[{"data":1,"prerenderedAt":616},["ShallowReactive",2],{"guide-facebook-ad-learning-phase":3},{"id":4,"title":5,"answer":6,"authorId":7,"body":8,"category":495,"ctaVariant":496,"dataset":495,"description":497,"examples":498,"extension":499,"faqs":500,"heroImage":525,"intro":526,"meta":527,"navigation":528,"path":529,"publishedAt":530,"seo":531,"sources":532,"stats":572,"stem":614,"updatedAt":530,"__hash__":615},"blog\u002Fblog\u002Ffacebook-ad-learning-phase.md","Facebook Ad Learning Phase Explained (2026)","The learning phase is the unstable window after you launch or significantly edit an ad set, when Meta's delivery system is still figuring out who to show the ad to and how to bid. To exit, Meta's guidance is that an ad set needs roughly 50 optimization events (purchases, leads, or whatever you optimized for) within about 7 days of the last significant edit. 'Learning Limited' is not a punishment: it means the ad set is unlikely to ever reach 50 events in that window, usually because the budget, audience, or event is too thin. On adding a new ad, the documented rule and the observed behaviour diverge: Meta's Significant Edits page lists adding an ad among the edits that reset learning, while Jon Loomer's 2025 testing found that adding a single ad to an already-established ad set often did not reset it in practice. Plan against the documented rule, but note that editing the budget or audience of a running ad is a reset either way, so a brand like Shakura is still better off spinning a winning before-and-after into fresh variations than tuning the ad already live.","likit-sae-lee",{"type":9,"value":10,"toc":476},"minimark",[11,16,20,23,26,30,33,36,41,44,48,51,58,62,65,79,87,91,94,102,106,109,112,179,187,195,199,202,205,279,282,288,292,295,298,311,314,318,321,324,327,377,380,383,386,424,432,436,439,447,455,459,462,465,468],[12,13,15],"h2",{"id":14},"what-the-learning-phase-actually-is","What the learning phase actually is",[17,18,19],"p",{},"When you publish a new ad set, Meta does not yet know who your best buyers are. The learning phase is the window where the delivery system experiments: it tries different people, different placements, different times of day, and different bid amounts, watching which combinations produce the outcome you optimized for. Until it has seen enough of those outcomes, it is guessing, and guessing is expensive. That is why your cost per result lurches around in the first few days, why the same audience can look like a bargain on Monday and a disaster on Wednesday, and why the \"Learning\" status sits at the ad-set level the whole time.",[17,21,22],{},"The phase ends one of two ways. Either the ad set gathers enough signal and the status flips to active (delivery stabilizes, costs settle), or it never gathers enough and the status becomes \"Learning Limited,\" which is a different problem we will get to. The thing to internalize first is that the volatility is not a malfunction. It is the system doing exactly what it is supposed to do: spend a little inefficiently up front to buy the information that makes the rest of the campaign efficient.",[17,24,25],{},"This reframes the whole emotional experience of a launch. The advertisers who lose money in the learning phase are usually the ones who treat early noise as a verdict and start intervening. The ones who come out of it cheap are the ones who decided their structure before launch and then left it alone long enough for the system to finish learning.",[12,27,29],{"id":28},"why-roughly-50-events-in-7-days","Why roughly 50 events in 7 days",[17,31,32],{},"Meta's documented guidance is that an ad set needs about 50 optimization events within roughly 7 days of its last significant edit to exit the learning phase. That number is not arbitrary, and it is not a quota you are being graded against. It is the minimum volume of outcomes the delivery model needs to make statistically reliable decisions. Below it, the system is drawing conclusions from too few data points, so it cannot confidently tell a good audience pocket from a lucky run of conversions.",[17,34,35],{},"The \"event\" is whatever you chose as your optimization goal. If you optimized for purchases, you need roughly 50 purchases; if you optimized for leads, 50 leads; if you optimized for add-to-carts, 50 of those. This matters enormously for budget planning, because, say, a $40 purchase is a far heavier lift to hit 50 times in a week than a $5 lead. The threshold is per ad set, and it resets to zero on every significant edit, which is the single fact most advertisers underestimate.",[37,38,40],"h3",{"id":39},"why-your-objective-decides-how-fast-you-exit","Why your objective decides how fast you exit",[17,42,43],{},"The 50-events math quietly explains a pattern that confuses a lot of advertisers: two ad sets with the same budget exit the learning phase at completely different speeds. The reason is the optimization event. A higher-funnel objective optimizes for a cheap, frequent action: a landing-page view or a link click happens many times for every purchase, so an ad set optimizing for it can clear 50 events in hours rather than days. A lower-funnel purchase objective optimizes for a rare, expensive action, so the same spend