Everyone watches the CTR to know when an ad is done. We took 271 image ads in Israeli online stores apart, week by week from their first day, to see what really wears out, when, and what that means for when to change the image.
Ad fatigue is the drop in an ad's results as the same people see it again and again: it is still running, still getting budget, and bringing in less than it did at the start. The usual advice is to watch the CTR and the frequency, and to change the image when they start to move. We wanted to know whether that is really what happens.
I lead graphic design and creative for online stores, and the same question comes back every week: is this image still working, or is it time to change it. Usually the answer comes from a feeling, or from whatever number the ads manager shows at that moment. We wanted an answer from the data.
We took every image ad on Meta in the Israeli online stores we work with, from April to October 2026, and followed each ad from the first day it spent money. The sample: 12 stores, 271 ads and 746 thousand shekels of cumulative spend, which brought 10,961 purchases. Only ads that spent at least 1,000 shekels over their life are in. Video ads are out: this article is about images. Purchases are what Meta reports on the ad, not the store's revenue, so every figure here compares an ad with itself, and none of it is ROAS. Each metric is computed per ad against its own first week, then the median is taken within the store, and then the median across the stores.
The easiest metric to see in the ads manager is the CTR, how many of the people who saw the ad clicked it. At the median it barely moves: 97% of the first week in the second week, 102% in weeks 3 to 4, and 99% in weeks 5 to 8. An image people clicked in its first week keeps getting clicked a month later.
What wears out is what happens after the click. The number of purchases the ad brings per shekel fell to 88% already in the second week, to 67% in weeks 3 to 4, and stayed below the first week after that.
The two charts together tell the story: the conversion rate from click to purchase fell to 71% of the first week in weeks 3 to 4. The same image, the same number of clicks, fewer buyers. Our explanation, from working on the ads themselves: in the first two weeks the image meets the people who were waiting for exactly this product, and Meta knows how to find them first. After that it reaches people who click out of curiosity, and people who have already seen it and bought. The click stays. The intent does not.
The ad does not stop getting clicks. It stops getting buyers.
So the CTR is not the signal to change. Of the 112 ads that ran four weeks or more, only 29% ever saw their CTR fall by more than a quarter from the first week. Whoever waits for the CTR to drop before changing an image usually never changes it.
An image ad that passed the spend threshold lived for a median of 4.5 active weeks. 63% of the ads reached their fifth week, 39% their ninth, and 26% passed three months.
That does not mean most ads "wore out" after a month. Most were closed: a sale that ended, a collection that changed, or an ad that did not make it through its first two weeks. The ads that did stay are the ones that held. In the stores we measured, ads in their ninth week or later brought 93% of the purchases per shekel of new ads in the same store. Close, but still less. An old ad that is still running is a good ad, not an ad immune to fatigue.
41% of all the money on image ads was spent in their first two weeks, and 20% at the median on ads nine weeks old or older. The budget already moves by itself to the new ads and to the old ones that hold. Our job is to make sure there is always a new ad to move to.
There are two kinds of image ads in an online store. A catalog ad takes the product images from the store's feed, and Meta chooses which product to show each person. A designed image is one ad we built: product, background, text, offer. A comparison across the 9 stores that ran both kinds in the same period:
| Metric | Catalog | Designed image |
|---|---|---|
| CTRshare of viewers who clicked | 3.5% | 1.6% |
| Click to purchasepurchases per 100 clicks | 0.5 | 2.3 |
| Cost per 1,000 impressionsshekels | 23 | 25 |
| Purchases per 1,000 shekelspurchases as Meta reports them | 11.8 | 13.7 |
| Cost per purchaseshekels | 85 | 73 |
The catalog wins on clicks in every single store, 9 out of 9: 1.7 times the CTR at the median. That makes sense, Meta shows each person the product they already looked at. But the conversion from click to purchase is half as high, and the line that matters, purchases per shekel, comes out 4% in favour of the designed image at the median, when the two kinds are compared inside the same store. In the table each column is a median of its own, which is why the gap there looks larger. The catalog brought more purchases per shekel in only 3 of the 9 stores.
What the catalog does do better is hold. In weeks 3 to 4, catalog ads kept 76% of their first week's purchases per shekel, and designed images 70%. In weeks 5 to 8, 89% against 79%. A catalog refreshes itself, because the products change. A designed image is one image, and that one has to be changed.
This is how we work on ad images after seeing these numbers.
Ad fatigue is the drop in an ad's results as the same people see it again and again. In the image ads we measured, the drop did not show in the CTR, which stayed at 102% of the first week even in weeks 3 to 4, but in purchases: 33% fewer purchases per shekel in those weeks.
In the stores we measured, purchases per shekel fell to 88% already in the ad's second week and to 67% in weeks 3 to 4. An image ad lives for a median of 4.5 active weeks, and 39% of ads reach their ninth week. An ad that is still selling is not closed because of its age.
When the ad's cost per purchase rises against its own first week, not when the CTR falls. Of 112 ads that ran four weeks or more, only 29% ever saw their CTR fall by more than a quarter. A replacement image should be ready by the second week, because that is where the drop in purchases begins.
Both, in different roles. In the 9 stores that ran both kinds, the catalog brought 1.7 times the clicks and held its results over time, but the designed image brought 4% more purchases per shekel at the median. The catalog always runs, and the designed image gets refreshed.
On the same measure and the same window: the new image's first week against the previous one's first week, in purchases per shekel as Meta reports them. An ad is measured against itself, so the comparison does not depend on budget or season. When the new image does not beat the old one, try a different direction rather than another version of the same thing.
Google's product ads have not gone anywhere, but the campaign that runs them has changed. We opened the Google Ads accounts of Israeli online stores over nine months of 2026: where the budget goes, what a click costs, and what each campaign type reports it returned.
Daniel Hirsh29 September 20268 min readEvery growth plan for an online store rests on an assumption about returning customers: how much they bring, and when. We took 36,000 orders, checked for each one whether the customer had bought from the store before, and followed the ones who came back. The answer changes where the money goes.
Daniel Hirsh21 October 20269 min readThe night sale is a very common promotion in ecommerce, and almost nobody measures what it does beyond the evening itself. We took 9 night sales apart, hour by hour, to see when the money actually comes in and what happens to the days around it.
Daniel Hirsh18 October 20267 min readEvery Preneuriat client gets a live dashboard: store revenue, all media spend, real ROAS, updated all day. The first call goes both ways.
Check the fit
