for a new loop. Until you change the new draft, Runs on in its header opens the
template on any other platform it runs on, and the loop's overview says once if it doesn't run on one of your connected
platforms ("Not available for Spotify"). With more than one platform, Runs on at the top of Templates shows the
templates of one platform, each in its own words. Instead of fixed amounts, templates compare with each ad account's own cost per purchase,
and schedules run in the ad account's time zone. Agent and the API start from the same templates
(GET /loop-templates, or list_loop_templates over MCP).
| Template | What it does |
|---|---|
| Daily optimizer | Every morning at 8:00 AM, an Agent step tunes budgets and pauses what isn't working after you approve, then reports to your Inbox. |
| Pause losing ads | Every morning, pauses live ads with a ROAS under 1 over the last 7 days that spent at least twice the ad account's cost per purchase. Ads under 3 days old or still learning are left alone. |
| Scale winners | Mondays and Thursdays, gives ad sets out of learning with a ROAS of 2 or more and at least 10 purchases in 7 days 20% more daily budget, once you approve. |
| Budget pacing | Tells you when a campaign spends 80% of its daily budget before 2:00 PM. |
| Creative fatigue alert | Every morning, tells you when an ad with at least 1,000 impressions has a frequency over 3 and a link CTR down 20%. |
| Weekly performance summary | Every Monday, sums up last week's spend and best ads for the team. |
| Reply to comments | Sorts each new comment on your ads: drafts answers to questions for your OK, tells you about complaints and hides spam. |
| New lead follow-up | Tells you about each new lead, in the Inbox and by email. |
| Pause losing ads on LinkedIn | Every morning, pauses LinkedIn ads clicked less than half as often as their ad account's ads (landing page CTR) after 2,000 impressions. |
| Scale winners on LinkedIn | Twice a week, gives LinkedIn ad sets clicked at least 1.5 times as often as their ad account's ads 20% more budget. |
| Budget pacing on LinkedIn | Tells you when a LinkedIn ad set spends 80% of its daily budget before 2:00 PM UTC. |
| New lead follow-up on LinkedIn | Tells you about each new lead from your LinkedIn lead forms, in the Inbox and by email. |
| Creative fatigue alert on LinkedIn | Every morning, tells you when a LinkedIn ad's frequency climbs and its CTR drops. |
| Weekly performance summary on LinkedIn | Every Monday, sums up last week's LinkedIn spend, leads and best ads for the team. |
| Daily optimizer on LinkedIn | Every morning, an Agent step tunes LinkedIn budgets and pauses what isn't working after you approve. |
The builder
- Blocks, on the left, grouped as Triggers, Find, Check, Flow, Do, AI and Notify. Search them in the field at the top. When nothing fits, type what the step should do and choose Ask Agent.
- The canvas, in the middle: the loop from its trigger down. Select Add a step between steps, or drag from the end of a path to connect it.
- The inspector, on the right: the selected step's settings. With nothing selected it shows the trigger, whom the loop runs as and what's left To finish. Select an item there to jump to the setting it needs.

The Ask the Agent bar at the bottom of the canvas is the loop's own chat with Agent. Ask for a change in plain words, or select a step and say "this step". Press Cmd+J to open it. See Agent.
Press Delete to remove the selected step, Cmd+Z to undo and Shift+Cmd+Z to redo. On Windows, use Ctrl instead of Cmd.
Blocks
| Group | Blocks |
|---|---|
| Triggers | Schedule (every 15 minutes, hour, day or week, or at set times), Meta data updated, Metric crossed threshold (such as cost per purchase above $20 today), Budget spent, New comment, Ad status changed, Changed in Meta, New lead, Creative uploaded and Manual run |
| Find | For each ad account, campaign, ad set, ad or comment, For each segment and For each creative match |
| Check | Check metrics, Creative has..., Check creative fatigue, Check profit (with Pluto Profit connected), Learning status, Check delivery issues and If/else |
| Flow | Split (runs several paths at once), Wait and Stop |
| Do | Pause, Turn on, Change budget, Change bid, Duplicate, Rebalance budget, Scale to target, Rotate creative, Hide comment, Delete comment, Reply to comment and Add tag |
| AI | Agent, Classify (into 2 to 8 groups, one path each), Write reply and Summarize |
| Notify | Notify people: an Inbox notification to you, specific people or the whole team, and an email with Also send an email |
- For each runs the steps after it once per item, for every item that matches. Right after a trigger about a campaign or ad set, it looks inside that one: Budget spent on a campaign, then For each ad, runs for that campaign's ads. After New comment, For each ad runs for the ads the comment is on.