produces a trickle of events and the phase drags. This is the single simplest lever for a stuck conversion campaign: if purchases are too infrequent to reach 50 in a week, optimize for an earlier event (add-to-cart or lead), let the ad set stabilize on volume, then move the optimization deeper once it has a delivery history to build on.",[37,45,47],{"id":46},"the-reduced-event-exception-kept-honest","The reduced-event exception, kept honest",[17,49,50],{},"There is a known exception worth flagging so you are not surprised by it. A few specific objective and event setups (some Purchase and app-install configurations) can exit on fewer events than the canonical 50. Meta does not publish one universal number that holds across every objective, so treat this as objective-specific rather than a new rule, and confirm what your own campaign type shows in Ads Manager. The number to teach yourself and plan around is still 50 in 7. Build for the canonical case and any lower threshold only ever helps you.",[17,52,53],{},[54,55],"img",{"alt":56,"src":57},"Diagram of the Facebook ad learning phase: a horizontal track running from Launch through a turbulent Learning window to an Active state, with a counter filling toward 50 optimization events in 7 days, and a red Reset arrow looping any significant edit back to the start.","\u002Fimages\u002Fblog\u002Ffacebook-ad-learning-phase-diagram.webp",[12,59,61],{"id":60},"tracking-is-the-input-the-learning-phase-actually-counts","Tracking is the input the learning phase actually counts",[17,63,64],{},"Budget and audience get all the blame for a stuck ad set, but there is a quieter culprit that fits every \"my numbers should be fine and it still will not exit\" case. The learning phase does not count conversions that happen on your site. It counts the optimization events Meta can match back to a real person who saw the ad. If your tracking sends weak signal, Meta credits fewer of those events, and the ad set crawls toward 50 even when the budget could buy them three times over.",[17,66,67,68,73,74,78],{},"Two things decide how much signal lands. The first is coverage: the browser Meta Pixel alone loses events to ad blockers, cookie limits, and iOS restrictions, so a server-side feed of the same events through the Conversions API keeps the data flowing when the browser drops it. The catch is that running both means each conversion arrives twice, once from the Pixel and once from the server, so you deduplicate them with a shared event ID and Meta counts the pair as one. Miss the dedup and you either double-count (inflated, misleading reports) or, worse, lose the cleaner server event entirely. The mechanics are worth getting right, and the full walkthroughs live in the ",[69,70,72],"a",{"href":71},"\u002Fblog\u002Ffacebook-conversions-api-setup","Conversions API setup guide"," and the ",[69,75,77],{"href":76},"\u002Fblog\u002Fmeta-pixel-capi-deduplication","Pixel and CAPI deduplication guide",".",[17,80,81,82,86],{},"The second is match quality. Meta scores how well the customer data you send (email, phone, name, IP, device) matches a known account on a scale Meta documents from 0 to 10, called event match quality, and higher quality matching may improve attribution and delivery. The practical link to the learning phase is direct: a low event match quality score means a chunk of your real conversions never get matched, so they never count toward the 50, and an ad set that should clear the threshold in four days stalls instead. If an ad set is stuck with adequate budget and a broad enough audience, check event match quality before you touch anything else. The ",[69,83,85],{"href":84},"\u002Fblog\u002Fmeta-event-match-quality-emq","event match quality guide"," covers how to read and lift the score.",[37,88,90],{"id":89},"the-attribution-window-decides-when-an-event-counts","The attribution window decides when an event counts",[17,92,93],{},"There is one more setting that quietly governs the 50: the attribution window, which is the period after a click or view during which Meta will still credit a conversion to the ad. Meta's default counts conversions on a 7-day-click plus 1-day-view basis, and on January 12, 2026 Meta removed the longer 7-day-view and 28-day-view windows, leaving one day as the longest view window available. The windows that remain are 1-day click, 7-day click, 28-day click, 1-day engaged view, and 1-day view.",[17,95,96,97,101],{},"Why this matters for learning: if your buyers take several days to decide, a conversion that fires on day six only counts toward the 50 if your window is still open when it lands. Choose a short 1-day-click window on a product with a long decision cycle and you will see events fire in your store that never register against the ad set in time, so it looks slower to exit than it really is. Match the window to how long your customers actually take to buy, and read the ",[69,98,100],{"href":99},"\u002Fblog\u002Ffacebook-ads-attribution-window","attribution window guide"," before you change it, because the setting also reshapes every number in your reports.",[12,103,105],{"id":104},"what-learning-limited-really-means","What Learning Limited really means",[17,107,108],{},"\"Learning