You did the work in the library. Four videos, one competitor, all live for months. You open the first clip, 38 seconds, watch it three times, write down the opening line, and send it to your editor.

Two weeks later your version is dead and theirs is still running.

Short answer: Analyzing a competitor's video ad means writing down its structure and its timing, not its words. Watch the longest-running video twice, once muted and once with the screen covered, log where the promise, the proof and the call to action land, then check whether the same pattern holds across three of their durable videos.

The takeaways

Pick by runtime. A video kept live past 30 days has cleared a month of that advertiser's own kill decisions. The best-produced one may have launched on Monday.
Watch it twice, once muted and once with the screen covered. Feed video autoplays silent, so the visual hook and the spoken hook are two openers competing for the same three seconds.
One video is an anecdote. A beat that appears in a single durable ad is a coincidence. It becomes a pattern when it repeats across three of their long-runners.

Which competitor video is worth breaking down?

The one that has been running longest. Runtime is the only performance signal a public ad library hands you, and a video still live after 30 days has survived roughly a month of its owner's kill checks. Production quality tells you about their budget and nothing about their results.

Most teardowns go wrong right here. You pick the video that impressed you, which selects for craft, and craft was never your bottleneck. Sort by runtime, let the boring one win.

The honest limit: a competitor who kept a weak video live for 90 days is telling you more about their review process than about their profit. Some shops kill below a ROAS floor every Friday, some brands never look. More on that in what a long-running ad does and does not prove.

Why do teardown checklists leave you with nothing to brief?

Because they ask for categories. Hook archetype: curiosity gap. Proof type: testimonial. CTA coherence: strong. Every field is answerable after one viewing, no answer can be wrong, and what you end up with is a filled-in form nobody can build from.

Categories are also the cheapest thing in the market. Everyone knows the hook archetypes. Nobody can tell you off the top of their head that this competitor states the promise inside 5 seconds in all four of its durable videos and holds the product back until halfway.

That second thing is a number, and numbers compare: against their other videos, against a rival's, against the one running in your own account today. A label can only be agreed with. So log timestamps and let the categories fall out of the timeline afterwards.

What do you write down in the first three seconds?

Two hooks, on two lines. Meta's video metrics count a play at the three-second mark, and feed video starts muted, so the opener you are studying was doing two jobs at once for an audience that mostly never heard it.

Run the clip with the sound off first. Note what stopped you: a face, a hand entering frame, hard motion, burned-in text, a product mid-use. Then play it again with the window covered. Does the spoken line stand on its own, or is it a caption for a picture you can no longer see?

Most lifted hooks die exactly here. The buyer copies the sentence out of an ad whose stopping power lived in the picture, then wonders why the words did nothing. Which channel carries your own opener is worth deciding on purpose, and the writing side of that sits in how to write Facebook ad hooks.

How do you log the structure without stealing the script?

Write the beats in your own words, each with the second it starts. A beat is any point where the video changes subject: problem to demonstration, demonstration to price, price to proof. Four to eight of them will cover most direct-response videos.

Then mark four positions. When the promise gets stated. When the product first appears. When the call to action arrives. Whether the proof came before that call or after it.

What you hold at the end is a shape: visual hook at 0, promise at 4 seconds, first demonstration at 9, objection handled at 22, offer at 31. No sentences of theirs on the page. Hand that skeleton to an editor with your own offer inside it and you have built something original on a structure the market already pays to run.

When does a beat count as a pattern?

When it repeats across three of their durable videos. One video is one editor's decision on one Tuesday, and a teardown of it produces a story rather than a finding. Three videos that all put the promise before the 5-second mark are a house style someone kept funding.

Lay the three timelines next to each other and read down the columns. Beats that line up are the pattern. Beats that appear once get crossed out, however clever they looked.

If you have the appetite, run the same pass on a second competitor. A rhythm two advertisers arrived at separately says more about the market than about either of them.

What a teardown can never tell you

Whether any of it made money. A public library carries no spend, no ROAS and no watch-through, so the exercise produces a hypothesis and your own test settles it. The hook is the least portable piece of what you decoded: it is carried by their offer, their price and the recognition their brand already has with that viewer, and none of those three travel with the sentence. The skeleton moves. The words stay behind.

Keeping this current by hand costs an evening per competitor, which I only ever managed for the accounts that mattered. That upkeep is why the pass is built into Adscalr: it pulls the Meta, TikTok and Google ad libraries into one daily-refreshed dataset (nothing here is real-time), flags anything past 30 days as a durable winner, and analyzes videos frame by frame, longest-running first, for hook, pacing, emotional arc, CTA placement and script structure. Transcript where one exists, selective by design.

Same timeline this post asks you to build by hand, which is the whole job of competitor intelligence.

How to analyze competitor video ads