Creative analysis
Not a wall of AI prose. A structured breakdown of one short-form video — eight dimensions, each line labeled by how confident we actually are. We separate what we saw from what we measured from what we’re guessing. We never tell you a video caused a result.
Observation · Public metric · Calculated signal · Model classification · Strategic hypothesis
The honesty labels
Most “AI analysis” blends a fact, a number, and an opinion into one confident paragraph. We don’t. Each statement gets one of five labels so you always know how much weight it can hold.
Something visible in the video. Anyone watching could confirm it. No interpretation.
“The speaker breaks eye contact at 0:03 and looks down before the first claim.”
A number the platform shows publicly — views, likes, comments, saves. Reported, not inferred.
“412k views, 9.1k saves, 38k likes as displayed on the post.”
A figure we derive from public metrics — like an outlier multiple against the account’s own baseline. Math, not magic.
“8.4× the account’s median view count over its last 30 posts.”
A category a model assigned — hook type, format, pacing bucket. A guess with a name, and it can be wrong.
“Hook classified as “confession / abandoned best-practice.””
Our opinion about why it may have worked. The least certain layer. Never a claim of cause.
“The reversal likely created curiosity because it contradicts a belief the audience holds.”
The rule
Correlation is labeled as correlation. We will not write “this hook drove the views.” We’ll write what we observed, what the public metrics were, and where our hypothesis begins.
What a breakdown contains
We read each video along the same axes so analyses are comparable across accounts and weeks. Here’s a real-shaped example, abbreviated — the confession hook from a demo skincare account.
Hook
What the first two seconds do, and the move that earns the next two.
- ObservationOpens mid-sentence with no logo, no intro, no music ramp — straight into a first-person admission.
- Model classificationHook type: confession that abandons a widely-accepted best practice.
- Strategic hypothesisContradicting a held belief likely opens a curiosity gap the viewer wants closed.
Structure
The order of beats — setup, tension, payoff — and where attention is spent.
- ObservationThree beats: claim (0–3s), reasoning (3–18s), single concrete proof (18–24s). No recap.
- Calculated signalProof arrives at roughly 75% of runtime, late relative to the account’s usual midpoint payoff.
- Strategic hypothesisDelaying the payoff may hold viewers who would otherwise drop after the claim.
Format
The container — talking head, voiceover-over-broll, screen record, skit.
- Model classificationSingle-take talking head, handheld, no cuts.
- ObservationOn-screen captions track speech; no B-roll inserts.
- Strategic hypothesisUnedited single-take may read as candid, which can fit a confession framing — though edit style alone rarely drives reach.
Visual execution
Framing, lighting, motion, and on-screen text — the craft layer.
- ObservationTight chest-up framing, soft window light from camera-left, subject centered.
- ObservationCaption emphasis (bold + color) lands on “stopped” and “every expert.”
- Strategic hypothesisEmphasis on the reversal words may reinforce the hook for sound-off viewers.
Emotional mechanism
The feeling the video moves through, and why that sequence might land.
- Model classificationEmotional arc: skepticism → recognition → relief.
- Strategic hypothesisNaming a frustration the audience already feels can build trust before the claim is justified.
- Strategic hypothesisRelief at the end may be what makes the idea feel worth saving rather than just watching.
Why it likely worked
Our best read — stated as a hypothesis, sitting next to the public number that makes it worth taking seriously.
Calculated signal The post reached 8.4× the account’s median. Strategic hypothesis The most likely contributor is the belief-reversal hook, which opens a curiosity gap the audience wants closed. We can’t isolate it from timing, distribution, or the sound — so we hold this loosely.
Reusable mechanism
Open by abandoning a best practice your audience accepts, justify the reversal with a single concrete proof, and let relief — not the product — be the payoff.
Preserve the mechanism
- The hook mechanism — abandon a held best-practice, then justify it
- The emotional sequence — skepticism, recognition, relief
- The late, single-proof structure
- The candid, unedited delivery register
Replace the execution
- The exact words and the specific “best practice” named
- The product, audience, and the proof itself
- The face, voice, location, and wardrobe
- The caption styling and on-screen emphasis
What not to copy
The most useful part of a breakdown is often the warning. Copying the wrong layer is how a winning pattern turns into a flop.
Anything tied to one creator
Their face, accent, and personal story are not the mechanism. Lifting them reads as imitation and rarely transfers.
The literal claim
Repeating the same “stop doing X” line in your category can be inaccurate or off-brand. Reuse the move, not the sentence.
Surface trends with a short half-life
A trending sound or transition may explain a spike that won’t repeat. We flag these so you don’t build a brief on them.
Anything we labeled as a hypothesis only
If the strongest support for a takeaway is our opinion, treat it as a bet to test — not a rule to follow.
Our standing rule
Reach is shaped by the hook, the timing, the account’s distribution, the audience, the sound, and luck we can’t see. We can describe what a video did and report what the platform showed. We can’t prove which part earned the views — and we won’t pretend to.
So a breakdown is a starting hypothesis, not a guarantee. The honest path is to reuse the mechanism, ship your version, and let your own numbers settle the argument.
Said plainly
- ✓“This post reached 8.4× the account’s median.” — a measured fact.
- ✓“The reversal hook likely contributed.” — a labeled hypothesis.
- ✕“This hook went viral and will for you too.” — we won’t write this.
- ✕“The lighting drove the saves.” — unfalsifiable, so we don’t say it.
Share a Reel or TikTok. We’ll return the eight dimensions, the reusable mechanism, and a clear list of what not to copy — with every claim labeled.