Limited\" reads like a failing grade, and most advertisers respond to it with panic. It is not a penalty. It is a diagnosis. The status means the ad set is unlikely to ever reach about 50 optimization events inside the 7-day window, so Meta is telling you, plainly, that there is not enough signal here for delivery to fully optimize. The fix is to remove whatever is starving it of events.",[17,110,111],{},"The causes fall into four buckets, and each has a clear remedy:",[113,114,115,131],"table",{},[116,117,118],"thead",{},[119,120,121,125,128],"tr",{},[122,123,124],"th",{},"Cause of Learning Limited",[122,126,127],{},"What it looks like",[122,129,130],{},"The fix",[132,133,134,146,157,168],"tbody",{},[119,135,136,140,143],{},[137,138,139],"td",{},"Budget too low",[137,141,142],{},"Spend cannot plausibly buy 50 events in a week",[137,144,145],{},"Raise the daily budget, or pick a cheaper optimization event",[119,147,148,151,154],{},[137,149,150],{},"Audience too small",[137,152,153],{},"A narrow retargeting pool or tight interest stack",[137,155,156],{},"Broaden the audience, or lean on Advantage+ Audience expansion",[119,158,159,162,165],{},[137,160,161],{},"Event too rare",[137,163,164],{},"Optimizing for a deep, infrequent action like purchase",[137,166,167],{},"Optimize for an earlier event (add-to-cart, lead) that fires more often",[119,169,170,173,176],{},[137,171,172],{},"Too many ad sets",[137,174,175],{},"Five small ad sets each running a separate learning cycle",[137,177,178],{},"Consolidate into fewer, better-funded ad sets",[17,180,181,182,186],{},"Budget fragmentation is the quiet killer on that list. Five ad sets at a small daily budget each run five separate learning cycles, all of them starved, none of them reaching 50. Collapse them into one consolidated ad set and the same total spend concentrates into a single signal that can actually clear the threshold. This is also where campaign-level budgeting earns its keep: Advantage+ Campaign Budget (Meta's current name for what used to be CBO) lets the system push spend toward whichever ad set is generating efficient events fastest, so the campaign reaches 50 sooner than rigidly splitting budget by hand ever would. One caution comes with that convenience: because an Advantage+ Campaign Budget is shared across every ad set under it, a single campaign-level budget edit can ripple to all of them at once, so a change you meant for one ad set can restart learning across the whole campaign. The trade-offs between campaign-level and ad-set budgets are worth understanding before you pick one, and the ",[69,183,185],{"href":184},"\u002Fblog\u002Fcbo-vs-abo-facebook-ads","CBO versus ABO guide"," lays them out.",[17,188,189,190,194],{},"Audience overlap is the subtler version of fragmentation, and it bites hardest exactly when you think you are scaling. Duplicate a winning ad set to \"run it twice\" and the two copies often target overlapping people, so they enter the same auction and compete against each other for the same impressions. You end up bidding against yourself, splitting the signal that each ad set needs to reach 50, and paying more for the privilege. The tell is two sibling ad sets whose costs both creep up after you launch the duplicate. Meta's audience overlap tool shows how much two audiences share, and the fix is to consolidate, exclude one audience from the other, or simply not run near-identical ad sets side by side. The ",[69,191,193],{"href":192},"\u002Fblog\u002Ffacebook-audience-overlap","audience overlap guide"," covers how to detect and resolve it.",[12,196,198],{"id":197},"the-edits-that-send-you-back-to-zero","The edits that send you back to zero",[17,200,201],{},"Here is the part that costs advertisers the most money, because it is the part that feels most like progress. Every significant edit discards the signal accumulated so far and restarts the learning phase from zero. Per Meta's \"Significant Edits and the Learning Phase\" guidance, the edits that trigger a reset include changing the optimization event, changing the targeting or audience, changing the creative, changing the bid strategy or your bid and cost controls, and making a large budget change.",[17,203,204],{},"That last one needs a caveat, and it is a bigger one than most articles admit. A budget change resets learning once it crosses a \"significant\" threshold, but Meta publishes no percentage for that threshold anywhere. Its Significant Edits page gives a magnitude illustration instead: $100 to $101 \"isn't likely\" to reset, $100 to $1,000 \"may\". The roughly 20% figure you see quoted is a practitioner rule of thumb that Meta never published, so treat it as a useful ceiling rather than a line the platform computes against. A tiny budget nudge is usually safe; doubling a budget is not.",[113,206,207,217],{},[116,208,209],{},[119,210,211,214],{},[122,212,213],{},"Edit",[122,215,216],{},"Resets learning?",[132,218,219,227,234,241,248,255,263,271],{},[119,220,221,224],{},[137,222,223],{},"Change the optimization event",[137,225,226],{},"Yes",[119,228,229,232],{},[137,230,231],{},"Change the audience or targeting",[137,233,226],{},[119,235,236,239],{},[137,237,238],{},"Change the creative on the existing