- Checks end the path for an item that doesn't match. A number Meta didn't report never counts as 0.
- Check metrics reads any metric the Analytics dashboard has, including every action Meta counts (your custom conversions too) as a count, a value or a cost per action. A condition can look at one segment only, such as cost per purchase among women 25 to 34, or link CTR on Instagram Stories: choose Only for and the segment.
- For each segment splits the ad account, campaign, ad set or ad it runs for by Age, Gender, Country, Region, Placement, Platform, Device, Hour of the day, a creative part (primary text, headline, image, video, call to action, website URL, ad format) or Product, over the period you choose, and runs the steps after it once per segment. Amount spent over leaves out segments that spent less, and Most segments caps how many it runs for, most spent first. A Check metrics after it reads that segment's numbers. Segment numbers use the ad set attribution setting (as in Ads Manager) and reach back 13 months for an ad account and 14 days for a campaign, ad set or ad, so a split of one takes a period of 14 days or less, with the period it's compared with; reach and frequency aren't available per segment, and neither are purchases and other website or app conversions by region, product or the device people acted on: the builder says so before you save. A loop judges only the days the segment numbers cover: today's arrive the next day, so a period that includes today ends yesterday, and a period they don't cover yet ends the step with the reason instead of acting on part of it. On LinkedIn it splits by Job title, Job function, Seniority, Industry, Company size, Company, Country, Region, County, Conversion action (conversions only), Placement, Device, Serving location and Carousel card (see Loops on LinkedIn).
- New comment reads comments on the posts your ads run as, on the Pages you choose and their Instagram accounts, on Facebook, Instagram or both. With one Page connected, it's that Page; with several, choose them, so a reply loop never answers one brand's customers in another's voice. Ad accounts and Campaign name contains narrow it further.
- For each comment reads the comments of the ad, ad set, campaign or ad account it runs for, or else of the Pages you choose. Each run takes every comment that came in since the last run, newest first. Summarize with Comments reads the comments of the item it runs for in the same way.
- Changed in Meta starts when someone changes a campaign, ad set or ad in Ads Manager, Meta's apps or another app, or changes an ad account's spending limit. Pluto reads each ad account's activity log every 2 minutes while it has an active campaign or one changed in the last 30 days, and every 10 minutes otherwise. Choose the Level and What changed, such as a budget, status, bid or targeting; with nothing chosen, any change starts it. Changes made in Pluto, including your loops' own, and changes Meta makes by itself never start it.
- Ad status changed, Changed in Meta, Metric crossed threshold and Budget spent watch the Ad accounts you choose on the trigger. With none chosen, they watch the accounts the loop's steps name, or your only ad account. They watch every account only when you choose All ad accounts; an Agent step set to all accounts never widens them. Campaign name contains narrows them to some campaigns.
- Change budget and Change bid take amounts (Set to, Never below, Never above) per currency. When the loop's ad accounts bill in several currencies, enter an amount for each. An ad set in a currency without one is left alone, and the run says why. Amounts are never converted between currencies, and a percentage works in every currency.
- Creative has... and For each creative match use what Creative Memory already found. A file it hasn't read yet counts as unknown and stops the path. They never use credits.
- Rebalance budget moves budget between the ad sets or campaigns the For each above it found, in the same currency. What the better ones gain is what the others give up, so their total stays the same. Move at most caps how far each one moves in a run.
- Rotate creative pauses the weakest ad in each ad set and turns on the next. Unless you change it, it judges ads with at least 1,000 impressions and keeps a rotated-out ad off for 7 days.
- If it fails, on an action step, either stops the run or sends the item down an error path.
Only AI steps use credits: Agent, Classify, Write reply and Summarize. Each has Instructions in plain words, and Use the recommended instructions fills in a starting point. Each also has a Model, the same choice as in the Agent's chat, with its effort where the model has one. A step you haven't set runs on Pluto 1 at Medium effort, or on the first model your workspace allows when Pluto 1 is turned off; it never follows anyone's last chat choice. See Models and rates.
When your organization's credits run out, the loop's AI steps wait, shown as Waiting for credits, and carry on by themselves once credits arrive. When a loop reaches its own Credit limit, its AI steps stop. Either way, its other steps keep running. See Billing.
