You Edit Your Ads Every Day and They Don't Work: Meta's Learning Phase

You open Ads Manager, see fewer leads than yesterday, and change the text. The next day you swap the image. Then you narrow the audience. The ads “somehow” don’t work. The ads were learning, you were editing, and who gave up waiting for whom is an open question. Below: what Meta calls the learning phase, what Meta’s help page and practitioners say about it, what to do on a small budget, and how to keep a simple change log.

We checked Meta’s help page “About the learning phase” on 28 October 2026. The page loads its text with a script, so we do not quote it word for word: what follows is a paraphrase of its content. Button names depend on the interface version and may differ.

1. What the learning phase is

When you launch a new ad set, or change one significantly, Meta’s system spends several days finding out who to show the ads to. During this time results are unstable and leads are usually more expensive. That is the learning phase.

The official benchmark from Meta’s help page: to exit learning, an ad set needs about 50 optimization events within roughly 7 days after its last significant edit. An optimization event is whatever the ad is set up to get: a lead, a purchase, a message and so on. If Meta sees that an ad set will not reach this threshold, it shows the status “Learning limited”.

An important clarification: 50 is a benchmark per ad set, not per account. Three ad sets with 20 leads a week each are three ad sets that are not learning, not 60 “collected” events.

2. What exactly resets learning

According to the help page and to what practitioners describe, learning can restart after:

• a change of audience or targeting;
• editing or adding an ad (new text, image, video);
• a change of the event the ad optimizes for;
• a change of bid strategy;
• pausing the ad set for a week or longer;
• a sharp change of budget.

On budget, Meta’s help page, as we understood it, gives no single number: whether a change is significant depends on its size. Practitioners say a change of roughly up to 10–30% usually does not reset learning. That is the observation of several practitioners, not a Meta rule, so test it in your own account. Jumping from $10 to $50 a day is an example of the sharp change we mean.

3. How many events you really need: where opinions differ

Meta’s official page holds to “about 50 within 7 days”. Practitioners have other observations, and we present them as observations, not as a rule.

• Some practitioners wrote that some advertisers saw a shortened learning phase: about 10 events in 3 days instead of 50 in 7. By their own account this did not apply to everyone and later went back to 50 in 7. We found no official confirmation from Meta.
• Others report that for purchases and app installs in Advantage+ the threshold is lowered to 10. Meta does not confirm this; the source is an article by a third-party author.
• And the reverse: in practice, they say, stability comes at 70–100 events, not 50.

The takeaway for a business owner is simple: plan for 50 a week per ad set, and if Meta exits learning earlier, treat it as a pleasant bonus you should not count on.

4. Why daily edits hurt twice

The first harm: the ads never get to finish learning. If you change something noticeable every day, each edit may restart learning. The ads spend almost all their time in the phase where leads cost more.

The second harm: you cannot tell what caused what. If you changed five things in a week and the result got better or worse, you do not know what worked. You can no longer tell the effect of an edit from ordinary fluctuation.

So the rule is: one significant change, then at least a week of waiting. Judge the result on day 5–7 by cost per lead, not on day two by number of clicks. We wrote about how long to wait for ad results in the article How Long to Wait for Ad Results.

5. What to do if you cannot reach 50 leads a week

The first step is arithmetic. Multiply your cost per lead by 50: that is roughly the weekly budget one ad set needs to reach the threshold. An example purely for the calculation: if a lead costs $10, 50 leads is $500 a week per ad set. If you have less, learning may not finish even in a week. For a small business that is normal and no reason to panic.

What to do then (this is practice, not a guarantee):

1. Fewer ad sets. One ad set with the whole budget learns faster than five with pocket change.
2. An event closer to the start of the customer journey. If there are too few purchases, you can optimize for an earlier step, such as a lead or a message. The downside: such leads may be of lower quality, so check them against your CRM.
3. Fewer edits. On a small budget every learning reset costs more, because there are few events to begin with.
4. A longer evaluation window. Look at the trend over 2–3 weeks, not a single day.
5. No budget jumps. If you raise the budget, do it gradually.

We wrote about what Meta shows in the account versus what ends up in the CRM in Meta Shows 50 Leads, the CRM Shows 40.

6. How to keep a change log

The log needs no special service: a spreadsheet or a note is enough. Record:

• the date of the change;
• what exactly you changed (text, image, audience, budget, in one line);
• why you changed it (a hypothesis: “I think nobody reads this image”);
• the “before” figure: cost per lead and number of leads over the last week;
• the check date, 5–7 days later, and the “after” figure.

Ads Manager itself has a change history (Activity history; the name in your interface may differ): it shows who changed what, but it does not say why and does not lead you to a conclusion. That is why your own log with a “why” column is more useful. And for when an ad has really “burned out” and the image should be changed, see our article on creative fatigue.

From our practice

Most often we see not “bad ads” but ads that were not allowed to finish learning. The owner logs in every day, sees today’s noise and rushes to fix it. When we agree on the rule “one change, one week” and keep a log, the conversation has facts instead of feelings: “from 5 to 12 November this text ran, and the cost per lead was this”. This is our working experience, not statistics.

Summary

1. Every significant change in a Meta ad set can restart learning: a new audience, image, text or a sharp budget jump.
2. The official benchmark is about 50 optimization events in 7 days per ad set. Claims that “10 is enough” are unofficial practitioner observations, not a rule.
3. One change at a time and at least a week of waiting; judge the result by cost per lead on day 5–7.
4. On a small budget: fewer ad sets, an event closer to the start of the customer journey, a longer evaluation window.
5. The log: date, what changed, why, the figure before and after.

See how we set up and evaluate advertising in the Google Ads for business section.

Want a free audit of your Google Ads account? Send us a request: we’ll look at your account and tell you where you can not only change settings but also wait for the result.