ad",[137,240,226],{},[119,242,243,246],{},[137,244,245],{},"Change bid strategy, bid cap, or cost cap",[137,247,226],{},[119,249,250,253],{},[137,251,252],{},"Large budget change (scale-dependent; the roughly 20% line is a rule of thumb, not Meta's)",[137,254,226],{},[119,256,257,260],{},[137,258,259],{},"Pause the ad set for ~7 days, then resume",[137,261,262],{},"Yes (re-enters learning)",[119,264,265,268],{},[137,266,267],{},"Duplicate the ad set (creates a new one)",[137,269,270],{},"Yes (the copy learns from zero)",[119,272,273,276],{},[137,274,275],{},"Fix a typo, small budget nudge",[137,277,278],{},"Generally no",[17,280,281],{},"Read that table next to the volatility of a fresh launch and the picture sharpens: an advertiser who reacts to a bad first day by swapping the audience has not fixed anything. They have thrown away three days of accumulated signal and started a new bad first day. Constant tinkering is what keeps an ad permanently expensive. It is the optimization equivalent of digging up a seed every morning to check whether it has rooted.",[17,283,284],{},[54,285],{"alt":286,"src":287},"Editorial illustration contrasting two advertiser habits: on the left a hand endlessly tweaking dials and sliders on a wilting ad set that keeps resetting, on the right a steady conveyor adding fresh creative cards into a thriving, stable ad set, with the right side glowing in indigo-to-magenta gradient.","\u002Fimages\u002Fblog\u002Ffacebook-ad-learning-phase-velocity.webp",[12,289,291],{"id":290},"the-2026-shift-add-ads-do-not-edit-them","The 2026 shift: add ads, do not edit them",[17,293,294],{},"For years the standard advice was that any change to a live ad set, including adding a new ad, restarted learning, which left advertisers terrified to touch a winner. That advice has aged. Independent testing by Jon Loomer in 2025 found that adding a single new ad to an ad set already running 22 ads kept the ad set active rather than resetting it. The honest framing is \"often no longer resets,\" with a real caveat: adding many ads at once to a thin ad set is still more likely to trigger learning, and this comes from industry testing rather than an explicit Meta policy statement.",[17,296,297],{},"But the direction of the finding changes the optimal play. If editing the existing ad's budget, audience, or copy each wipes your signal, while dropping a fresh creative into an established ad set usually does not, then the cheapest way to chase performance is no longer to optimize the ad you have. It is to ship the next one. You keep the learning you paid for and put a new contender into the same pool, where Meta can test it without resetting the rest.",[17,299,300,301,305,306,310],{},"This is why creative velocity, not manual tinkering, is the real lever. The independent meta-analysis NCSolutions ran, reported by Marketing Charts, attributes about 49% of advertising's incremental sales lift to the creative itself, more than brand, targeting, reach, and recency. Half the result lives in the ad, not the settings. So the smartest thing you can do with a stabilized ad set is feed it new variations rather than edit the one already winning. The same discipline that beats ",[69,302,304],{"href":303},"\u002Fblog\u002Ffacebook-ad-fatigue","ad fatigue"," (refresh with new creative, the reset-free way to fight it) is the discipline that keeps your learning intact, and it pairs naturally with a structured approach to ",[69,307,309],{"href":308},"\u002Fblog\u002Ffacebook-ad-creative-testing","creative testing"," (ship variations instead of editing one ad to keep learning intact).",[17,312,313],{},"Real ad examples show what that variation engine looks like in practice. Shakura runs before-and-after creative on stubborn dark spots faded without laser, anchored to a concrete two-session offer, and that single proven angle can spin into a dozen fresh framings, different hooks, different opening frames, each one a new ad you add rather than an edit you make. Glad2Glow runs a price-led bundle, a five-piece set at RM58.50, where the optimization event is a clean purchase and the 50-events math is tangible for a small direct-to-consumer budget. Mootiara puts a founder on camera pulling on a wrinkle-free hijab in one motion, the kind of low-cost user-generated variation you can reshoot weekly. Desire Gym runs transformation carousels where each sign-up enquiry is a lead event accumulating toward the threshold. Four brands, four reasons to add the next creative instead of editing the last one.",[12,315,317],{"id":316},"a-budget-and-structure-plan-that-exits-learning-fast","A budget-and-structure plan that exits learning fast",[17,319,320],{},"Most learning-phase pain is decided before launch, in the structure. Get the structure right and the no-touch period takes care of itself. Get it wrong and no amount of mid-flight editing rescues it, because the editing is the problem.",[17,322,323],{},"Start with the math. The planning rule of thumb is to size each ad set's daily budget at roughly (your target cost per event times 50) divided by 7. A $20 target cost per event implies around $140 a day to have a realistic shot at 50 events inside the week. That is a community heuristic derived from Meta's own