When the AI model is busy or its reply comes back incomplete, an AI step waits and asks again, up to three times over about a minute and a half, before it gives up. If the AI model still doesn't answer before the step has changed or proposed anything, the step fails and the run stops there, so the steps after it, such as a report by email, never send an answer it didn't give. After a change, the step stops early, and the run keeps what it changed. An Agent step whose answer comes back empty, or that has used all its model calls without writing its result, is asked once more for its result; if it still ends without writing anything, it fails the same way, so Notify people never sends an empty notification. A run that stopped early reads as Stopped early in its history, with what it did and why. A run that stops before it finishes, with nothing left to continue it, is ended by Pluto within a few minutes: as it went when every step had finished, at the time its last step finished, otherwise as failed, and the loop can start again.
The Agent step
An Agent step reads everything the person the loop runs as can see. What the Agent may do sets what it may change: Budgets, delivery and bids (on for a new step), Comments, Draft ad edits, Targeting, Duplicate, Delete, Notify, Save scripts and Research competitors. Open a switch to pick its actions one by one.
It also searches the web and reads pages when its instructions need them, and with Pluto and some other models it runs code in a private sandbox (see Agent). Files its code makes come with the step's result. With Save scripts on, it keeps the code it wrote with the loop, so the next run reuses it. It reads the competitors your workspace researched, their ads and funnels, and with Research competitors on it researches them itself; Creative Memory's reading of their ads uses credits (see Agent).
- Draft ad edits edits copy, creatives and tags of ads managed in Pluto as drafts in their launch. Nothing changes in Meta until someone publishes them there, and a published ad goes back to Meta's review.
- Targeting changes live targeting. The step looks up interests and audiences by name to find them.

Its limits sit under Budgets, delivery and bids:
- Change a budget or bid by at most: how far one budget or bid moves in a run, 20% unless you change it.
- Keep each account's daily budgets under: a cap in the ad account's currency. Turning something back on counts as spending more.
- Judge only what spent at least: spend over the last 7 days before it judges an object.
Set its Ad accounts and Goal (Agent decides, Profit with Pluto Profit connected, Cost per result or ROAS; on LinkedIn Leads, Cost per lead, Cost per conversion or CTR) next to the instructions. When a new action is added to a switch you turned on, the step shows it as New with Turn it on; it's never on by itself. Deleting on an ad platform always asks first.
Its report
The Agent step delivers what its instructions ask for as a report: a document with sections, lists, tables, quotes and links, charts and tables of the numbers it read, and the ads it names as cards with their pictures or videos and numbers (as in Agent), with a one-line summary. Ask for the result you want, such as "5 to 10 new angles, each with who it's for, the evidence and three hooks", and the report holds all of it. It also lists what the step changed or proposed and why.
Reports read like a finished document: the result first, a heading per section and one per item inside it, short paragraphs, bullets that lead with the key figure, tables for comparisons, a chart when the result rests on numbers, and numbers and dates written as people read them. Before a report is kept, Pluto tidies its layout: section headings start under the title (which is never repeated), empty sections and leftover labels such as "Answer:" are taken out, every table gets its header row, and a chart appears only where one was made.
A Notify people step after it sends the report. In the Inbox, the notification is from the loop, its row shows the summary, and opening it shows the whole report, with its charts and ads. An email gets the report as text, without the charts and ads; a long one ends with a line that the full report is in the Inbox.
To take a report elsewhere, open the More actions menu at the top of it. Copy (or ⌘ShiftC, CtrlShiftC on Windows, after clicking in the report) copies the whole report: pasted into Google Docs, Notion or Word it keeps its headings, lists, bold, tables and charts, and pasted as plain text it's Markdown, with each chart as a table of its numbers and each ad as its linked picture and name. Copy as Markdown copies the Markdown alone, for a field that would keep the formatting you don't want. The menu also has Download as Markdown (.md), Download PDF and Copy link, which links to the report in the Inbox. A step's report on the Runs tab has the same menu, with Copy report first, where Copy link links to the run, in live and test runs alike; a long one's Open report opens it as the same document as in the Inbox, tables and charts included.
