You have automation on your Meta account. Meta's own automated rules, or a Claude session with the Ads connector attached, or a tool you pay for. You check at the end of the day and nothing happened: no budget moved, no ad was switched off, nothing was flagged. The natural reading is that the account was fine and there was nothing to do. That reading is often correct. The problem is that in most setups it looks exactly the same as automation that never looked at your account at all.
That is not hypothetical. It happened on our own ad account, and we read it the wrong way round for most of a day.
A run that found nothing to do and a run that never happened look identical in most logs, because almost every log is a record of changes rather than a record of runs. In Ads Manager the difference is real but it lives in one specific place: Automated rules → Activity reports a rule result of “No action” when a rule ran and changed nothing. The per-object “See history” page cannot tell you this, because it only lists changes that were actually made. For anything that is not a Meta rule — a chat session, a script, a tool — ask for the denominator: how many ad sets it evaluated, not how many it changed.
What happened on our account
In August we launched two lanes on our own ad account through our own product. The ads on the sales lane cleared review and began delivering the same afternoon, so the half of the system that builds campaigns had genuinely worked against real Meta.
Some days later the account was quiet and our database held zero optimization recommendations, for its entire history. I read that as the optimizer running its rules and correctly deciding to hold — the account was small, and no sensible significance threshold would have fired on it. It is a comfortable reading and it was wrong. Finding out took shipping a new log line, and on the run we could then observe, the loop evaluated zero ad sets. Every ad set on the account was paused, so the query for active ad sets returned an empty list and no rule was ever consulted.
“Ran the rules and held” and “never looked at a single ad set” had produced identical records. I could not tell them apart, so I guessed, and I guessed in the direction that flattered the system.
Why silence is the default failure mode
This is not a defect peculiar to us. It falls out of how almost every automation writes its records:
- Audit entries are written per event. If nothing was killed, cut or capped, there is no event, so nothing is written.
- Summaries are often produced and then dropped. Ours were built by the engine, returned to the scheduler that called it, and discarded there — nobody was assigning the return value.
- Job runners record success or failure. A run that evaluated ten ad sets and a run that evaluated none both record success.
The result is a system whose failure mode is silence rather than an error. Errors get noticed. Silence gets interpreted — and people interpret it in the direction they already believe.
The two logs in Ads Manager, and why the popular one is the wrong one
If your automation is Meta's own automated rules, Meta does answer this question. It just answers it in a different place from the one most people open.
The page most people open is the per-object history. Select a campaign, ad set or ad, click Options below the name next to Compare, then View history. Its columns are Activity, Activity details, Item changed, Changed by, and Date and time, and if a rule caused the change the rule's own name appears under Changed by. Read what those columns are: every one of them describes a change that was made. A rule that ran and decided to hold changes nothing, so it appears nowhere on this page — and neither does a rule that never ran. This is the page that cannot answer your question.
The page that can is the rules' own activity view. Open Automated rules, then select Activity on the left to see rule results. Meta documents the result types, and the first one is the one that matters here:
- No action — no action was taken on your campaign, ad set or ad. This is the row that means the rule ran and held. Its presence is the good news; its absence is the thing worth investigating.
- Notification sent — a notification went to your ad account.
- Objects turned off, budgets increased or decreased, manual bids increased or decreased — the rule acted, and these changes will also show up on the per-object history under the rule's name.
So the check is thirty seconds: open Activity, pick the window you are asking about, and see whether there are rows at all. Rows of No action mean the rule is running and your account genuinely gave it nothing to do. No rows at all is a different statement entirely — the rule did not run, or it has no schedule, or its conditions cannot match anything currently active. Those are the cases people mistake for a healthy quiet day.
If Claude is your ads manager
A chat session with the Ads connector has no equivalent view, and it is worth being clear about why. The connector acts as the token that authorized it, so anything it changes lands in the per-object history under the name of the person whose login it is using. That covers the changes. It does not cover the reads. A session that pulled your account, looked at it and concluded there was nothing to do leaves no trace anywhere, and neither does a session you meant to run and never opened.
For a chat, the denominator has to come out of the transcript, so ask for it explicitly and ask for it before the conclusion: how many ad sets did you evaluate, over what window, and against what threshold. A model that has read nothing will still write you a confident paragraph about your account being in good shape if you let it summarize first and justify later.
What to ask of anything that manages a spending account
The general form of the fix is to make the different kinds of nothing say different things. There are three, they are not equally reassuring, and they are almost always printed identically:
- Held before any rule — the object was never eligible to be judged. An ad still in review, an ad set that is paused. No rule was consulted, so this is not a performance judgement about anything, and it is the one most likely to be misread as one.
- Below the threshold — the rules ran and the data was too thin to act on. This should print both sides of the comparison: how many days of data exist and how many are required. “Not enough data” on its own is an answer that answers nothing.
- Ran and held — the rules ran against sufficient data and chose to do nothing. This is the only one of the three that is actually good news, and it is the only one people think they are getting.
One caution on building this yourself, learned the tedious way: have the reporting ask the decision path what it concluded, rather than recomputing the same judgement a second time for display. A second copy of the logic drifts from the first, and it will start lying at the worst possible moment — the moment the two disagree is exactly the moment you are relying on the readout.
Where we sit
Everix is a scheduled version of this loop, and the log line described above is ours — we built it because we could not answer this question about our own account. Being straight about the state of it: the building half has run against real Meta, and the optimization half has not yet run against meaningful spend, so read this as a report on the instrumentation problem rather than a claim about results.
Before you trust automation on an account that spends money, find out what it writes on a day when it does nothing. If the answer is nothing, you do not have automation you can supervise — you have automation you can only hope about.