threshold, not an official Meta formula, so treat it as a sanity check rather than a guarantee, but it stops you from launching an ad set that was mathematically doomed to Learning Limited from the first hour.",[17,325,326],{},"Put it in a table so the floor is obvious before you spend. The figures below are illustrative, run in USD, and come straight from the (cost per event times 50) divided by 7 heuristic, not from live spend:",[113,328,329,342],{},[116,330,331],{},[119,332,333,336,339],{},[122,334,335],{},"Target cost per event (illustrative)",[122,337,338],{},"Daily budget to aim at 50 events in 7 days",[122,340,341],{},"Why it matters",[132,343,344,355,366],{},[119,345,346,349,352],{},[137,347,348],{},"$5",[137,350,351],{},"~$36 a day",[137,353,354],{},"A cheap lead or add-to-cart clears the threshold on a small budget",[119,356,357,360,363],{},[137,358,359],{},"$20",[137,361,362],{},"~$143 a day",[137,364,365],{},"A mid-priced lead or sign-up needs a real daily commitment",[119,367,368,371,374],{},[137,369,370],{},"$50",[137,372,373],{},"~$357 a day",[137,375,376],{},"A direct purchase event demands serious budget, or a cheaper event",[17,378,379],{},"Read down that column and the lesson lands on its own: if your only realistic budget is $40 a day but your purchase costs $50, the math says you cannot buy 50 purchases in a week, so optimizing for purchase was never going to exit learning. Pick a cheaper event or raise the budget. The arithmetic decides it before launch, which is exactly the point of doing it first.",[17,381,382],{},"The benchmark numbers matter here because they ground the math in reality. WordStream's 2025 Facebook Ads benchmarks (drawn from data spanning April 2024 to June 2025) put the average cost per lead at $27.66, up 20.94% year over year, against an average lead conversion rate of 7.72%. Traffic objectives ran cheaper at a $0.70 average CPC, down 6.67% year over year. Click-through averaged 2.59% on Leads and 1.71% on Traffic. Plug your own target cost into the sizing rule and you can see immediately whether your budget can buy enough events, or whether you should optimize for an earlier, more frequent action instead.",[17,384,385],{},"Then make four pre-launch decisions and commit to them:",[387,388,389,397,403,409],"ol",{},[390,391,392,396],"li",{},[393,394,395],"strong",{},"Consolidate, do not fragment."," Fewer, better-funded ad sets concentrate the signal. Lean on Advantage+ Campaign Budget so Meta routes spend to the ad set clearing events fastest.",[390,398,399,402],{},[393,400,401],{},"Size the audience to feed the threshold."," Advantage+ Audience is the default in most objectives now, treating your age, location, and interest inputs as suggestions it can expand beyond, which keeps the pool large enough to generate events. Note that Meta consolidated detailed-targeting interests from June 23, 2025, with discontinued options ceasing delivery on January 15, 2026, so verify your interest selections still exist.",[390,404,405,408],{},[393,406,407],{},"Pick an event you can actually hit 50 times."," If purchases are too rare for your budget, optimize for add-to-cart or lead and let volume do its job.",[390,410,411,414,415,73,419,423],{},[393,412,413],{},"Decide the creative slate up front."," Launch an ad set with enough variety for delivery to learn, a small handful of genuinely distinct concepts (a spread of roughly three to five is a common starting point), not one lonely ad and not twenty thin ones that fragment a modest budget. Dynamic creative is the built-in way to feed that variety: you upload several headlines, images, and primary texts and Meta mixes them into combinations, which is a clean way to test variation inside a single ad set. The counts and the trade-offs sit in the ",[69,416,418],{"href":417},"\u002Fblog\u002Fhow-many-facebook-ad-creatives-per-ad-set","creatives-per-ad-set guide",[69,420,422],{"href":421},"\u002Fblog\u002Ffacebook-dynamic-creative","dynamic creative guide",". Decide the slate now so you add new variations into the running ad set later rather than editing the originals.",[17,425,426,427,431],{},"Setting all of this correctly before you publish is the whole game, and the full ",[69,428,430],{"href":429},"\u002Fblog\u002Ffacebook-ad-campaign-setup","campaign setup walkthrough"," covers it step by step (set the ad set up right before launch so you do not have to make resets-triggering edits).",[12,433,435],{"id":434},"the-no-touch-week-and-reading-the-status","The no-touch week, and reading the status",[17,437,438],{},"Once it is live, the best optimization is restraint. Hold a no-touch policy for about 7 days unless a critical issue forces your hand: a rejected ad, a billing failure, a tracking break. Outside of those, leave the budget, the audience, and the creative alone and let learning finish. Judging an ad set in its first 48 hours is reading noise, and the cost of reacting is a full reset that buys you another 48 hours of noise.",[17,440,441,442,446],{},"To watch progress, add the Delivery status column in Ads Manager so you can see \"Learning,\" \"Learning Limited,\" or active at the ad-set level, and track the event count climbing toward 50. If you want to understand how to read