A competitor's ad library shows ads, not results

A competitor's ad library tells you what they ran, not what worked. For ordinary commercial ads, Meta publishes the ad and its delivery dates, plus some reach and targeting details for ads shown in the EU and UK. It does not publish spend or results. Every teardown that reads winners out of it is inferring performance from survival.

Funnel teardowns are doing well on ecommerce YouTube this week, including one that screen-shares a mushroom coffee brand's active ads and walks through what each type is doing, from UGC to comparison statics (the video). As a catalog of ideas, that is useful. As evidence of what performs, the library cannot carry the weight.

What the library actually shows

Meta's documentation for the Ad Library API lists each field and which ads it is available for (Meta for Developers):

Reach tells you an ad was delivered to people. It does not tell you anyone bought.

What the EU reach numbers can and cannot tell you

If your competitor advertises in the EU, the library adds an estimated EU reach and a breakdown of the people reached by age, country and gender. That is the closest thing to a budget signal it offers: an ad seen by many people was delivered widely, and wide delivery costs money. It is still not a result. An ad can reach a lot of people because it is being tested hard, because it is cheap to deliver, or because it works. The reach number cannot tell those apart, and the one report that can, the brand's own ads account, is not public.

Three ways a teardown misreads it

Each of these turns a list of ads into a list of supposed winners:

1. Long-running is read as winning. An ad that has run for months may be a winner. It may also be a small evergreen ad nobody switched off. The library shows both the same way. That is survivorship bias with a start date.
2. Many active ads is read as success. A large count of live ads shows a brand that ships a lot of creative. It says nothing about which ads carry the budget or the sales.
3. Today's ads stand in for the whole history. A teardown reads what is live now. The ads that were tried and dropped carry half the lesson, and they rarely make the video. The sample tilts toward whatever lasted, which is selection bias.

The same trap catches product research. Copying bestsellers is survivorship bias for the same reason: you see the survivors and not the graveyard.

Who the teardown is for

A teardown has to end in a lesson, and "we cannot know from here" is not one. That is the format, not dishonesty. The library itself was built for transparency about who advertises and to whom, not as a benchmark of what performs. Meta sees the results and does not publish them. The brand sees them and has no reason to share. You see ads.

What you can take from it

Used as a list of hypotheses, the library is worth an hour a month:

• Angles and hooks: which claims competitors lead with, and how that changes over time.
• Formats: UGC, founder video, comparison statics, demos.
• Offers and landing pages: click through. The landing page shows the offer structure, which the ad often hides.
• Timing: start dates show when a competitor launched a push, which helps you read your own numbers for the same weeks.

Everything on that list is an idea to test, not a finding. When a whole category adopts the same angle at once, the idea also stops being different, which is what happens when everyone clones the same offer block.

How to test a borrowed angle

• Borrow one element at a time: one hook, one offer, one format. A test that changes five things teaches you nothing reusable; see vary one element, learn something reusable.
• Run it as a split with non-overlapping audiences. Meta's split testing divides the audience so the groups do not overlap (Meta for Developers), which is what makes the comparison a split test rather than two campaigns competing for the same people.
• Judge it on your own store's revenue and margin over the test window, not only on the platform's attributed purchases.
• Keep a record of what you borrowed and whether it worked. Your own library of results, with dates, is worth more than theirs; see own the test results, not just the winners.

A borrowed angle that wins in your account is yours now, with evidence behind it. One that loses has cost you a test budget instead of a quarter.

Where a read fits

Your own GA4 export already holds what their library never will: which of your channels drove your sales. Later, when you want that read, a causal attribution tool like Causality Engine takes one file, your GA4 Attribution paths export, and shows what each channel caused next to what last-click gave it, in 1 to 2 minutes. The read is EUR 99 once per upload, excluding VAT.

Related answers

• Copying bestsellers is survivorship bias
• When everyone clones the same offer block
• A bestseller is not evidence of why
• Find the mechanism, not the winning product
• Own the test results, not just the winners