How long should you run a Facebook ad before judging it?

What does Meta’s learning phase require?

About 50 results in a week. Meta’s help page About the learning phase says the delivery system is “exploring the best way to deliver your ad set” after you create it or make a significant edit, and that ad sets leave the phase “as soon as they can deliver stably. This usually occurs after about 50 results in the week after the ad set’s last significant edit.”^[1] The result is whatever the ad set optimizes for: a purchase, a lead, a link click. Meta’s page on the Optimization events metric adds a softer line: “We recommend around 50 optimisation events, but some ad sets stabilise earlier.”^[4]

Meta also tells you what to expect while you wait. “During the learning phase, ad sets are less stable and usually have a higher CPA,” and results during learning “aren’t necessarily indicative of future performance.”^[1] An ad set that will not get there is marked Learning limited in the Delivery column, which Meta defines as one “unlikely to receive about 50 optimization events in the week after your last significant edit.” The same page says plainly: “Learning limited isn’t a penalty.”^[2]

How much data do you need before judging, by objective?

Enough that the metric you are judging has settled, and that differs between a click and a purchase. Meta publishes one threshold, the 50 results in a week above, and it applies to whatever the ad set optimizes for.^[1] The floors below for impressions, clicks and purchases are practitioner rules of thumb from three agencies and tools, not Meta rules, and they disagree with each other on how many purchases are enough.

The spread on purchases is real. Fifteen to 20 is the floor for a cost per purchase you can act on at all; 50 is the count at which Meta says delivery has usually stabilized.^[1][10] Wait for 50 if you can. If you cannot, treat 15 to 20 as the earliest honest read, and look at click-through rate and click volume first; they arrive days sooner and say whether the creative is the problem.

When should you turn off a Facebook ad early?

Only when the data already rules it out: the ad is broken, or it has spent far past your target cost with nothing to show. AdManage’s kill guide puts broken tracking, a wrong URL or a wrong price in the first group; those are not tests, so there is nothing to wait for.^[12] The second group is the zero-conversion ad. Datadrew’s rule is to “turn it off early if it has spent 3x your target CPA with zero purchases.”^[10] AdManage reaches a similar number from probability: an ad that really converts at target CPA has only about a 5% chance of showing zero conversions after three times that spend, so it kills at about 3.0 times target CPA, or 2.3 times for the impatient.^[12]

Two cautions. First, the rule only holds on settled data: AdManage says conversion reporting can lag 24 to 72 hours, so a zero on this morning’s spend is not yet a zero.^[12] Second, a low click-through rate is a creative verdict, not a purchase verdict. AdManage treats a link CTR under 0.5% on prospecting as a sign the creative is not working, and cuts the bottom 20 to 30% of ads by CTR once each has about 1,000 impressions.^[12] That verdict arrives days before any purchase verdict can.

Why does day-3 ROAS mislead?

Because on day 3 the ad has not finished earning credit for the clicks it already paid for, and the delivery system has not finished learning. Meta’s default attribution setting for a new ad set optimized for conversions is 7-day click or 1-day view.^[6] Under click-through attribution Meta counts events “that occurred within 1-day or 7-day after a link click on your ad.”^[5] A person who clicks on Tuesday and buys on Sunday is a conversion for Tuesday’s ad. Meta’s reporting page notes that you can report conversions “by time of conversion or time of exposure when the impression was served,” which is why the early days of a report keep filling in after the fact.^[7]

The learning phase compounds this. Meta says CPA is usually higher while the ad set learns, so a day-3 cost per purchase is both incomplete and biased upward.^[1] Meta’s A/B testing guidance draws the same conclusion: “we recommend a minimum of 7-day tests,” because “tests shorter than 7 days may produce inconclusive results,” and if your customers typically take longer than 7 days to convert, run the test longer, “such as 10 days.”^[8] A full week also covers each day of the week once, so a weekend lull or a payday spike does not stand in for the whole ad.

Wait, kill or scale: the decision checklist

1. 1

   Check the ad is a real test

   Tracking fires, the link is right, the price on the page is right. If not, fix it and restart; there is nothing to judge.^[12]
2. 2

   Read the Delivery column

   Learning means wait. Meta’s advice is to “wait to edit your ad set until it’s out of the learning phase.”^[1] Learning limited means the setup cannot reach about 50 events a week; Meta’s fixes are a bigger audience or budget, a higher cost control, a more frequent optimization event, or combining ad sets.^[2] Add the Optimization events and Last significant edit columns to see both numbers.^[1]
3. 3

   Apply the early kill

   Zero conversions at about three times target CPA, on data at least 48 to 72 hours old: turn it off.^[10][12] Otherwise keep going.
4. 4

   Check the data floors

   Below the floors in the table above, the verdict is wait. Above them, compare cost per result to your target over the whole run.
5. 5

   Scale without restarting learning

   Meta says a budget change may or may not be a significant edit “depending on the magnitude of the change”: $100 to $101 is unlikely to restart learning, $100 to $1,000 may.^[3] Datadrew waits for 20 purchases before raising budget.^[10] Raise in steps, and leave targeting, creative and the optimization event alone; any change to those restarts learning.^[3]

How does budget change the timeline?