those signals alongside the rest of your numbers, the ",[69,443,445],{"href":444},"\u002Fblog\u002Ffacebook-ad-reporting","reporting guide"," covers it (read the learning status and event volume in your reports). Expect costs to run higher while an ad set is still learning, and remember that a strong opening hook helps you hit 50 faster by pushing more people toward the event that exits learning.",[17,448,449,450,454],{},"It helps to remember why all this rigor pays off at scale. Meta's full-year 2025 results put worldwide advertising revenue at $196.18 billion, up 22% year over year, and worldwide average revenue per user reached a record $57.03 for the year. That is the demand pressure behind the auction every learning ad set fights through, so the system rewards advertisers who give it clean, concentrated signal and taxes the ones who keep starving and resetting it. The whole approach fits inside the larger picture in ",[69,451,453],{"href":452},"\u002Fblog\u002Fhow-to-run-a-facebook-ad","how to run a Facebook ad"," (the full walkthrough this fits into), and a platform like AdPlay.ai keeps research, generation, editing, and Meta launch in one place, which shortens the gap between deciding to ship the next variation and actually shipping it. The habit is the point: when an ad set is learning, the move is patience; when it has learned, the move is the next ad, not the next edit.",[12,456,458],{"id":457},"how-to-scale-after-you-exit-the-learning-phase","How to scale after you exit the learning phase",[17,460,461],{},"Exiting learning is not the finish line, it is the green light. The ad set is stable and profitable, and now the instinct is to pour budget in. That instinct is the second most expensive mistake in this whole topic, right behind tinkering during learning, because a big budget jump is a significant edit, so the very move meant to grow a winner resets it and drops you back into volatility. Scaling well is mostly about respecting the same reset rules you just learned, on the way up.",[17,463,464],{},"There are two directions to scale, and they answer different questions. Vertical scaling means putting more budget into the winning ad set you already have: simplest, but capped by the reset threshold and by how much your audience can absorb before frequency climbs. Horizontal scaling means widening the surface: duplicating the winner into new audiences or placements, or adding fresh creative into the same pool, so you grow reach without leaning the whole budget on one ad set. Most accounts that scale cleanly do a bit of both, nudging the winner up while opening new lanes beside it.",[17,466,467],{},"For the vertical move, the community heuristic (guidance, not a Meta-published number) is to raise the daily budget by roughly 10% to 20% every few days. That keeps each step small enough to sit at the harmless end of Meta's own magnitude illustration, so delivery re-stabilizes between bumps instead of resetting. Make one change, give it a few days of data, then decide the next step on results rather than on the first nervous afternoon. Tripling the budget overnight does the opposite of what you want: it resets learning and hands you a fresh, expensive exploration window at the higher spend.",[17,469,470,471,475],{},"For the horizontal move, know exactly what duplicating does. Editing the budget of a live ad set keeps its accumulated signal as long as you stay under the threshold. Duplicating it does not edit anything: the copy is a brand new ad set, so it starts the learning phase from zero while the original keeps running untouched. That is not a bug, it is the trade. Duplicate only a genuinely proven winner, expect the copy to be volatile for its own 7-day window, and avoid pointing it at an audience that overlaps the original, or the two will compete in the auction and split the signal. The full playbook for growing a winner without breaking it sits in the ",[69,472,474],{"href":473},"\u002Fblog\u002Fscale-facebook-ads","scaling guide",". The through line never changes: feed the system clean signal, change one thing at a time, and let it learn before you ask it to do more.",{"title":477,"searchDepth":478,"depth":478,"links":479},"",2,[480,481,486,489,490,491,492,493,494],{"id":14,"depth":478,"text":15},{"id":28,"depth":478,"text":29,"children":482},[483,485],{"id":39,"depth":484,"text":40},3,{"id":46,"depth":484,"text":47},{"id":60,"depth":478,"text":61,"children":487},[488],{"id":89,"depth":484,"text":90},{"id":104,"depth":478,"text":105},{"id":197,"depth":478,"text":198},{"id":290,"depth":478,"text":291},{"id":316,"depth":478,"text":317},{"id":434,"depth":478,"text":435},{"id":457,"depth":478,"text":458},null,"neutral","What the Facebook ad learning phase is, why an ad set needs roughly 50 events in 7 days to exit, what Learning Limited really means, and why shipping variations beats editing in 2026.",[],"md",[501,504,507,510,513,516,519,522],{"question":502,"answer":503},"How many conversions do I need to exit the Facebook ad learning phase?","Meta's guidance is roughly 50 optimization events within about 7 days of your last significant edit. The event is whatever you optimized for: 50 purchases, 50 leads, 50 add-to-carts, and so on. It is a