Three competitors’ Meta ads, read in full

Competitor Ad Audit Report

Their ad count is not their creative count, and their longest-running ad is usually not their best one. We download their ads and review them with computer vision. That's how we detect re-uploads and read the hooks.

How it works Get the report — €149

You name three libraries. If they are too thin to audit, we swap them or refund you — before any analysis begins. No call, no subscription.

01 Who it is for

You are planning the next creative cycle.

This is for you if:

Four conditions

  1. You are deciding what to test next, and two or more competitors are visibly busy on Meta.

  2. You do not have hours to scroll, sort and compare hundreds of ads.

  3. You want claims tied to evidence, not a generic AI summary.

  4. You spend enough that avoiding one weak test pays for this.

02 Full library, not a sample

You saw 20 ads — they have run 423.

Fifteen minutes of scrolling gets you the top two rows.

A dense grid of 423 thumbnails, one per ad, from a single cosmetics advertiser's public Meta Ad Library — the whole capture, thirty-nine tiles to a row, dissolving along the bottom edge where the final row stops short. The grid is dominated by pinks and product-on-colour studio shots, with visible clusters: the same percentage-off offer creative repeating across many tiles, a band of talking-head creator videos, and runs of near-identical variants of one palette launch.

423

ads in one competitor’s library

263 already stopped

The patterns are in the whole library. The same creative running a dozen times. The split between talking-head UGC and studio flat-lay. The run of near-identical variants that means a dynamic-creative test. We gather all of the data to see the patterns. Every number says what it was counted from, so you can reconfirm.

03 Creative survival vs output

The ad that has run longest is usually not the ad delivering most per day

In a third library, the nine longest-running ads had each reached under 200 accounts — against a library maximum of 28,790, on an ad that had been live three days.

The obvious way to rank a competitor’s library is by how long each ad has run. Survival looks like proof. It is also the only ranking Meta gives out without our recalculation.

Divide reach by days live and the library re-sorts.

Rank a library by how long each ad has run and you put its quietest ads at the top.

Run span and EU reach per day are different observable signals. Neither proves performance. The report reads both instead of collapsing them into a one-column “winner” rank.

The demonstration

Now sort it the way Meta will not let you

Library explorer Run duration ↓ · max 701d · 64 of 64 rows
Sort
Show

Longest-running first. Every ad in the slice carries a run-span, so this sort ranks all 64. Bars run to the column maximum, 701 days.

Creative Ad Runmax 701d Variantsmax 45 EU reachmax 99,526 Per daymax 16,588 Status
🔒All Your Marketing Analytics for a 1-Time FeeIMAGE Run701d Var1 Reach21,431 /day30.6 Stopped
Elevate Your Ad Game 🚀IMAGE Run298d Var1 Reach185 /day0.6 Stopped
🧠 The SuperApp for Meta Advertisers (7-Days Free)VIDEO Run290d Var1 Reach1,202 /day4.1 Stopped
AI Will Eat Your Ad Homework! Try for 7-Day Free 🐶🤖IMAGE Run288d Var5 Reach601 /day2.1 Stopped
Meet the first MEDIA-BUYING AI 👋🤖VIDEO Run277d Var2 Reach668 /day2.4 Stopped
🌠 Try 7-Day Free: EXPERT AI for Meta AdsIMAGE Run266d Var1 Reach3,100 /day11.7 Stopped
🦅 Upskill & Become The Sheikh of Your Meta AdsVIDEO Run210d Var6 Reach418 /day2.0 Stopped
Running Ads without AI? 🙈VIDEO Run141d Var4 Reach1,968 /day14.0 Stopped
🦅 Ready to Deploy AI for Your Meta Ads?IMAGE Run114d Var20 Reach44,590 /day391.1 Stopped
BUILT BY MARKETERS, BOOSTED BY AIVIDEO Run114d Var6 Reach8,568 /day75.2 Stopped
CHAOS OUT, AI IN CONTROLIMAGE Run113d Var17 Reach4,751 /day42.0 Stopped
SURGICAL TRACKING PRECISION. REVEAL HIDDEN REVENUE.IMAGE Run112d Var5 Reach2,738 /day24.4 Stopped

madgicx.com · Meta Ad Library · 64 of 500 captured ads · country=ALL, 2026-08-12

04 How it works

Four steps. Including human review.

Where the five days go 3 machine · 1 person

  1. 01 Machine

    Order

    Fill in your details and pay €149.

  2. 02 Machine

    Capture

    We capture and compare the public evidence.

  3. 03 A person

    Specialist check

    A paid-media specialist checks the recommendations.

  4. 04 Machine

    Delivery

    Your report arrives within five business days.

05 The deliverable

What you get

Around 20 pages, private to you, built to inform a decision.

Executive verdict

The situation, the catch and what to test.

One page

Three competitor snapshots

Product and offer, funnel, creative production, testing velocity, dynamic-creative use.

One page each

6–8 tests, ranked

Every claim shown beside the ads behind it, with run-span, daily reach and the on-screen text each carried.

Each one cited

Cross-competitor pattern matrix

What shows up in one library, two or all three.

8 patterns × 3 libraries

Retired creatives

What each advertiser stopped paying for, how long it ran and what that does and does not prove.

Per advertiser

Evidence and limits

Links to every ad behind a major claim, the method and where each proxy misleads.

Every major claim

The artifact itself

See the example pages from a report

Extracted pages from an audit we ran on three public libraries as a demonstration, redacted. Nothing from a buyer’s report is ever published.