Budget sets how many results a week can buy, and that number, not the calendar, decides whether a week is enough. The arithmetic is weekly budget divided by expected cost per result. At $20 a day, $140 a week, and a $35 cost per purchase, you buy about 4 purchases a week. Meta’s threshold is about 50, so the ad set sits in Learning limited and a seven-day read rests on four sales.^[1][2] Reaching 50 purchases a week at $35 each takes $1,750 a week, or $250 a day.

Meta’s fixes for a small budget are on its learning-limited page: a bigger audience, combining ad sets, and “choosing an optimization event that occurs more frequently,” for example moving from purchases to add to cart.^[2] Its learning-phase page adds “avoid high ad volumes,” because “the delivery system learns less about each ad and ad set” when there are many, and a Meta guide puts the number at “no more than five ads per ad set.”^[1][9] That is what the four versions from Showfolk’s AI ad generator are for: four layouts of the same offer in one ad set, one broad audience, sharing the 50 events, so the budget runs a creative test instead of judging a single ad too early. The best way to create Facebook ads covers what to vary between them.

Questions people ask

How long should Facebook ads run before you judge them?

At least seven days, and until the ad set has about 50 results, the point at which Meta says an ad set usually leaves the learning phase. Practitioners judge a traffic ad after 3 to 5 days and 1,000 to 3,000 impressions, and a conversion ad after 7 to 10 days and at least 15 to 20 purchases, 50 if the budget allows.

How long is the Facebook ads learning phase?

Meta does not give a number of days. It says an ad set usually exits the learning phase after about 50 results in the week after its last significant edit, and some stabilize earlier. At a small budget an ad set may never reach 50 in a week, and the Delivery column then reads Learning limited.

When should you turn off a Facebook ad?

Turn it off at once if tracking is broken or the link is wrong. Turn it off early if it has spent about three times your target cost per acquisition with zero conversions on data two to three days old; that is a practitioner rule, not Meta’s. Otherwise wait for the data floors, then compare cost per result to your target.

Does changing the budget restart the learning phase?

Sometimes. Meta says a budget change may or may not be a significant edit depending on its size: $100 to $101 is unlikely to restart learning, $100 to $1,000 may. Any change to targeting, creative or the optimization event, adding an ad, or pausing for seven days or more always restarts it.

How long should you test Facebook ads?

Meta recommends at least 7 days for an A/B test and allows up to 30; it says shorter tests may be inconclusive, and suggests longer, such as 10 days, when customers take more than a week to buy. Judge the test only when it has ended.

Sources

Every figure above points at one of these. Facts were last checked on October 7, 2026.

1. [1]Meta Business Help Center: About the learning phase. “About 50 results in the week after the ad set’s last significant edit”; higher CPA and less stable results during learning; the advice to wait to edit and to avoid high ad volumes
2. [2]Meta Business Help Center: About learning limited. The definition of Learning limited, “isn’t a penalty”, and Meta’s fixes: combine ad sets, expand the audience, raise budget or cost control, choose a more frequent optimization event
3. [3]Meta Business Help Center: Significant edits and learning phase. The list of significant edits, and the $100 to $101 versus $100 to $1,000 budget example
4. [4]Meta Business Help Center: Optimization events. “We recommend around 50 optimisation events, but some ad sets stabilise earlier.”
5. [5]Meta Business Help Center: About attribution models and attribution settings. Click-through attribution counts events within 1 or 7 days after a link click; view-through within 1 day after an impression
6. [6]Meta Business Help Center: Differences between conversion optimizations in Meta Ads Manager. Lists 7-day click or 1-day view as the default attribution setting for maximize number of conversions
7. [7]Meta Business Help Center: Troubleshoot common third-party reporting differences. Ads Reporting can report conversions by time of conversion or by time of exposure
8. [8]Meta Business Help Center: What are best practices for A/B tests for Meta ads?. Minimum 7-day tests, 30-day maximum, and the 10-day example for slower converters
9. [9]Meta for Government and Nonprofits: Driving optimisation with machine learning. “Run no more than five ads per ad set for optimal delivery.”
10. [10]Datadrew: How long to test Facebook ads before you kill them. Practitioner floors: 5 delivery days, 1,000 impressions, 100 link clicks, 15 to 20 purchases; the 3x target CPA early kill; 20 purchases before raising budget
11. [11]LeadEnforce: How long Facebook ads should run before judging results. Practitioner floors: 3 to 5 days and 2,000 to 3,000 impressions with 100+ clicks for traffic; 7 to 10 days and 50 to 100 conversions for conversion campaigns
12. [12]AdManage: When to kill a Facebook ad. Immediate kills for broken tracking or wrong destination; the 2.3x and 3.0x target CPA thresholds at zero conversions; 24 to 72 hour reporting lag; CTR screening at 1,000 impressions