Download PDF saves the report as a PDF file that looks the way it does in Pluto, in the same typeface: its charts, tables, and ads as cards, three to a row, each with its picture, name and numbers. A video shows its first frame with a play mark, and clicking a card in the PDF opens the ad in Pluto. The first page has the title, the summary, and the workspace, who sent it, the date and the days its numbers cover. Every page has the same footer: the workspace and the report's title at the start, and "Made with Pluto Ads" and the page number at the end. A card, a chart, a table row or a quote never splits across two pages, a heading stays with what follows it, and a table that continues on the next page repeats its header row. The PDF is always light, whichever theme you use. It's US Letter where your browser's language names a country that uses it, such as the United States or Canada, and A4 everywhere else. It takes a few seconds: the menu row reads Preparing PDF until the file downloads.
Skills
Skills are instructions your team writes once and reuses, such as a brand voice or a review checklist (see Guidance, skills and models). Agent and Write reply steps use them the way the Agent does in a chat:
- The step knows every skill shared with the workspace, and Pluto's own skills that work in loops, by name and description. It loads one when the item or its instructions match.
- Skills on the step pins up to 5 it always follows, such as your brand voice on every reply.
- A loop runs for the whole team, so it only uses skills shared with the workspace. To use one that's only yours, an owner or admin shares it in SettingsAgent (Settings, then Agent) first.
- A run uses each skill as it is at that moment. The step in the run says which it used, such as "Used skills: Brand voice v3".
- If a pinned skill is deleted, renamed or no longer shared, the step stops and says why, and Skills on the step says so until you choose another one or remove it.
- A skill never changes what the step may do, its limits, what asks first or what it may spend.
Scripts and agents list the skills a step can use with GET /loop-skills or list_loop_skills over MCP, and pin
them by name in the step's skills.
Turn a loop on
Select Turn on at the top right of the builder, or the Loop on switch next to its name. A loop turns on only when To finish is empty. Once it's on, Run now runs it right away. A loop that starts with Manual run has no switch: Run now runs it whenever you select it.
When the loop changes budgets, delivery or comments, Turn on this loop asks how it acts: Ask every time (the default) or Act without asking, the same choice as the Agent's. Its changes then run as you, so it needs a role that includes all of them, and loops: changing budgets for a loop that changes budgets, deleting for one that deletes, never more than its steps do. It stays allowed while your role does. Without one of them, you don't get the choice: the dialog, and Ask before acting on its steps, name in one line what your role doesn't include, such as "Pause and resume live ads". When you ask Agent to let a loop act without asking, it asks you to approve that first; other apps and API keys can't allow it.

Each action step and Agent step also has Ask before acting:
- Always: every change waits for your approval.
- Only above limits: lowering a budget and pausing an ad or ad set go ahead. A raise bigger than Budget increases over, turning anything back on and pausing a campaign ask first.
- Never: acts on its own within the step's limits, once someone whose role includes its changes allows it.
Letting a loop act without asking covers the version you saw. After an edit, or once someone else saves one of its scripts, it asks again, and the header reads Asks again after edits. A loop with an Agent step can also run the scripts your workspace keeps, so a new version of one of those counts too. While it may act without asking, running it by hand with Run now, from another app or with an API key needs everything its changes need, such as changing budgets or pausing ads. The same goes for a loop that starts it. When it asks, a run's changes arrive together as one plan in your Inbox. See Reviewing a plan of changes.
A loop acts for its owner. Its work shows as a loop's, for example Loop for Anna, in activity and the Inbox. It never does more than the person who last changed its steps or a script it can run may do. When another app or an API key changed them, that's the person who made it. So saving a change refuses a step the loop couldn't then perform, and says which step and what it needs, and the list of blocks shows only the ones your role can use.
A loop that's on but has nothing to watch reads Needs a Page or Needs ad accounts in the list instead of On. That happens when your workspace has several Pages or ad accounts and its trigger names none. Its owner gets one Inbox notification that opens the trigger; choose the Pages or ad accounts there and it starts again.
Test runs and runs
A loop without AI steps shows what it would do now on its canvas, without doing it: each step's card says what it found, such as "3 match" or "Would pause 3 ads". The counts update by themselves after each change, cost nothing, and the header says when: Updated just now. Counts over an hour old, or of an earlier version of the loop, say when they're from.
A loop with AI steps has Test run in its ⋯ menu instead. A test run runs on real data and sends and changes nothing. It opens in the Runs tab like any run, marked Test run, with each step's result and the Agent's report. A test run uses credits like a real one. Each person can start up to 20 test runs an hour on a loop.
