# Video Creative Metrics Benchmarks 2026

Cited 2026 benchmark ranges for thumbstop rate, hold rate, and ThruPlay on Meta — with segmentation rules, source links, and interactive analyzers to score your creative.

**Category:** Marketing Benchmarks  
**Difficulty:** intermediate  
**Last Updated:** 2026-06-12

## Summary

Compare video creative on three thresholds — 3s hook, 15s hold, and ThruPlay — only after segmenting by length, placement, and audience temperature. Every cell below cites a public source.

## Steps

1. **Segment before you compare**

Lock video length bucket, placement (Feed vs Reels vs Stories), and audience temperature (cold vs warm) before reading any benchmark cell.

2. **Score the hook (3-second threshold)**

Pull thumbstop / hook rate first. Below ~18% on Meta cold prospecting Feed usually means the opening frame fails before the story starts.

3. **Score mid-funnel retention (15-second threshold)**

Hold rate isolates viewers who survived the hook. A strong thumbstop with weak hold points to pacing or story structure — not the first frame.

4. **Score engagement depth (ThruPlay)**

ThruPlay counts completions of sub-15-second videos plus 15-second-plus plays of longer ones — so it qualifies more views than a raw 15-second-video-views count. Use it for creative ranking, but pair with hold rate so short, shallow creative does not look falsely healthy.

## Methodology: How These Benchmarks Are Built

![Segment before you compare — lock video length, placement (Feed/Reels/Stories), and audience temperature (cold→buyers) before reading any benchmark cell.](/images/resources/guides/video-creative-metrics-benchmarks-2026/segment-before-compare.webp)
*Lock all three — length × placement × audience — before reading a single benchmark cell. Apples to apples, or the number lies.*

This page consolidates **cited public ranges** from platform documentation, practitioner reports, and an **independent web research pass** (June 2026) across adlibrary.com, Vaizle, Billo, Sepia Lab, Zeely, eonik.ai, MHI Growth Engine, Affect Group, Dynamoi, Triple Whale, and Nielsen-derived completion compilations. We do **not** publish a single “industry average” without context.

**Research process**

1. Start from Meta’s official metric definitions (3-second views; the 15-second-or-completion ThruPlay threshold)
2. Pull third-party tier tables that name their formula and year (Sepia Lab’s 2026 hook-rate roundup is the best cross-vendor comparison we found)
3. Cross-check practitioner reports that publish ranges **with** segmentation (length, placement, or audience temperature)
4. Flag when two sources use the **same metric name but different formulas** (hold rate is the worst offender — see the definitions section below)
5. Drop any cell where no 2024–2026 source publishes the full segmentation — we note dropped cells explicitly rather than inventing numbers

**Inclusion rules**

- Every table row must name a source and the segment it applies to
- Meta ThruPlay ranges use **Video Plays** as the denominator (what Meta reports) unless a row notes otherwise — on Meta, autoplay keeps Video Plays ≈ impressions, so the bands hold either way; confirm your dashboard’s denominator before comparing
- Thumbstop / hook rate = 3-second video views ÷ impressions on Meta (Meta has **no official hook-rate benchmark** — Sepia Lab, 2026)
- Hold rate on this page defaults to **15-second video views ÷ 3-second video views** unless a row explicitly notes a different formula

**How to use this page**

1. Pick your platform, length bucket, and audience temperature
2. Confirm your formula matches the cited source’s formula
3. Enter your metrics in the linked analyzers below
4. Compare range-relative verdicts — not raw leaderboard rankings
5. Document learnings in your [creative testing framework](/resources/guides/creative-testing-framework-guide)

> **Benchmarks are diagnostic, not targets** (info)

A “below benchmark” hook on a retargeting audience may still deliver strong ROAS. Use benchmarks to localize creative problems (hook vs body vs CTA), not to override downstream conversion metrics.

## Metric Definitions (Meta)

| Metric | Formula | Threshold | Source |
| --- | --- | --- | --- |
| Thumbstop / hook rate | 3-second video views ÷ impressions | 3 seconds | Meta Ads Manager columns; practitioner convention ([Billo](https://billo.app/blog/hook-rate-to-hold-rate/), [Vaizle 2026](https://insights.vaizle.com/hook-rate-hold-rate/)) |
| Hold rate (15s retention) | 15-second video views ÷ 3-second video views | 15 seconds among hooked viewers | [adlibrary.com Hold Rate guide](https://adlibrary.com/posts/hold-rate); [Sepia Lab 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks) (Motion 15s dataset cited therein) |
| Hold rate (ThruPlay funnel) | ThruPlays ÷ 3-second video views | 15 seconds or completion (sub-15s) | [Vaizle 2026](https://insights.vaizle.com/hook-rate-hold-rate/); [Sepia Lab hook roundup 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks) |
| Hold rate (watch depth) | Average video play time ÷ video length | Continuous (% of ad watched) | [MHI Growth Engine DTC guide](https://mhigrowthengine.com/blog/hold-rate-video-ads-dtc/) |
| ThruPlay rate | ThruPlays ÷ video plays | 15 seconds or completion (sub-15s) | [Meta Business Help Center — About ThruPlay](https://www.facebook.com/business/help/2051461368219124) |
| Video completion (VCR) | 100% video views ÷ video plays | Full watch | Nielsen completion data (via [Amra & Elma compilation](https://www.amraandelma.com/best-video-ad-completion-rate-statistics/)) |