signal-volume threshold, not a magic switch, so a clean, concentrated signal can exit a little sooner and a noisy one can stall. Some specific objectives reportedly exit on fewer events, but 50 in 7 days is the canonical number to plan around.",{"question":505,"answer":506},"How long does the Facebook ad learning phase last?","Up to about 7 days, but the clock is really counting events, not days. If your ad set hits roughly 50 optimization events in three days, it can exit early; if it never reaches 50 inside the 7-day window, it falls into Learning Limited instead. The phase also restarts from zero every time you make a significant edit, so an ad set you keep tweaking can stay volatile indefinitely.",{"question":508,"answer":509},"What does 'Learning Limited' mean and how do I fix it?","Learning Limited means the ad set is unlikely to ever get roughly 50 optimization events in the 7-day window, so Meta cannot fully optimize delivery. It is a diagnosis, not a penalty. The usual causes are a budget too low for your cost per event, an audience too small, an optimization event that fires too rarely, or too many overlapping ad sets splitting the signal. Fixes: raise the budget, broaden the audience, optimize for a more frequent event (add-to-cart instead of purchase), or consolidate fragmented ad sets into one.",{"question":511,"answer":512},"Does editing an ad reset the learning phase?","Significant edits do. Changing the optimization event, the audience or targeting, the creative, the bid strategy or cost controls, or making a large budget change (commonly cited at roughly 20% or more, but that percentage is practitioner guidance: Meta publishes no figure at all) restarts learning and discards the signal gathered so far. Minor edits, like fixing a typo in ad copy or a small budget nudge, generally do not. The safe rule is to treat any change to who, what, or how much as a reset.",{"question":514,"answer":515},"Does adding a new ad restart the learning phase in 2026?","The documented rule and the observed behaviour diverge here, and it is worth knowing both. Meta's Significant Edits page lists adding a new ad among the edits that reset learning. Independent testing by Jon Loomer in 2025 found that adding one new ad to an ad set already running 22 ads kept the ad set active rather than resetting it, so in practice a single fresh variation into an established ad set often does not reset. Meta has not updated the page to match, so treat the exception as undocumented and unreliable: plan against the documented rule, and batch creative rather than dripping it. The caveat compounds this, since adding many ads at once to a thin ad set is more likely to trigger learning. None of that changes the core play, because editing the budget, audience, or copy of the ad already running is a reset under both the rule and the testing, which is why shipping variations still beats editing the winner.",{"question":517,"answer":518},"How much budget do I need to get out of the learning phase?","Enough to plausibly generate 50 events in a week. A useful planning rule of thumb is daily budget per ad set of about (your target cost per event times 50) divided by 7. If your target cost per event is around $20, that implies roughly $140 a day to have a realistic shot at 50 events in seven days. This is a community heuristic derived from Meta's threshold, not an official Meta formula, but it keeps you from starving an ad set before it can learn.",{"question":520,"answer":521},"Can I increase my budget without resetting the learning phase, and how much is safe?","You can, within limits. A budget change only restarts learning once it crosses Meta's 'significant' threshold, but Meta publishes no percentage for where that threshold sits. What it publishes is a magnitude illustration: a move from $100 to $101 'isn't likely' to reset learning, while $100 to $1,000 'may'. So the roughly 20% line you see quoted everywhere is a community rule of thumb, not a Meta rule, and you should not write it into an SOP as though Meta published it. The community heuristic for scaling a winner without re-triggering learning is to raise the daily budget by roughly 10% to 20% every few days, which keeps each step small and gives delivery time to re-stabilize between bumps. If you need to grow faster than that, duplicate the proven ad set and scale the copy instead, but expect the duplicate to enter a fresh learning phase from zero because it is a brand new ad set.",{"question":523,"answer":524},"Will a broken Meta Pixel or poor tracking keep my ad set stuck in the learning phase?","Yes, and it is the cause most advertisers never check. The learning phase counts matched optimization events, not raw conversions, so if your Meta Pixel and Conversions API send weak signal (low event match quality, missing deduplication, events that never fire), Meta credits fewer of your conversions and the ad set crawls toward 50 even when budget and audience are fine. Confirm the events fire, send server-side events through the Conversions API, deduplicate them against the Pixel with a shared event ID, and push your event match quality up. Clean signal is often the difference between an ad set that exits in days and one that