Every run is kept on the Runs tab of the inspector. Open one to see:
- Needs approval: changes that need your OK, each as one row with Reject and Approve, as in the Agent. Deciding one needs a role that includes that change, such as changing budgets for a budget change; anyone else reads which, in place of the buttons. An approval nobody decides within 24 hours expires.
- Steps: each step with what it did.
- Items: everything it found, marked Changed, No change, Skipped, Postponed or Failed.


Stop run ends a run that's still going. To take back a run's changes, open the step and use Undo, or Undo all, under Changes it made.
Organize many loops
When you have more than a handful of loops, give them structure on the Loops page:
- Folders: select loops and choose Move to folder, or drag a loop onto a folder. New folder in that menu creates one and moves the loops in one step. A loop is in one folder at a time, and folders don't nest. Deleting a folder keeps its loops; they move to No folder.
- Group, filter and sort: Display groups the list by folder, status, trigger, owner, ad account or last run, and sets the order. Filter narrows it to loops that failed, are waiting for approval, ran recently, or belong to someone. While you have no folders, the list stays a plain list.
- Change many at once: select loops (Shift-click selects a range), then Turn on, Turn off or Move to folder in the bar at the bottom. Loops that still need setting up stay off: the notice says how many, and Select them selects them so you can finish them.
- Saved views: save what you see as a view, for you or for the whole workspace. The page's address keeps it too, so you can share a link to it.
You can also ask Agent, for example "Put all my comment loops in a Comments folder" or "Group my loops by ad account". Members can create folders and move loops; every change is recorded in the audit log.
Loops on LinkedIn
A loop runs on one platform, Meta or LinkedIn. With both connected, choose it in Runs on at the top of a new loop. Before you connect one, Runs on starts empty, and the loop is saved once you choose. The builder then offers only the blocks that platform runs, in its own words, such as Changed in LinkedIn and LinkedIn data updates. Once the loop has a block, its platform stays; to run the same loop on the other platform, open the loop's menu and choose Duplicate loop to LinkedIn (or Duplicate loop to Meta). The dialog lists the blocks the copy keeps and any it leaves out because the other platform doesn't run them. The copy starts off, and its ad accounts are chosen again.
On LinkedIn, a loop can:
- start on a Schedule, a Manual run, Creative uploaded, new LinkedIn data, Ad status changed, a change made in Campaign Manager, Metric crossed threshold, Budget spent or a New lead;
- find ad accounts, campaigns, ad sets and ads with For each, and check them with Check metrics, If/else, Check creative fatigue and Check delivery issues;
- Pause, Turn on, Change budget, Change bid, Duplicate, Rebalance budget, Scale toward a target, Rotate creative and Add tag;
- run an Agent step, Summarize performance, Notify people, Split, Wait, Run loop and Stop.
Actions on LinkedIn work on what you published from Pluto Ads; campaigns made in Campaign Manager are left alone, and a copy goes where its original is. A bid on LinkedIn is an amount: Change bid moves the bid of ad sets on the bidding strategy you choose (Manual bidding, Target cost or Cost cap) and leaves ad sets on another one as they are; LinkedIn changes a strategy in the ad set's settings, which you publish, and an ad set on maximum delivery has no bid to move. Amounts use the currency's own decimals. Comment text and lead answers from LinkedIn never go to AI, so Write reply, Classify on comments or leads and Summarize comments aren't offered there. LinkedIn reports in UTC days, so loops on LinkedIn use UTC: a schedule's UTC (LinkedIn's days) and Budget spent's time of day. New lead on LinkedIn reads the lead forms of your LinkedIn ad accounts; where your connection has no LinkedIn leads, the form picker reads LinkedIn leads aren't on yet.
Conditions and measures on LinkedIn are LinkedIn's: clicks, landing page clicks, leads, cost per lead, CTR (landing page clicks), website conversions where an ad set counts them, reach and frequency. Purchases, cost per purchase, Meta's attributes such as the learning phase, and Meta's statuses (processing, pre-approved, waiting for billing info) aren't offered, and a loop that names them is refused when it's saved. Statuses are LinkedIn's: a draft, a completed ad set and a canceled campaign, ad set or ad among them. An ad set or ad that LinkedIn holds because its campaign is off reads paused by its campaign, an ad whose ad set isn't serving reads paused by its ad set, and an ad set past its end date reads completed; only the other holds, such as billing, a spent budget or a lead form on hold, read has issues. Changed in LinkedIn has no spending limit to follow, since only Meta's change history has one. A run that waits for LinkedIn says Waiting for LinkedIn.