Official ThruPlay definition: a ThruPlay counts when a viewer watches a video to completion (videos under 15 seconds) or for at least 15 seconds (longer videos); Meta treats ~97%+ as completion. It does **not** count ad clicks — a common misconception worth unlearning. Source: [Meta Business Help Center — About ThruPlay](https://www.facebook.com/business/help/2051461368219124).

Because ThruPlay also fires on sub-15-second completions that a raw 15-second-video-views count never captures, pair ThruPlay with hold rate when diagnosing watch depth. See [Hold Rate vs ThruPlay Rate](/resources/comparisons/hold-rate-vs-thruplay-rate).

## Hold Rate Has Four Common Definitions — Pick One

Independent research (June 2026) across [adlibrary.com](https://adlibrary.com/posts/hold-rate), [Vaizle](https://insights.vaizle.com/hook-rate-hold-rate/), [Billo](https://billo.app/blog/hook-rate-to-hold-rate/), [Sepia Lab](https://sepia-lab.com/en/blog/hook-rate-benchmarks), and [MHI Growth Engine](https://mhigrowthengine.com/blog/hold-rate-video-ads-dtc/) found **four** formulas all called “hold rate.” They are **not interchangeable** — a dashboard showing 45% and one showing 18% can describe the identical ad.

| Definition | Formula | Typical cold range | Who uses it |
| --- | --- | --- | --- |
| 15-second retention | 15-second video views ÷ 3-second video views | 15–25% healthy; <10% kill floor | [adlibrary.com](https://adlibrary.com/posts/hold-rate) (Foreplay + Triple Whale composite); [Sepia Lab 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks) |
| 75% completion retention | Video plays at 75% ÷ 3-second video views | 8–15% healthy; <5% kill floor | Meta Ads Manager `video_p75_watched_actions`; [adlibrary.com](https://adlibrary.com/posts/hold-rate) |
| ThruPlay funnel hold | ThruPlays ÷ 3-second video views | 40–50% average; 60%+ strong | [Vaizle 2026](https://insights.vaizle.com/hook-rate-hold-rate/); [Motion via Sepia Lab 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks); Billo Meta reporting |
| Watch-depth percentage | Average video play time ÷ video length | 28–40% on 30s DTC ads | [MHI Growth Engine 2026](https://mhigrowthengine.com/blog/hold-rate-video-ads-dtc/) |

All four are legitimate — they answer different questions. Lock one per review.

**Why the numbers look “wrong” side by side:** Vaizle’s ~45% “hold” and adlibrary’s ~18% “hold” can describe the **same ad** because ThruPlay counts sub-15-second completions in addition to 15-second-plus plays, inflating the numerator versus raw 15-second video views. MHI’s 35% on a 30-second ad means ~10.5 seconds average watch time — a third metric entirely.

**Default on this page:** segmented hold tables below use **15-second views ÷ 3-second views** unless a row says otherwise. When a vendor dashboard shows 40%+, check whether it is the ThruPlay funnel definition before comparing to our 15–25% cold band.

> **Lock one formula per creative review** (warning)

adlibrary.com notes that mixing 15-second and 75% definitions in the same review wastes time and leads to wrong kill/scale calls. Create separate custom columns in Ads Manager and never compare Vaizle-style ThruPlay÷3s hold to adlibrary-style 15s÷3s hold without converting.

