Drop in your TikTok ad.
Get a brutally honest review.

Hook score, scored rubric, top issues to fix, and rewrites — in under a minute. No fluff. Just the work.

Video (MP4 / MOV / WebM)

Drop an MP4 / MOV — frames are extracted in your browser, only the frames upload

Planned caption (optional)

// THE_RUBRIC

What it scores you on

Every analysis returns numbers calibrated to 2026 TikTok industry benchmarks — not vibes.

Hook rate
Predicted % of viewers who watch past 3 seconds. <20% is weak, 30+ strong, 40+ viral-tier.
Six-dimension creative rubric
Hook (heaviest), clarity, CTA, brand fit, native feel, pacing. Every score cites a specific frame, line, or word as evidence.
Hook formula classification
Pattern interrupt · curiosity gap · controversy · social proof · PAS · value promise · or “none” (the kiss of death).
Ad format classification
Talking head · UGC testimonial · unboxing · before/after · GRWM · POV · demo · founder pitch · skit · tutorial · list/countdown.
10 predicted performance metrics
Hook rate, hold rate, completion rate, avg watch seconds, engagement, save, comment, share, follow CVR, CTA CTR.
Production analysis
UGC vs. hybrid vs. studio-polished, sound-off comprehension %, on-screen text coverage, brand chrome density, voiceover style, cuts per minute.
Algorithm signals
FYP distribution likelihood, sound trend alignment, hashtag effectiveness, first-frame thumb-stopper, Spark Ad suitability.
Business projections
CPM / CPC / CVR brackets, creative lifespan in days, recommended daily budget range, variants to test.
Risk flags
Unsubstantiated medical/financial claims, missing before-after disclaimers, trademark, audio copyright, generic template clones, deceptive patterns.

// TIERS

Three ways in

Free during beta. Quick Check and Human review open after launch.

Mode	Cost	Limit	Latency	What you get
Deep BenchLive	Free during beta	2 / IP / 24h	~30–60s	Hook rate prediction, 6-cat rubric, 10 predicted metrics, production analysis, 5 algorithm signals, business projections, risk flags, ranked top issues with concrete fixes, hook + caption + CTA rewrites + shot-list edits, vertical benchmark match. Video: per-frame timing read + Whisper voiceover transcript.
Quick CheckSoon	Coming soon	—	~10s	5-category scorecard + 3 actionable fixes. For when you're iterating fast.
Human strategistSoon	$89 / ad	—	24–48h	A real TikTok performance marketer writes notes by hand and emails them.

Deep BenchLive

Cost: Free during beta
Limit: 2 / IP / 24h
Latency: ~30–60s

Hook rate prediction, 6-cat rubric, 10 predicted metrics, production analysis, 5 algorithm signals, business projections, risk flags, ranked top issues with concrete fixes, hook + caption + CTA rewrites + shot-list edits, vertical benchmark match. Video: per-frame timing read + Whisper voiceover transcript.

Quick CheckSoon

Cost: Coming soon
Limit: —
Latency: ~10s

5-category scorecard + 3 actionable fixes. For when you're iterating fast.

Human strategistSoon

Cost: $89 / ad
Limit: —
Latency: 24–48h

A real TikTok performance marketer writes notes by hand and emails them.

Soft-cap training mode: server hard cap is 10/day on Deep Bench. The cap modal pops at the visible limit (2) — dismissing acks for the session and runs go through silently up to the server cap.

// HOW_IT_WORKS

What if you knew, before launch, the ad was wrong?

TokBench gives you notes you can act on — built from how the actual top-performing TikTok ads work in 2026. Fix what's broken. Ship ads that convert.

YOU

Drop the ad. Pick a format.

Paste a script, drop an MP4, paste a TikTok URL, or upload a still. Frames extract in your browser, voiceover transcribes via Whisper, the URL pulls metadata + cover via TikTok's official oEmbed — pick whichever surface matches what you've got.

THE AI

TikTok-tuned rubric. Not generic AI.

Claude Opus 4.7 runs the work through a rubric calibrated to 2026 TikTok benchmarks. 71% of retention is decided in the first 3 seconds. Sound-off is the default assumption. Polished commercial-look caps at 50. “Hey guys today I want to tell you about” is an automatic sub-30 hook score.

YOU GET

Specifics, not platitudes.

Predicted hook rate, six-dimension scorecard, top issues with concrete fixes, three rewritten hooks, a caption, three CTAs, shot-list edits, and a benchmark match against the top ads in your vertical — with what those winners do that yours doesn't.

// PRIVACY

What happens to your ad

Uploaded videos and images go to Vercel Blob at unguessable paths. URLs are not enumerable.
Anthropic’s commercial API terms forbid training on submitted data.
We don’t keep an analytics fingerprint of your ads.
Manual delete: email legal@tokbench.ai.