How Long Should a Facebook Ad Run?

Most answers to this question are a number of days, and a number of days is the wrong shape of answer. An ad does not get better because a week has passed. It gets readable because enough people have done the thing you asked them to do, and how long that takes depends on your budget, your audience and what you are optimising for.

How long should a Facebook ad run? Long enough to collect about 50 optimisation events within seven days, which is the threshold Meta itself uses to decide an ad set has left its learning phase. Below that, the numbers you are reading are noise. Above it, they are a signal you can act on. For most accounts that lands somewhere between three days and two weeks, and the right way to plan it is to work backwards from your target cost per result rather than to pick a number of days in advance.

That is the decision rule. The rest of this page is what sits behind it: what Meta's threshold actually requires, when you may kill an ad before you reach it, what editing an ad costs you, and then something the rest of this conversation leaves out entirely, which is how long ads actually run in the wild once they are working. We publish that figure for 234 live advertisers in the Wilow Ad Index, and it is not the number most advice implies.

Why days are the wrong unit

Days measure how long you have waited. Optimisation events measure how much Meta has learned. Only the second one changes the quality of the decision in front of you.

When you launch or significantly edit an ad set, it enters what Meta calls the learning phase. In that period the delivery system is deliberately spending some of your budget on uncertainty: showing the ad to people it is not sure about, to find out. Meta's own About the learning phase documentation puts the exit threshold at roughly 50 optimisation events in a seven-day window, where the optimisation event is whatever you chose to optimise for. An ad set optimised for purchases needs purchases. One optimised for leads needs leads. Link clicks do not count towards a purchase target, however many of them arrive.

This is why two accounts can run the same ad for the same fortnight and only one of them has learned anything. An account spending enough to produce 50 purchases a week reads its results at the end of the first week. An account producing four purchases a week does not get a readable answer in seven days, or in fourteen, and no amount of waiting fixes that — the shape of the problem is budget and target, not patience.

So the useful version of the question is arithmetic. Take your target cost per result, multiply it by 50, and that is roughly what an ad set has to spend before its numbers mean anything. Divide by your daily budget and you have the number of days you should have planned for. If the answer is longer than you are willing to wait, the lever is not time. It is consolidating ad sets so one of them reaches the threshold instead of three of them starving, or choosing an optimisation event further up the funnel that occurs often enough to be learnable.

When can you first judge an ad?

Not before the threshold, unless the ad is failing badly enough that the threshold has stopped being relevant. Those are two different situations and they deserve two different rules.

Inside the learning phase, the numbers move for reasons that have nothing to do with your creative. Cost per result is usually at its worst in the first days and improves as delivery narrows, so a day-two cost per purchase tells you almost nothing about a day-ten one. Reading an ad early and killing it is the most common way advertisers throw away creative that would have worked, and it is expensive in a way that never shows up in a report, because the thing you lost is invisible.

If your account is data-poor — a long sales cycle, a high-priced product, a small market — you have a structural problem rather than a timing one, and there are two honest ways out. Optimise for an event that happens more often and accept that it is a proxy, or widen the ad set so the same budget buys more of the events you care about. What does not work is treating a thin week of data as if it were a thick one.

When should you kill an ad before the threshold?

When the ad is failing at a stage early enough that more data will not change the verdict. The test is whether the failure is upstream of the thing you are optimising for.

An ad that has had thousands of impressions and almost no clicks is not waiting for the learning phase to finish. People are seeing it and declining it, and that judgement is already statistically comfortable even though your purchase count is not. The same is true of an ad whose click-through is healthy and whose landing page converts nobody: the ad is working and the page is not, and leaving it running buys you more evidence of something you already know.

The failure mode worth naming is killing an ad whose cost per result is merely bad. Bad is what the learning phase looks like from the inside. Catastrophic at the top of the funnel is a different reading, and it is the one that justifies acting early.

What does editing an ad cost you?

It costs you the data you have already paid for. A significant edit sends the ad set back into the learning phase and the 50-event count starts again.

Meta counts changes to targeting, placements, the optimisation event and the bid strategy as significant, along with adding a new ad to the set and large changes to budget or bid. Pausing an ad set for more than seven days has the same effect. This is the mechanical reason the "let it run or turn it off" framing is better than it sounds: the middle option, tinkering, is the one that guarantees you never finish a test. Every tweak on day four buys you another week before you learn anything, and the account that tweaks weekly is permanently in learning.

If you want to change something, the cheaper move is usually a new ad set rather than an edit to the running one, so the original keeps its accumulated learning and the new idea gets its own clean count.

How long do Facebook ads actually run?

Longer than the advice implies, for the ads that survive at all. Across 234 brands in the Wilow Ad Index, 58% have at least one ad still running after 90 days. Nearly three in five advertisers, in other words, have something they have decided not to touch for three months.

