Google capped Learning at 7 days. Meta still counts events, not days

Google's Smart Bidding help page now says the "Learning" status "never lasts more than 7 days." Meta's learning phase has no such clock: an ad set exits after about 50 results in the week after its last significant edit, and if it cannot reach that pace it reads "Learning limited" instead. One platform put a ceiling on a label. The other set a volume threshold. Treat them as the same measurement and you will judge campaigns before they have finished telling you anything.

Google rewrote and renamed that documentation on October 21, 2026. The page that was called "Duration of the learning period for campaigns and what affects it" is now titled "About Smart Bidding learning", and Search Engine Roundtable, which broke the change, called it a major rewrite rather than a headline edit. The substance of the rewrite matters more than the title: Google now separates the status you can see from the time bidding actually needs to stabilise.

What Google actually put in writing

Three sentences carry the whole change.

"Learning status is a helpful reminder in your Google Ads account that your campaign configuration has recently changed. It never lasts more than 7 days."

"While you may see a 'Learning' status in your account for up to 7 days, your ads continue to run and serve as normal."

"Smart Bidding continuously optimizes and learns in the background using historical and new conversion data."

Read together, the seven days belongs to the label, not to the model. Google now lists exactly three triggers: a new or reactivated bid strategy, a composition change (campaigns, ad groups or keywords added to or removed from the strategy), and a goal change. It is equally explicit about what does not trigger it — budgets, ROI targets, targeting settings and creative edits never produce "Learning" status, and modifying creatives never removes historical data.

That second list is the practical gift. On Google you can change budget and creative without restarting any indicator you can see.

What Meta actually put in writing

Meta defines the phase in volume terms, not calendar terms.

"ad sets exit the learning 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."

Two qualifiers do the damage: about 50, and in the week after the last significant edit. There is no day on which Meta declares the phase finished. An ad set that trickles out five conversions a week simply never gets there, and Meta's troubleshooting page says exactly that — an ad set becomes Learning limited "when it is unlikely to receive about 50 optimization events in the week after your last significant edit". Meta is careful to add that Learning limited "isn't a penalty"; it is a diagnosis that your current setup cannot produce volume. (For Shops ads the bar is stated differently: a minimum of 17 purchases through your website and 5 through Meta.)

Meta is just as specific about what restarts the phase. Always significant: any change to targeting, to ad creative, to the optimization event, adding a new ad to the ad set, pausing the ad set for seven days or longer, and changing bid strategy. Significant only by magnitude: ad set spending limit, bid control or cost-per-result goal, and budget amount — Meta's own illustration is that going from $100 to $101 probably will not re-enter learning, while $100 to $1,000 may.

So the honest summary is short. Google guarantees the label disappears within a week. Meta guarantees only that a stable ad set is one that produced roughly 50 of the event you asked for, inside seven days, starting from the last thing you touched.

The same word, two measurements

The trap is that the word "Learning" appears in both interfaces. An advertiser running the two platforms side by side assumes one clock is ticking in both places. On Google the indicator is a receipt for a recent edit. On Meta it is a forecast about your volume. Day eight arrives, Google says nothing more because it never promised anything past the label, and Meta may be using the same column to tell you that your ad set will not stabilise at this pace at all.

Our own read: the demand was there, our answer was not earning it

We pulled 28 days from our own Search Console property (September 21 to October 19, 2026) across the four posts that already cover this ground:

That is about 3,135 impressions a month landing on pages sitting in positions 5.9 to 6.2, and the two pages that most directly answer "how long do I wait?" convert at 0.1% and 0.4%. We are not going to claim those numbers say anything about Google or Meta — they are different queries with different intents, and we are explicit about that. What they say is about us: people are asking the timing question in this cluster, the position is already good, and the answer we serve is not worth the click. That is the gap this piece is here to close, and the reason should you kill or keep a learning ad set was never enough on its own.

The rule we use before calling a number

We manage Meta and Google spend for clients, and this is the argument we have in the first week of almost every account: someone wants to judge a campaign on day three. The question underneath is never "how long is learning" — it is "when is this number safe to act on". Three gates, in order:

1. Status gate — is the indicator even relevant? On Google it disappears within seven days by documentation, and since budgets, targets and creatives never trigger it, a cleared status tells you a structural change happened, not that the model is done. On Meta, add the two columns Meta documents for this exact purpose: Last significant edit and Results. If you cannot see when the clock restarted, you cannot read the number that follows it.

2. Volume gate — is the sample big enough to mean anything? On Meta the threshold is published: roughly 50 optimization events per ad set in the week after the last significant edit, so the check is arithmetic — does this week's pace reach it? Google publishes no exit threshold for the model, which is precisely why we pre-register our own minimum sample with the client before launch: the number of conversions that must exist before any budget or structure decision is allowed. Set after launch, that number always bends to whatever the data happened to do.

3. Change gate — what did you touch? Keep a dated edit log per campaign. On Meta, any of the six always-significant edits restarts the count from zero, which is how a "stable" account is quietly reset every fortnight by someone improving it. On Google, budget, creative and targeting edits do not move the visible status at all, so a stalled result after those edits points somewhere other than learning.

If the status has cleared, the volume gate is not met, and you changed something significant yesterday, you do not have a performance problem. You have an unfinished experiment, and the honest action is to wait, not to relaunch.

That framing is not theoretical for us. The same question keeps surfacing in practitioner communities this month — accounts with a handful of installs asking whether to let the campaign "learn" or whether they are simply burning budget — and it is the exact question the three gates are built to settle in one meeting instead of three relaunches. With Black Friday and the holiday relaunch window weeks away, the temptation to restart something that is working is about to peak, and on Meta every one of those restarts costs you the week you just bought.

If you want the mechanics behind the kill-or-keep call, the Smart+ versus Advantage+ comparison covers where consolidation helps, and what counts as an auto-enhancement explains the creative edits that do not touch your data at all. When a number has to survive a board meeting, this is the Digital Ads & PPC engagement, and the fastest way in is a call.

Seven days is how long Google promises to show you a label. Fifty events is how many Meta promises to need. Neither one tells you whether your campaign works — that is what the gates are for.