Pluto Ads checks a loop the same way however it's made: in the builder, by the Agent, through the API or MCP, from a template or as a copy. A loop LinkedIn can't run as built is never saved, and the reason is shown on its step. On LinkedIn:
- Duplicate makes the copy where its original is. LinkedIn has no way to copy into another ad set or campaign, so Another ad set or campaign isn't offered.
- Pause can't pause an ad while LinkedIn is still reviewing it: pause its ad set instead. A loop that finds ads in review and pauses them is refused when it's saved, and a run leaves an ad in review as it is, with the reason.
- Turn on can't restart a completed ad set (it needs a later end date, which you set in its settings) or anything that was canceled. An ad set holds at most 15 active ads: a run leaves a 16th ad off and says why, and a copy that would be the 16th is made paused.
- Change budget changes an ad set's daily budget, or its lifetime (total) budget when it has no daily one. An ad set with both keeps its total budget as it is when the step changes lifetime budgets. LinkedIn sets its minimum budget and bid by your audience and currency, so a change below them is refused by LinkedIn and the run says so.
- Change bid has no ROAS goal or ROAS limits on LinkedIn.
- LinkedIn counts each conversion in the windows set on it in Campaign Manager, so a loop can't choose an attribution window.
- The Agent step can pause and turn on, change budgets and bids, move a live ad set's start or end, copy where the original is, delete, and draft ad edits for you to publish. It can't change an ad set's audience: that's in the ad set's settings, which you publish. Changes to comments aren't available on LinkedIn.
A loop saved before one of these checks shows the reason on its step in the builder. It can't be turned on, or saved with other changes, until the step is fixed. A loop that was already on keeps running, and its runs skip what LinkedIn can't do, with the reason.
Segments on LinkedIn
For each segment splits an ad set, campaign or ad account by one of LinkedIn's segments and runs the steps after it once per value, with that value's numbers for the period you choose; a Check metrics after it, or a condition's Only for, reads one segment. For example: job titles with fewer than 2 leads from more than 5,000 impressions in the last 14 days, then Notify people with the list. LinkedIn's rules, checked when the loop is saved:
- Job title, Job function, Seniority, Industry, Company size, Company, Country, Region and County are LinkedIn's demographic segments: approximate (values under 3 events left out) and 24 to 48 hours late, so a period of only today or yesterday has none yet. Each run reads LinkedIn's numbers for exactly the loop's period and item, never days added up.
- Demographic segments check counts and rates (impressions, clicks, leads, conversions, CTR and the like), never amount spent, a cost or Amount spent over: LinkedIn doesn't adjust demographic cost for over delivery, so it isn't what you were billed. Conversion action, Placement, Device, Serving location and Carousel card take amount spent and costs.
- One demographic segment at a time: a loop can't combine two segments in one check.
- Segments reach back 365 days, as far as LinkedIn's terms let Pluto keep them. A demographic period that starts more than 6 months ago is whole months (the 1st to a month's last day), as LinkedIn keeps daily demographics 6 months.
- Reach and frequency aren't available per segment. Demographic segments have no conversion value, ROAS or carousel numbers, and Device has no conversions.
- Demographic segments are reported per ad set, so a loop splits ad sets, campaigns or ad accounts by them, never ads.
- LinkedIn doesn't allow its demographic data as input to AI, so an Agent step, Classify or Summarize isn't offered after a demographic segment, and a loop that has one there is refused when it's saved. Conversion action, Placement, Device, Serving location and Carousel card can go to AI like LinkedIn's other numbers.
Loops on X
With X connected, choose X in Runs on. The builder then offers only the blocks X runs, in X's words: ad groups, X accounts, New reply and Hide reply.
On X, a loop can:
- start on a Schedule, a Manual run, Creative uploaded, new X data, Ad status changed, a change made in X Ads Manager, Metric crossed threshold, Budget spent, a New reply to one of your ads or a New lead;
- find ad accounts, campaigns, ad groups, ads and replies with For each, and check them with Check metrics, If/else, Check creative fatigue, Check profit and Check delivery issues;
- Pause, Turn on, Change budget, Change bid, Duplicate, Rebalance budget, Scale toward a target, Rotate creative, Hide reply, Reply and Add tag;
- run an Agent step, Notify people, Post to Slack, Split, Wait, Run loop and Stop.
