The short answer
Ad Radar reads public Meta Ad Library ads, records their engagement and how long they stay live, and turns those signals into an estimated spend with a calibrated model. Only ads past $50k in estimated spend make the board (2,443 ads from 749 brands on October 9, 2026). Top tier is the best 10% by winner score among ads live 21 days or more. Every ad is tagged on ten dimensions. The estimates rank ads well and can miss any single ad by a wide margin.
In this article8 sections
What our ad spend estimation methodology covers
This page explains the ad spend estimation methodology behind every number on the Ad Radar blog: where the ads come from, how a spend figure is put on each one, what "winning" and "top tier" mean, how ads are tagged and where the data stops being reliable. It is written by the Ad Radar Research Team and applies to every article that cites the board.
The short version: we read public ads, we record public signals about them over time, and we turn those signals into an estimate. Nothing on the board comes from an advertiser's account.
2,443
winning Meta ads on the October 9, 2026 snapshot
749
brands behind them
$50K+
estimated spend behind every ad
10
dimensions every ad is tagged on
Where the ads come from
Every ad on the board is a public ad from the Meta Ad Library, the transparency archive Meta keeps for ads on Facebook and Instagram. Ad Radar follows a set of DTC and e-commerce brands, collects their active ads, and checks them again on a schedule. Each check stores what the library shows at that moment: the creative, the copy, the start date, the platforms, the number of versions and the public engagement on the ad.
Storing those checks over time is what makes the rest possible. A single look at an ad tells you it exists. A series of looks tells you whether it is still getting delivery.
Video ads are transcribed, so the hook and the script can be searched and tagged like any other text. The landing page each ad sends to is recorded too.
How estimated spend is built
Meta publishes spend only for ads about politics, elections and social issues. For commercial ads, the library shows the ad but not the budget. Any spend figure for a commercial Meta ad, in any tool, is a model. We explain the general options in how to estimate competitor ad spend.
Ad Radar's estimate rests on two kinds of public signal:
- Engagement over time. Reactions, comments and shares accumulate as an ad is delivered, and delivery costs money. Recorded repeatedly, the growth of that engagement says how much budget is still flowing through the ad.
- Duration. How long the ad has been live. Brands don't keep paying for an ad for months unless it earns its keep.
A calibrated model turns those signals into a dollar figure. We don't publish the model or its parameters, for two reasons: it changes as we improve it, and any fixed coefficient would give a false sense of precision. The relationship between engagement and spend moves with format, niche, offer and audience.
That is why every figure is written as estimated spend, in the app and in every article. It comes with a margin of error that can be wide for a single ad. It is much more reliable as a ranking: an ad estimated at $2M has almost always spent more than one estimated at $200K, even when neither number is exact.
On the October 9, 2026 snapshot the estimates spread like this:
| Estimated spend | Ads | Share |
|---|---|---|
| $50K to $100K | 511 | 21% |
| $100K to $250K | 820 | 34% |
| $250K to $500K | 536 | 22% |
| $500K to $1M | 311 | 13% |
| $1M to $5M | 237 | 10% |
| $5M or more | 28 | 1% |
The median ad on the board is estimated at about $223K, and the top 10% start at about $1.07M.
What "winning" means
An ad is winning when its estimated spend has passed $50k. That is the only way onto the board.
The logic is simple. Testing is cheap and scaling is expensive. A brand that keeps putting money behind an ad past that line has seen it work in its own numbers, which we can't see. Sustained spend is the most honest public signal of that decision.
Two things "winning" does not mean. It does not mean profitable: we can't see margins, returns or lifetime value. And it does not mean the ad is still growing: some winners are old ads coasting on a small daily budget. The other signals, below, separate those cases.
What "top tier" means
Top tier is the best 10% of the board by Ad Radar's winner score, counting only ads live 21 days or more. On the October 9, 2026 snapshot that is 235 ads.
The winner score ranks ads on signals advertisers can't easily fake: how long the ad has been live, how many variants the brand launched around it and how fast its engagement is still growing. The four signals for finding winning ads explain each one.
When an article says a group of ads has a "top-tier rate" of 14%, it means 14% of the ads in that group reach the top tier. Because the top tier is the top 10%, about 10% is the board average. Above it, the group beats the board. Below it, it trails.