Four pages · any one opens full size

The audit's first page: an executive verdict laid out as situation, complication, question and answer, followed by four key findings, each with a figure, a confidence label and a caveat.
Fig. 06.1 — Executive verdict · the one-page owner brief · n=423 captured ads
A single recommendation shown beside the three ads that motivated it, each ad carrying its run-span, its EU reach and its on-screen headline.
Fig. 06.2 — A recommendation, with its evidence
A ranked table of a competitor's library: every ad as a row, with run-span, EU reach, quadrant, status, destination host and the angle it carried.
Fig. 06.3 — The full ranking · every ad a row · quadrant is the total-reach grid, not the per-day re-sort
Every ad in the library placed into four quadrants by how long it ran against how far it reached, with the ads carrying no EU reach counted separately rather than treated as zero.
Fig. 06.4 — How the capture divides · 272 of 311 ads carry EU reach

06 Before you decide

What you would do instead.

5 Alternatives, including doing nothing

  1. Free · unreadable at scale

    Browse the Ad Library yourself

    Free, and fine for looking up one ad. It cannot be read as a body of work: no sort by run length, no grouping of variants, no view of what was retired. Six hours of scrolling gets you four saved ads. You are not short of access. You are short of a way to read three hundred ads at once.

  2. Real · undeclared slice

    Buy a $25 automated report

    Real, fast, and good value for one competitor. If you have never seen one, buy one before you buy this. Then ask it two questions: how much of the library did you read, and did you include the ads that have stopped?

    We captured a brand one of these had covered. Their published median run was 8 days. Ours was 21, same advertiser, 25 days apart.

    Both numbers are real. Only one of them arrives with the count of ads it was drawn from.

  3. No files · no denominator

    Ask ChatGPT

    Handing a chatbot the library link returns nothing at all. Meta blocks them before a card renders. A better model does not fix it.

    If you have a browser extension that drives your own logged-in tab, use it. It is free, and it does read the library — the ad list and the EU reach behind each details panel. That is the real alternative to this page. It is worth an hour.

    Two things it does not do.

    It never takes the creative off Meta’s servers. So it cannot check its own account of what is in a video, and it cannot tell you which ads are the same file.

    And Meta throttles it partway through. The second pass is a different sample from the first, so what it hands back has no denominator either.

  4. The place, not the work

    Subscribe to an ad-spy tool

    €50–150 a month buys search, filters, saved boards and a large archive. The tools are good at that. This costs about one month of that subscription, once.

    What you buy is a better place to do the work, not the work. You still open the libraries. You still decide what counts as a pattern. You still write the brief.

    If you will do that every week, a subscription is the right purchase, and this is not. This is for the cycle where you need the answer once, in writing, with the evidence attached.

  5. Free today · expensive later

    Skip it this cycle

    The real default, and worth naming. The account keeps repeating what it already knows, the category moves on, and you find out when it is expensive to fix.

07 The order

Founding offer

€149

One-time

Paid upfront, by card. No discovery call. No invoice chase. You buy it the way you buy software.

The receipt and the finished report both go here.

Next screen is Stripe's card page — your card details never touch this site. You name your three competitors straight after, and nothing is analysed until they are checked. The report reads EU delivery data, so the three libraries had better be advertising in the EU.

Adracle is a trading name of Volodymyr Holovachov, a sole trader registered in the United Kingdom, who is the seller and merchant of record for this order; analysis is delivered by 99rabbits under contract. Prefer an invoice first, or a bank transfer? Write to stan@99rabbits.com and it is arranged by hand — the slot is held either way.

What we aren’t doing

What this report cannot tell you.

Meta’s Ad Library does not show a competitor’s ROAS, their conversions or who they targeted — so neither do we. What it does show is what they ran, how long each ad stayed up, how often the same file came back, and where it pointed. Where we are inferring rather than reading, we say so on the line. An ad running everywhere is proof they kept it, not proof it worked.

The report reads EU delivery data — the only place per-ad reach is published and retired ads stay archived. For a market outside the EU we can read what the libraries show, but the daily-reach ranking has no vertical axis there, and we would tell you that before taking your money.

Questions

7 · answered here

How much of each library do you actually read?

We capture against Meta’s own count, and the report prints it.

Meta publishes a result count for every library search. We capture against that count and state it as N of M on the first page. If a capture comes in short, the report says so rather than calling a slice the library.

It is the one question worth asking any competitor report you buy. Including this one.

Can you tell which competitor ads are profitable?

No. Public libraries do not expose ROAS or conversions. The audit identifies visible persistence, recurrence, change, messaging, offers, and testable implications.

Will you copy competitor ads for us?

No. The deliverable translates observed patterns and gaps into original test directions.

How do I choose which three?

Pick the brands competing for the same buyer with a similar mechanism — not necessarily the largest names in the category. A competitor one size up from you is usually more instructive than the market leader, whose budget and constraints are nothing like yours.

If one of them turns out too thin to audit, we swap it or refund you before any analysis begins.

I would be forwarding this to my founder. Will it hold up?

That is what the confidence labelling is for. Every claim links to the ad behind it, and the report never presents prevalence as proof of performance. When someone asks how you know a direction will work, the honest answer — here is what we observe, here is what we are inferring, here is what the test would tell us — is already written into the document.

How do I know this isn’t an AI summary with a price tag on it?

Read the sample pages above first. They are redacted pages from a real audit, run on three public libraries as a demonstration.

A paid-media specialist checks the final recommendations for unsupported leaps.

And the report keeps four things visually distinct throughout: machine synthesis, observed evidence, analyst inference and reviewer comment. You can see which is which rather than take our word for it.

Is there a call?

No. Pay, name your three competitors and the decision you are facing, and the report arrives within five business days.

If you would rather talk first, email stan@99rabbits.com. Nothing about buying requires it.