To analyze a competitor’s ads, sort their public ad library by how long each ad has been running, then break the longest runners into hook, angle, format and destination. Run length is the only performance signal the libraries publish, and it is a good one: nobody keeps paying to serve an ad that loses. What transfers to your own account is never their wording — it is the pattern underneath it.
Last reviewed September 7, 2026. Every ad quoted below is a real creative recorded in the Sprites ads library, with its run length as measured on that date.
Step 1: Open the right library
Three libraries are public, they disclose different things, and knowing which is which saves an hour of confusion:
| Library | What it shows | What it withholds |
|---|---|---|
| Meta Ad Library | Every active ad for a Page, with the creative, the start date and the platforms it runs on. | Spend and impressions, except for ads about social issues, elections or politics, and EU-targeted ads. |
| Google Ads Transparency Center | Search, Display and YouTube creatives by advertiser or domain, with first- and last-shown dates and a rough impression band in some regions. | Keywords, budgets and which campaign an ad belongs to. |
| LinkedIn Ad Library | Ads by company, with the ad type and, for EU-served ads, a date range and an impression band. | Dates and impressions for everything outside the EU, which is most of it. |
Detailed walkthroughs of each live in how to see competitors’ Facebook ads and the Google Ads Transparency Center guide. If you would rather not open three tabs, the Sprites ads library merges all three onto one page per advertiser.
Step 2: Sort by how long each ad has been running
Ignore the newest ads. A creative launched last week is an experiment; a creative still serving after four months has survived a budget review. Sort descending on run length and take the top five.
Calibrate against the category rather than your instinct. Across the 84 advertisers in the ads library on September 7, 2026, the median advertiser’s longest-running Meta ad had been live 92 days, and only 15 of the 59 advertisers with dated Meta ads had anything past 180 days. Three months is a proven creative here. Six months is unusual enough to be worth an hour of your attention.
Two traps. Localised duplicates inflate the list — Captions has effectively the same video running past 400 days in English, Spanish, French and German, which is one creative and four rows. And a long run on a brand-awareness ad tells you far less than a long run on an ad pointing at a signup page, so check the destination before you get excited.
Step 3: Break the long runners into elements
Four columns, five rows, ten minutes. For each surviving ad, write down:
- Hook — the opening line, quoted exactly. Not a paraphrase: the wording is the evidence.
- Angle — the argument in five words. “Stop losing time to X”. “People like you already use this”.
- Format — video, static or carousel, plus the length and whether a face appears in the first second.
- Destination — the exact path. A homepage and a comparison page imply completely different funnels.
Worked example, from Grammarly’s ads: the hook is “Be a Better Writer, Starting Today”, the angle is aspiration with the price objection removed (the body copy leads on “free writing app”), the format is a single static on Facebook only, and the destination is grammarly.com/a. That ad ran 328 days. The transferable lesson is not the sentence — it is that a static, aspirational, price-neutralising opener outlasted every video the same account was running.
Step 4: Categorize the hooks into styles
Now count. Sort the hooks you collected into a small taxonomy and tally them; the tally is the finding, not the individual ads. Seven categories cover almost everything, and here is one real long-running ad for each, drawn from the ads library:
The full taxonomy, with the tells that put an ad in one bucket rather than another, is in ad hook styles. What you are looking for in the tally is not the winner. It is the empty column: the style your competitor leans on that your account has never run, or the style nobody in the category uses at all.
Step 5: Generate variations against your own promise
This is the step where most competitive research dies, because “write six ads” is real work. The rule that makes it tractable: take the style, not the sentence. A hook style is a structure — a question, a number, a named pain, a before-and-after — and structures are not ownable. The specific promise has to be yours, because it has to be true.
A batch of six that keeps the test readable:
- Two ads in the style your competitor’s longest runner uses, applied to your strongest proof point.
- Two ads in a style the whole category ignores. This is where the outsized wins come from, and where most of the losses do too.
- One ad that changes only the format — same hook, same angle, static instead of video — so you learn something about format independently.
- One wildcard nobody in the room quite believes in.
Write the expected result down before launch. A test you cannot be wrong about teaches nothing, and “we learned a lot” after the fact is what people say when they wrote no hypothesis.
Sprites’ ad generator does this step from the analysis: hand it the angle and it returns headlines, image concepts and video scripts for your product, with the research each one came from attached.
Step 6: Launch the batch and read the cohort
Launch the six together, tag each one with its hook style and angle at launch, and judge the group. A single ad’s cost per result at a few hundred dollars of spend is a coin flip; six ads sharing one style, read together, is a signal. When the cohort’s cost per result drifts up while spend holds steady, the creative is fatiguing and it is time to run the loop again.
Keep the tags. The second time you do this the analysis takes twenty minutes instead of two hours, because your own account has become the second library you read.
What not to copy
- Their footage, images, music or script. Copyright applies to ads exactly as it does to anything else.
- Their brand name as a keyword or in copy. Several platforms’ policies prohibit it, and it invites a trademark complaint.
- Their claims. Their “#1 rated” may be substantiated. Yours has to be too, separately.
- Their offer, without their margins. A competitor’s free tier or discount may be losing them money on purpose.
The structure is the part you are allowed to learn from, and it happens to be the part that transfers. A question hook works because of what a question does to attention, not because of whose question it was.
Running the loop without doing it by hand
Every step above is manual work that has to happen weekly to be worth anything, which is why it usually stops after the first enthusiastic month. Sprites’ creative analytics runs it continuously: it reads the public libraries and your own accounts, sorts by what keeps getting funded, breaks the winners into elements, writes the next batch, and launches it — with every change waiting for your approval. Start free in the ads library, or book a demo.
Frequently asked questions
How do I analyze a competitor's winning ad?
Find their ads in the relevant public ad library, sort by how long each has been running, and take the longest runners as the winners. Then break each one into four elements: the hook (the opening line or first three seconds), the angle (the argument it makes), the format, and the destination it sends the click to. The pattern across those four columns is what transfers to your own ads.
Is there a tool that finds winning ads and recreates similar creatives?
Yes. Swipe-file tools such as Foreplay and Motion’s Inspo find and organise them; Sprites goes a step further and generates new creative for your product from the pattern behind the winner, then launches the test in your account once you approve it. What no tool should do is reproduce a competitor’s creative — their ad was tuned to their audience, their price and their brand.
How do AI tools generate new ad hooks from winning patterns?
By separating the pattern from the wording. A winning hook is a style — a question, a number, a named pain point, a before-and-after — applied to one specific promise. An agent that has read a few hundred ads in your category can see which styles keep getting funded, then apply the styles you are not using to the promise you can make. The generated hook is a hypothesis until the test comes back.
How long does an ad have to run before it counts as a winner?
It depends on the category, which is why the comparison should be within it. Across the 84 advertisers in the Sprites ads library on September 7, 2026, the median advertiser’s longest-running Meta ad had been live 92 days, and only 15 of the 59 with dated ads had anything past 180 days. Against that distribution, three months is a proven creative and six months is an outlier worth studying.
Can I get in trouble for copying a competitor’s ad?
Reproducing their footage, images, script or trade dress is a copyright and trademark problem, and running their brand name as an ad keyword breaks several platforms’ policies. Nothing stops you learning from the structure: hook styles, angles and formats are not ownable, and they are the part that actually transfers.
What if my competitor runs almost no ads?
Widen the frame. Analyse the advertisers your buyer sees rather than the companies you sell against — adjacent categories that target the same person are usually a richer source of hook styles than a direct competitor with a small budget.