Your best ad just died. Now what?

Four things that look like fatigue and are not

The morning your best ad tanks is the most panicked moment in a small account, and the first instinct is to blame the creative. Half the time it is fine. Check these first, in order; each takes five minutes and each has fooled a good media buyer.

1. A tracking break. A pixel or Conversions API change, a Shopify app update, a tweaked consent banner. Compare Meta’s reported purchases against Shopify’s orders for the same day. If Meta sees half of what Shopify sees, the ad did not change.
2. A landing page change. Someone edited the product page, the price moved, the free shipping threshold changed, a bestselling variant went out of stock. Click your own ad on your phone.
3. A budget jump that reset learning. A significant edit resets the learning phase; Meta wants roughly 50 optimisation events in a week to exit it[2], and performance is unstable until then. Check what changed in the account the day before the drop.
4. The calendar. Payday cycles, a competitor’s promotion, the CPM spike that comes with a shopping season. If every ad in the account got worse on the same day, it is the market, not the ad.

What Meta itself calls fatigue

Meta puts two labels in the Delivery column[1]. Creative limited means cost per result has risen above the ad’s own history. Creative fatigue means it has at least doubled. Both are lagging signals: the cost damage is already paid by the time the label appears.

Use the labels to confirm fatigue after the fact, and the frequency band below to see it coming.

The frequency band to watch

Frequency is how many times, on average, one person has seen the ad. On cold prospecting, a rising seven-day frequency means the algorithm is running out of new people and repeating itself. Practitioner benchmarks put the early-warning band at 2.5 to 3.0, and call 3.5 late[3]. Your own account’s history beats these thresholds once you have a few winners to compare.

Pair frequency with two other lines. Falling click-through with rising CPM is the fatigue signature. Flat click-through with falling conversion is a landing page problem in creative’s clothing.

How long a winner should live

Practitioner data puts a typical Meta creative at two to three weeks before fatigue shows, TikTok at one to two[4]. A winner that ran six weeks did its job. But the asset dies in weeks while the concept can live for a year.

When we read haircare in the Meta Ad Library, Polar Haircare’s gray-coverage hook had been live since August 2025, relaunched with new footage on five separate dates. Based Supplies did the same with its onion script, four times over. The ad you are mourning is one cut of an idea that may have a long way to go.

The headwind you are fighting anyway

Some of what feels like fatigue is the market getting pricier underneath you. Triple Whale’s 2025 panel saw Meta CPMs rise 20% year over year while cost per purchase rose only 1%[5]. Reach got a fifth more expensive; the brands that held cost per purchase flat did it with creative. The ones that did not keep shipping paid the full 20%.

A bid change rarely fixes a tired winner. New creative does.

Iterate or go net-new: the rule

First, decide whether the ad was actually a winner. Motion’s bar: at least ten times the account’s median creative spend, and at least $500[6]. If it never got there, do not iterate it. It was a guess that ran a while; the next guess should come from research, not from this ad.

If it was a winner, work the ladder in this order, and do not skip rungs:

1. New first three seconds, same body. Three openings, one launch. The hook is the part of an ad that fatigues fastest and costs least to replace.
2. New presenter, same script. If the openings moved the numbers, the concept is alive and the face is tired. Different person, different room, same words.
3. New format, same angle. The advertorial becomes a twelve-second demonstration, or the other way round. Same promise, different shape.
4. Net-new angle. Only after two rungs in a row fail. Pick the next row on your angle map, which is why the map has to exist before the panic, not after.

The iteration ladder on a real concept

Rung one, on one of our own ads. The shampoo ad on our work page opens with a size line; the body (open, pour, work it in, set it down) cost the most to get right, so it stays. These are the four openings we would cut against it, in test order. Each is something a synthetic presenter may say: a demonstration, a question or an enemy, never a testimony.

Notice what is not on the ladder: a before-and-after, a “six weeks later”, a customer quote. A presenter who has not used the product cannot claim a result from it, and we would rather lose a rung than cross that line.

The pipeline that prevents the next panic

Every step above moves faster if the next concept already exists. Write the replacement only after the winner dies and you are always two weeks behind. Brands that never panic ship a weekly number of new creatives regardless, so a scored concept is always waiting. How many, by spend, is in how many ads should you test a month.

If you cannot produce that pipeline with your current team, that is the gap we exist to close. See whether our output is good enough: have us make one ad for your product, free.