Conditions on X use X's own metrics: impressions, clicks, link CTR, CPM, conversions and their cost, purchases and their cost, video views, reach and frequency for campaigns, and X's engagements, likes, replies, reposts, follows, app clicks and installs, with the rate and cost X defines for each campaign objective: engagement rate and cost per engagement, click rate and cost per link click, follow rate and cost per follow, app click rate, cost per install, video view rate and cost per view. X sends no value for its conversions, so ROAS, purchase value and AOV aren't offered. X counts each conversion in the windows set on its conversion event, so a loop can't choose an attribution window. Thresholds are checked after each X data update, every 15 minutes while an ad account delivers. Days are each ad account's own time zone, and spend can still change for 3 days.
Actions on X work on what you published from Pluto Ads:
- Pause on an ad removes it from its ad group at X, and Turn on promotes its post again. It stays one ad in Pluto Ads with one history. An ad group past its end date turns on only with a later end date, which you set in its settings.
- Change budget changes an ad group's daily or total budget, or a campaign's daily or total budget. An ad group's budget stays within its campaign's.
- Change bid moves the bid of ad groups on Maximum bid or Target cost and leaves ad groups on Autobid as they are. X changes a bid strategy in the ad group's settings, which you publish, and X has no ROAS goal.
- Duplicate makes the copy where its original is: it promotes the same post again.
- Hide reply hides replies to your own ad posts, at most 50 every 15 minutes per X account. X can't delete someone else's reply, so Delete comment isn't offered.
- Reply posts the text you write in the step, as your X account. X doesn't allow AI-written replies without its approval, so Write reply isn't offered.
- The Agent step can pause and turn on, change budgets and bids, move an ad group's start or end and copy where the original is. It can't change an audience or delete anything live.
X's terms keep its data in Pluto Ads. A loop's checks read X's numbers inside Pluto Ads, and everything a loop sends
names X campaigns, ad groups, ads and replies by their IDs, with counts of what it did: notifications, emails,
Slack messages, and runs read through the API or MCP. The Agent step reads X objects by their IDs too. For the same
reason Classify and Summarize aren't offered on X, and a trigger narrows by ad account rather than by
campaign name: use For each campaign with Name contains instead. Your webhooks hear X's events by their IDs
alone, marked withheld: X's names, numbers and texts stay in Pluto Ads.
Run another loop
Run loop, under Flow, starts another loop, which can run on another platform: a Meta loop can hand off to a LinkedIn loop. The loop it starts gets what the path is about: the item's IDs, its latest numbers and what started the run. Its run says which loop started it, and the Run loop step links to the run it started. A loop never runs itself, directly or through the loops it runs, loops run loops at most 3 deep, and a loop that's off or archived doesn't start.
A loop that may act without asking makes its changes as the person who allowed it. So a loop starts it only while it runs as someone whose role includes everything those changes need, such as deleting at the platform. Saving a Run loop step says what's missing, and a run checks again: its Run loop step stops with the reason, and the other loop doesn't start.
Limits that keep your ads safe
- Loops make at most 60 changes per ad account per hour and 4 budget changes per campaign or ad set per hour. Changes past that wait, and the run shows when.
- One change raises a budget by at most 100% or lowers it by at most 90%, never to $0. A step set to move more (Change budget's percent, Scale to target's step, Rebalance's most to move) is refused when it's saved.
- If someone changed a campaign, ad set or ad outside Pluto in the last hour, the loop leaves it alone.
- Loops act on every campaign, ad set and ad Pluto syncs from Meta, including ones made in Ads Manager. Add tag works on ads managed in Pluto. Loops never touch members, billing, API keys or integrations.
Good to know
- Everyone who views the workspace can open loops and see their runs. Building loops and turning them on takes a role that includes Create and manage loops: Owner, Admin, Manager and Media buyer. Guests don't see loops.
- A loop with AI steps doesn't start when your organization can't use AI, or while it's read-only. Runs shows one skipped run a day that says why, and missed times aren't made up. See Billing.
- A draft that never ran can be deleted with Delete loop. Any other loop is archived with Archive loop; turn on Show archived loops under Display to find it, and Restore to bring it back.
- A loop holds up to 200 steps, and its name up to 120 characters. A folder name holds up to 80.
- Scripts and agents build and run loops with the same rules. See the Public API and the MCP server.