One caveat runs through the articles: variant count is one of the score's inputs, so any comparison by number of variants is partly built into the score. We flag it where it matters.
How every ad is tagged
Every ad on the board is tagged on ten dimensions. Each dimension answers one question, and each value is defined by what the ad does, not by what it says about itself. Tags are assigned from the ad's copy, its transcript and its landing page, against fixed written definitions.
| Dimension | The question it answers | Example values |
|---|---|---|
| Hook pattern | What form does the first line take? | Problem/Benefit Opener, Question Opener, Specific-Number Opener |
| Hook mechanism | Why does the opening grab attention? | Problem-to-Solution Bridge, Expert Explanation, Offer Pressure |
| Narrative angle | How does the ad build its case? | Root Cause, Systematic Breakdown, Story/Case Study |
| Reader awareness | What does the opening assume the viewer knows? | Unaware, Problem-Aware, Solution-Aware, Product-Aware, Most-Aware |
| Proof level | How much evidence backs the main claim? | Unproven, Medium, Strong |
| Production format | How was the ad made? | UGC selfie, Founder or expert, Footage + voiceover, Static image |
| Video type | What kind of video is it? | Voiceover Listicle, Demo / Mechanism, Talking head |
| Landing page type | Where does the click go? | Product page, Dedicated landing page, Listicle, Quiz, Advertorial |
| Niche | Which market or problem? | Pet, Skin, Joints, Digestion |
| CTA type | What does the ad ask for? | Buy, Lead |
The five dimensions the articles use most, in more detail:
- Hook. The opening line or the first seconds. We split it in two: the pattern is the form (a question, a number, a warning), the mechanism is the psychology (a promised fix, an expert about to explain, a deal that won't last). See hook pattern vs hook mechanism.
- Angle. The line of argument after the hook. There are eight; an ad can carry up to two, and about a quarter of ads carry none, usually short offer-led ads. See marketing angles.
- Format. The production technique, not the content. A founder reading a script and a founder in an interview are both "Founder or expert".
- Awareness. Eugene Schwartz's five levels, judged by what the opening of the ad assumes, not by where the ad ends up. About 12% of ads are too short to classify. See awareness levels in ads.
- Landing page. The type of page the click lands on, recorded with a snapshot of the page. See message match.
Hook pattern and hook mechanism take one value per ad. Angle, proof level and niche can take two, because real ads mix them: 36% of tagged ads carry two narrative angles. That is why shares in a table can add up to more than 100%. The taxonomy guide covers the design rules in full.
Limits of the data
Use the numbers with these limits in mind:
- Meta only. The board covers Facebook and Instagram. TikTok, YouTube and Google ads are not on it.
- Active ads only. Outside the EU, the Meta Ad Library drops ads once they stop, so the board reflects what is live, not every ad that ever ran.
- A tracked set of brands. The board follows DTC and e-commerce brands, mostly selling in the US. It is not a census of all Meta advertisers, and categories like B2B, finance and local services are thin or absent.
- Estimates, not spend. Shared posts can give several ads the same estimate. Formats and niches that attract comments look bigger per dollar. Engagement bait inflates the count. Compare within a niche and a format.
- Survivors only. Every ad on the board already won. The data says what winners have in common, not how often a tactic fails, because the failures never reach the board.
- Association, not cause. When a tag reaches the top tier more often, the tag and the result travel together. It doesn't prove the tag caused the result.
- Small groups. Where a group has only a few dozen ads, the article says so. Treat those rates as directions, not findings.
- Snapshots move. Each article names its snapshot date. The live board changes every day, so the counts in the app will differ from the counts in an article.
How to read the numbers in our articles
A few conventions repeat across the blog:
- "Estimated spend" is always the model above. Medians are used instead of averages, because a handful of very large ads would pull an average far from the typical ad.
- "Top-tier rate" is the share of a group that reaches the top tier. Compare it with the 10% baseline.
- "Days live" counts from the start date shown in the Meta Ad Library.
- Examples are real ads from the board, quoted as they ran, with their estimated spend and days live on the snapshot date.
If you find a number that looks wrong, write to hello@ad-radar.dev and we will check it against the snapshot and correct the article.
Figures marked as estimated spend come from Ad Radar's model of engagement on public Meta Ad Library ads. They are estimates, labeled as such, and are best used to rank ads against each other.