October 2026. Data collected 28–29 July 2026.

The first page of the Ad Library is not the market

What 3,282 cash-on-delivery ads shown in Morocco say about competitor research on Meta, and why three of the usual shortcuts break.

Most competitor research on Meta starts the same way. Search a keyword in the Ad Library, scroll the first results, pick the ads that have run the longest, and copy what they do. I wanted to know how well that holds up in a market I know: cash-on-delivery e-commerce in Morocco, where paying in cash at the door is still the default for many online orders.

So I collected 3,282 ads with my own Ad Library collector and checked each shortcut against the full data instead of the first screen.

What I collected

• Meta Ad Library, ads shown in Morocco, all statuses and media types, collected 28–29 July 2026.
• Six searches in the languages these ads are written in: الدفع عند الاستلام, الدفع عند التوصيل, خلص ملي يوصلك (Darija), “paiement à la livraison”, “livraison gratuite” and “commandez maintenant”.
• 3,282 unique ads, deduplicated by ad ID, from 1,426 Facebook pages. 2,223 were still running on the collection date; 1,059 had stopped.

The limits matter, and they are listed at the end.

1. The first results oversample the biggest advertisers

The Ad Library does not return ads in random order. In every search, the ten pages with the most ads take up a much larger share of the first 100 results than of everything I collected. The deeper you scroll, the more fragmented the market turns out to be.

Across the whole dataset, 905 of the 1,426 pages (63%) ran exactly one ad, and the ten biggest pages made only 12.6% of all ads (about 14–15% if pages from the same brand are merged). If you research from the first screen, you are studying the heaviest spenders, not the market. And “copy the market leader” assumes there is one. In Moroccan cash-on-delivery, there mostly isn’t.

2. “It has run for months, so it works”: check what you can’t see

The most repeated advice about the Ad Library is that long-running ads are winners, because nobody keeps paying for an ad that loses money. Two things in this data complicate it.

First, two thirds of the ads were still running. Among the ads that had stopped, the median ran 21 days, 41% ran past 30 days and 15% past 90. But stopped ads are a biased group: the longest runners are mostly still live, so any “average lifespan” computed from stopped ads comes out too short, and any long-versus-short comparison mixes finished tests with ads that simply started last week. Of the ads still running, 30% had already been live for more than 90 days.

Second, run time measures an advertiser’s patience and budget, not results. Meta shows no spend, reach or sales for these ads. Page size barely moves it: stopped ads from pages with ten or more ads ran a median of 22 days, against 17 for pages with a single ad (42% versus 37% past 30 days).

This matters more now than it used to. Since Meta’s Andromeda changes to ad delivery, the standard advice from media buyers is to launch many genuinely different creatives and let delivery sort them. In that setup a short run is often a planned test rather than a failure, and a long run can just as easily be an untouched catalog ad. Use run time to choose which ads to study, not to decide which ones work.

3. Small samples invent trends

To see how much a quick scroll can mislead, I drew 500 random samples of different sizes from the dataset and measured the top-10 share in each.

Formats are worse, because splitting by format shrinks every group. In an early cut of this data, dynamic creative ads seemed to outlast image ads by 23 points, on just 55 ads. With 182 of them, they did worse than video: 34% ran past 30 days against 44%. In random samples of 150 stopped ads, the gap between the two swings from 27 points in video’s favor to 7 in dynamic creative’s. A 100-ad scroll can “prove” almost anything about formats.

The practical rule: before comparing two groups, make sure each has a few hundred ads, and run the comparison again as the sample grows. If the answer flips, you don’t have one yet.

4. For many sellers, WhatsApp is the checkout

30% of the ads (992, from 522 pages) send people to WhatsApp, through a wa.me link or a “Send WhatsApp message” button, not to a product page. For a cash-on-delivery seller the chat is where the order is taken and confirmed. The share depends heavily on the search: 46% of ads found with الدفع عند الاستلام used WhatsApp, against 14–19% for the French phrases. That compares ads within each search, not languages across the market. If your research only follows ads to landing pages, you miss a large part of how this market sells.

How I’d research competitors in this market

1. Collect past the first page: several hundred ads per search before describing a market.
2. Keep running and stopped ads apart, and never average a lifespan across both.
3. Use run time to pick ads to study, not to declare winners.
4. Follow the conversion path: WhatsApp, Messenger, website or marketplace.
5. Write down the denominator: which searches, which dates, and what got cut short.

Limits

• These are ads shown in Morocco, not ads from Moroccan advertisers. The sample includes an Algerian shop, French retailers and global brands.
• After about 2,300 ads in one day the collection hit an IP-level block, so some searches stopped early and the dataset leans toward the Arabic search that ran deepest. For that reason I don’t report language or offer shares: the searches chose them.
• The Ad Library has no spend, reach or sales data for these ads. Nothing here measures performance.
• Advertisers are counted by Facebook page. Merging pages from the same brand moves the top-10 share from 12.6% to about 14–15%; none of the conclusions change.