That rate is not even across industries, and the spread is the interesting part:

• Pet Care — 72% of brands have an ad still running past 90 days
• Food & Beverage — 64%
• Home & Lifestyle — 63%
• Health & Supplements — 52%
• Fashion & Apparel — 52%
• Beauty & Skincare — 51%

Figures as of the Index's 27 August 2026 update, across six published industries, read from live Meta ads through the free Wilow Ad Library. A brand joins the panel when somebody looks it up in the tool, so it is a panel of real advertisers rather than a hand-picked list, and an industry only gets a page once 25 brands sit behind it. The Index publishes its method and updates weekly.

Read against the arithmetic above, those numbers say something fairly specific about how the accounts that are working actually behave. A brand launching a few new ads a week and holding one for three months is not running a timer on its creative. It is letting the losers stop early and leaving the one that works alone — which is the opposite of the weekly refresh cycle that a lot of advice recommends, and it is what the public data shows the majority of advertisers in these categories doing.

What this figure cannot tell you, and why that matters here

Meta publishes the date an ad started running. It does not publish the date an ad stopped. That single gap decides what any longevity number can honestly claim, including ours.

Because the start date is published and the end date is not, what we can say is that a brand has an ad still running past 90 days, or that it has nothing old still running. What we cannot say is that a brand's ads died, or when. Every figure you will read in this field of the form "half of ads stop within four days" is inferred from ads disappearing from the library between one observation and the next, not read off a stop date, because there is no stop date to read. That inference can be reasonable and it is still an inference, and the studies that make it are usually clear about that if you read their method sections.

Public ad data also carries no spend, no clicks, no conversions and no return on ad spend, for any commercial advertiser. Meta publishes spend and impression ranges for ads about social issues, elections and politics, and nothing of the kind for a product ad. So you will not find a benchmark cost per result on this page: we do not have one, and anybody publishing one from library data has modelled it rather than measured it. Where reach is concerned, Meta publishes delivered reach for ads shown in the EU and the UK.

How do you find out how long a competitor's ad has been running?

You look it up, in about ten seconds, free. This is the part of the question that is answerable directly rather than by inference, and most writing on ad duration never mentions that it is possible.

Meta's own archive publishes every active ad with the date it started, and the free Wilow Ad Library reads that archive: search any brand by name, pick a country, and you get every ad the brand is running right now, with no account and no payment. Its long-runners are collected under a "Live for 90+ days" heading, so the ads a competitor has chosen to keep for three months are separated out from everything they launched last week. A brand with no such row has nothing old still running, which is its own answer about how their testing is going.

Doing that to three or four competitors in your category is a more useful calibration than any industry benchmark, because it tells you what longevity looks like for your kind of product rather than for advertising in general. Meta documents the full set of archive tools it publishes in its Transparency Centre. If what you are after is which of a competitor's ads are worth studying rather than how long they have been up, that is a different job and we have written it up separately in how to find winning ads.

Does a long-running ad mean it is working?

It is the strongest signal public ad data carries, and it is still a signal rather than a measurement. Meta stops spending on ads that do not perform, so an ad a brand has kept live for three months has almost certainly earned its place — but you are reading the advertiser's revealed preference, not their results.

Two things keep it honest. A long-running ad may be an always-on piece of a different job, such as a catalogue or retargeting ad that is never meant to be refreshed, in which case its age says more about its role than its performance. And a brand with a long-runner may simply be a brand that is not testing much, which is not a thing to copy. The way to use longevity is as a shortlist: the ads worth taking apart are the old ones, and what you take from them is the angle, the offer and the structure rather than the ad itself.

Common questions

How long does it take for a Facebook ad to start running?

Review is usually a matter of hours and can be close to immediate, though a first-time ad account or a sensitive category can take longer. That waiting period is not the learning phase and it does not count towards it: the 50-event clock starts when the ad begins delivering, not when you submit it. If an ad sits in review for more than a day, the practical move is to check it against Meta's advertising policies rather than to resubmit it repeatedly.

Does pausing an ad reset the learning phase?

A pause of more than seven days does. A short pause does not, so turning an ad set off over a weekend is not the problem that a week-long pause is. If you need to stop spending for a stretch longer than a week, plan for the ad set to re-learn when it comes back.

Should a campaign run longer than an individual ad?

Usually, yes, and they are different questions. A campaign is a container for a goal and can stay open indefinitely; an individual ad is a test inside it with a verdict at the end. Most of the duration decisions that matter are made at the ad set level, because that is where the learning phase lives and where the budget is set.

Do seasonal and retargeting ads follow the same rule?

The threshold is the same and the planning is not. A seasonal ad with a fixed end date has to reach 50 events inside the window you actually have, which means the budget has to be sized for the window rather than for the month — a two-week promotion on a trickle budget never produces a readable result. Retargeting ad sets usually reach the threshold quickly because the audience converts at a higher rate, but they also exhaust that audience, so the thing to watch after the first week is frequency rather than duration.

The short version

Plan the duration backwards from 50 optimisation events, not forwards from a number of days. Leave the ad alone while it gets there, unless it is failing at the top of the funnel badly enough that more data cannot save it. Do not edit a running ad set when a new one would do. And when it works, let it run — because across 234 brands in the Ad Index, 58% of them have exactly one thing they have chosen not to touch for three months, and in pet care that figure is nearly three brands in four.