Teardowns
Other people's advertising, taken apart
Every ad any company runs is public. We read them the way we read a client account, write down what we find, and publish it – including when the answer is that they are doing it well.
Why we do this instead of publishing case studies
We cannot show you our own client numbers. Every account is under a confidentiality agreement signed without a publication clause, and inventing impressive percentages is what we would be criticising if someone else did it.
So we work in public on companies who are not our clients. It demonstrates the same thing a case study would – how we think, what we notice, what we would change – on material anyone can go and verify.
Where the material comes from
Both major platforms are legally required to publish what advertisers run. Google's Ads Transparency Center shows every ad an advertiser is running, by format and region. Meta's Ad Library shows the same for Facebook and Instagram, with dates. Between them you can reconstruct a competitor's creative strategy without any special access at all.
What you cannot see is spend, performance, or targeting. So a teardown is a reading of strategy and craft, not of results. We say so in every one, because the alternative – asserting what worked based on what is visible – is the most common failure of this genre.
What we look at, in order
| # | What we examine | What it reveals |
|---|---|---|
| 1 | Creative volume and refresh rate | Whether there is a real production pipeline or a one-off batch |
| 2 | Concept variety versus variant count | Fifteen cuts of one idea is not a testing programme |
| 3 | Message-to-landing-page match | The most common and most expensive disconnect |
| 4 | Offer construction | Trial length, gating, price framing – where visible |
| 5 | Format mix by placement | Whether creative is adapted per surface or just resized |
| 6 | Longevity of individual ads | Long-running ads are the closest public proxy for what works |
| 7 | What is missing | Absent formats and untouched surfaces are usually the real finding |
The rules we hold ourselves to
We never claim to know performance. We do not pretend a competitor is failing to make a point. We name what we would do differently and why, so the reasoning is checkable. And if a company we take apart wants a right of reply, they get one on the same page.
-
Cross-section · Google
Seven advertisers, one table
Every account below, re-counted on the same day so the figures are comparable. The volume spans forty times, and the tidy conclusion – B2B buys text, DTC buys video – does not survive the whole table: two B2B companies here differ by that same factor of forty.
September 2026
-
Google · subscription apps
Duolingo
Sixty-six search text ads in the United States, fewer than a running app a fraction of its size. Half of them are not for the language app at all, but for a $70 exam on a different subdomain – with the competitor price written into the copy.
September 2026
-
Google · B2B software
monday.com
Eight thousand live US ads and nearly two thirds of them plain search text – the inverse of the DTC account below. One platform sold as four separate products, two claims of being number one, and no price anywhere.
September 2026
-
Google + Meta · ecommerce & DTC
Gymshark
Roughly five thousand live US ads, of which only eight percent are search text. Every text ad defends the brand term, the price appears on Meta and never on Google, and one segment gets a page of its own.
September 2026
-
Cross-section · Google + Meta
Eight advertisers, one decision
Whether the offer belongs in the creative at all. Live copy from eight advertisers across apps, payments, hardware and team software, and the two product facts that decide the answer.
September 2026
-
Google · fintech
Wise
Roughly two thousand live US ads answering one objection, Chinese-language creative targeted inside the United States, and two licensed entities running separate ad accounts.
September 2026
-
Google · B2B software
Notion
No trial, no urgency, no offer. Instead one description doing work on two different people – the person who wants the tool and the person who has to approve it.
September 2026
-
Google + Meta · subscription app
Headspace
Around 400 search ads and a Meta split by symptom rather than by product, a promise with a number and a deadline, and a separate campaign line for a new AI product.
September 2026
-
Google + Meta · subscription app
Runna
Around 200 search ads segmented by race distance, trial terms stated in the creative rather than saved for the paywall, and three Meta formats running at once – including the parent brand doing credibility work for free.
September 2026
How to run one yourself
Nothing here requires us. Open the Ads Transparency Center, search the advertiser, and filter by region and format. Then open the Meta Ad Library and search the same brand. Export what you find, sort by how long each ad has been running, and read the ones that have survived longest – those are the ones someone decided to keep paying for.
Do it for your three closest competitors and you will learn more about your category in an afternoon than any benchmark report will tell you.
Or have us take apart your own account
The free audit is a teardown of your advertising, except we can see the performance data as well as the creative.