stalls in Learning Limited.","\u002Fimages\u002Fblog\u002Ffacebook-ad-learning-phase-hero.webp","Your ad set finally went live, and the numbers are all over the place: one day the cost per result looks brilliant, the next it doubles, and the 'Learning' tag just sits there mocking you. The instinct is to fix it, to nudge the budget, swap the audience, rewrite the copy. That instinct is the most expensive mistake in Meta advertising, because almost every fix you can think of sends the algorithm back to zero. The learning phase is not a problem to solve. It is a process to feed, and once you understand what it is counting, you stop sabotaging it.",{},true,"\u002Fblog\u002Ffacebook-ad-learning-phase","2026-09-11",{"title":5,"description":497},[533,537,540,544,547,550,553,557,560,563,566,569],{"label":534,"url":535,"year":536},"Meta Business Help Center, About the Learning Phase","https:\u002F\u002Fwww.facebook.com\u002Fbusiness\u002Fhelp\u002F112167992830700","2026",{"label":538,"url":539,"year":536},"Meta Business Help Center, Significant Edits and the Learning Phase","https:\u002F\u002Fwww.facebook.com\u002Fbusiness\u002Fhelp\u002F316478108955072",{"label":541,"url":542,"year":543},"WordStream \u002F LocaliQ, Facebook Ads Benchmarks 2025","https:\u002F\u002Fwww.wordstream.com\u002Fblog\u002Ffacebook-ads-benchmarks-2025","2025",{"label":545,"url":546,"year":543},"Meta, Reports Fourth Quarter and Full Year 2025 Results","https:\u002F\u002Fwww.prnewswire.com\u002Fnews-releases\u002Fmeta-reports-fourth-quarter-and-full-year-2025-results-302673127.html",{"label":548,"url":549,"year":543},"Statista, Meta Average Revenue Per User (ARPU) Worldwide (secondary; figures verified via Meta FY2025 release)","https:\u002F\u002Fwww.statista.com\u002Fstatistics\u002F234056\u002Ffacebooks-average-advertising-revenue-per-user\u002F",{"label":551,"url":552,"year":543},"Statista, Global Meta Advertising Revenue Worldwide (secondary; figures verified via Meta FY2025 release)","https:\u002F\u002Fwww.statista.com\u002Fstatistics\u002F271258\u002Ffacebooks-advertising-revenue-worldwide\u002F",{"label":554,"url":555,"year":556},"Marketing Charts, Creative Still the Biggest Driver of Sales (NCSolutions meta-analysis)","https:\u002F\u002Fwww.marketingcharts.com\u002Fadvertising-trends-230468","2024",{"label":558,"url":559,"year":543},"Jon Loomer Digital, New Ads and the Learning Phase","https:\u002F\u002Fwww.jonloomer.com\u002Fqvt\u002Fnew-ads-learning-phase\u002F",{"label":561,"url":562,"year":536},"Meta Business Help Center, Updates to Detailed Targeting","https:\u002F\u002Fwww.facebook.com\u002Fbusiness\u002Fhelp\u002F458835214668072",{"label":564,"url":565,"year":536},"Meta for Developers, Conversions API Dataset Quality and Event Match Quality","https:\u002F\u002Fdevelopers.facebook.com\u002Fdocs\u002Fmarketing-api\u002Fconversions-api\u002Fdataset-quality-api\u002F",{"label":567,"url":568,"year":536},"Supermetrics, Facebook Ads Attribution Window and Metric Removals (January 12, 2026)","https:\u002F\u002Fdocs.supermetrics.com\u002Fdocs\u002Ffacebook-ads-new-historical-limitations-attribution-window-and-metric-removals-january-12-2026",{"label":570,"url":571,"year":536},"PPC Land, Meta Restricts Attribution Windows and Data Retention in Ads Insights API","https:\u002F\u002Fppc.land\u002Fmeta-restricts-attribution-windows-and-data-retention-in-ads-insights-api\u002F",[573,577,581,584,587,590,594,598,602,606,610],{"label":574,"value":575,"source":576},"Optimization events an ad set needs within ~7 days to exit the learning phase","~50 events in 7 days","Meta Business Help Center, 2026",{"label":578,"value":579,"source":580},"Average Facebook cost per lead (Leads objective, all industries)","$27.66, up 20.94% YoY","WordStream \u002F LocaliQ, 2025",{"label":582,"value":583,"source":580},"Average Facebook conversion rate (Leads objective, all industries)","7.72%",{"label":585,"value":586,"source":580},"Average Facebook CPC (Traffic objective, all industries)","$0.70, down 6.67% YoY",{"label":588,"value":589,"source":580},"Average Facebook CTR (Leads vs Traffic objectives)","2.59% vs 1.71%",{"label":591,"value":592,"source":593},"Share of incremental CPG sales lift driven by creative quality","49%","Marketing Charts \u002F NCSolutions, 2024",{"label":595,"value":596,"source":597},"Ads in an active ad set when adding one new ad did NOT restart learning","22 ads (stayed active)","Jon Loomer Digital, 2025",{"label":599,"value":600,"source":601},"Meta worldwide advertising revenue, full year 2025","$196.18 billion, up 22% YoY","Meta FY2025 Results, 2025",{"label":603,"value":604,"source":605},"Meta average revenue per user worldwide, full year 2025","$57.03, a record high","Statista, Meta ARPU Worldwide (secondary), 2025",{"label":607,"value":608,"source":609},"Longest Meta view-through attribution window after the January 12, 2026 change","1 day (7-day and 28-day view removed)","Supermetrics \u002F Meta, 2026",{"label":611,"value":612,"source":613},"Meta event match quality score that grades how well your Pixel and CAPI signal matches","0 to 10 (higher is better)","Meta for Developers, 2026","blog\u002Ffacebook-ad-learning-phase","5thlDnVRYBc0Oytb_ZrEIB8u0rGFHwmEWmIc5S9dy5E",1786093697821]