## Thumbstop & Hook Rate Benchmarks (Meta)

**Meta thumbstop / hook rate by placement and audience (3s ÷ impressions)**

| Segment | Typical range | Median | Source | Note |
| --- | --- | --- | --- | --- |
| Feed, DTC prospecting (cold) | 18% – 28% | ~23% | [Sepia Lab 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks) (Motion in-feed dataset); [Affect Group 2025](https://affectgroup.com/blog/understanding-meta-ads-video-metrics-how-to-measure-analyze-and-improve-performance/); [MHI Feed 2026](https://mhigrowthengine.com/blog/hook-rate-meta-ads-guide/) | Below ~18% usually signals opener failure on cold traffic |
| Reels, DTC prospecting (cold) | 24% – 36% | ~30% | [Sepia Lab 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks); [adlibrary.com Hook Rate 2026](https://adlibrary.com/posts/hook-rate) (30–40% “good” band) | **Placement conflict:** [MHI 2026](https://mhigrowthengine.com/blog/hook-rate-meta-ads-guide/) reports Reels 5–10pp below Feed — compare Reels-only to Reels-only |
| Stories, DTC | 22% – 32% | ~27% | [Sepia Lab 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks); [MHI Stories 2026](https://mhigrowthengine.com/blog/hook-rate-meta-ads-guide/) (12–22% avg, 22–32% strong) | Full-screen helps but skip behavior varies by creative format |
| Retargeting (warm) | 30% – 45% | ~36% | [Sepia Lab 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks); [adlibrary.com 2026](https://adlibrary.com/posts/hook-rate) | Brand recognition inflates hook rate — not a standalone creative-quality signal |
| Cold traffic floor (kill signal) | n/a | n/a | [adlibrary.com 2026](https://adlibrary.com/posts/hook-rate); [get-ryze.ai 2026](https://www.get-ryze.ai/blog/meta-video-ad-benchmarks-thruplay-hook-and-hold-rate-2026) | Below **15%** hook → pause or rebuild opener before scaling spend |

**Practitioner hook tiers (Meta, 3s ÷ impressions).** Meta publishes **no official hook benchmark**; these tiers come from creative-analytics vendors. The first four rows are the cross-vendor comparison in [Sepia Lab’s 2026 hook-rate roundup](https://sepia-lab.com/en/blog/hook-rate-benchmarks); the Zeely row is from [Zeely’s own Meta hook-rate analysis](https://zeely.ai/blog/boost-meta-ads-hook-rate/).

| Source | Needs work | Solid | Good / strong | Exceptional |
| --- | --- | --- | --- | --- |
| Motion (2025) | below 25% | 25–35% | above 35% | baseline target 30–40% |
| Vaizle (2025) | below ~20% | ~20–25% | above 30% (top ads) | n/a |
| AdManage.ai (2025) | below 20% | 20–25% | 30–45% | 45%+ |
| Billo (2025) | below 25% | n/a | 35–45% | n/a |
| Zeely Meta analysis (2025) | below 15% | 20–25% | 30%+ strong | 30–50% after optimization |

**Account-level reality check:** [Billo](https://billo.app/blog/hook-rate-to-hold-rate/) cites ~24% average Meta hook rate across reviewed accounts — top performers push past 28%, optimized campaigns reach 30–50%. When someone quotes “good hook = 40%,” compare against your segment and audience temperature first.

**TikTok hook benchmarks (2-second threshold — not Meta-comparable).** TikTok hook rate uses **2-second video views ÷ impressions**, so numbers run hotter than Meta 3-second rates for comparable attention. Source: [Billo — Hook Rate to Hold Rate](https://billo.app/blog/hook-rate-to-hold-rate/) (Tuff Agency 11-account study cited therein).

| Tier | TikTok (2s) | Meta (3s) |
| --- | --- | --- |
| Elite | 40%+ | 35–45% |
| Competitive | 30–39% | 30–39% |
| Fix-it zone | below 25% | below 25% |
| Account average (Tuff study) | **30.7%** | **~24%** (Billo Meta review) |
| Top quartile (Tuff study) | 40–45% | n/a |

Do not hold TikTok creative to a Meta 3-second number, or vice versa — set the analyzer’s platform toggle correctly.

> **Cells dropped — no citable segmented range found** (warning)

We did not blend TikTok 2-second bands into the Meta placement table above because thresholds and viewing modes differ. YouTube Shorts hook benchmarks were dropped for the same reason — most public reports lack placement-comparable segmentation for 2024–2026. Vertical-specific rows (e.g. Beauty/Skincare) were dropped when no 2024–2026 source published a URL-backed range with placement segmentation — [Benly.ai’s 2026 creative benchmarks](https://benly.ai/learn/ad-creative/ad-creative-benchmarks-2026) reports industry medians but not Meta placement splits by vertical.

**Hook Rate Benchmark Analyzer** — Calculate hook rate for Meta, TikTok, or YouTube Shorts and see where you land vs 2026 cited ranges.
Tool: /resources/tools/analyzers/hook-rate-benchmark

Deep dive: [Video Ad Hook Optimization guide](/resources/guides/video-ad-hook-optimization-guide) · [Thumbstop rate glossary](/resources/glossary/metrics/thumbstop-rate-tsr) · [How to read hook metrics in Meta Ads Manager](/blog/topics/ad-performance/how-to-read-video-metrics-in-meta-ads-manager)

## Hold Rate Benchmarks (Meta)

**Meta hold rate (15-second views ÷ 3-second views) by segment**

| Segment | Typical range | Median | Source | Note |
| --- | --- | --- | --- | --- |
| Feed, DTC prospecting (15–30s) | 15% – 25% | ~20% | [adlibrary.com 2026](https://adlibrary.com/posts/hold-rate) (Foreplay + Triple Whale composite); [MHI 2026](https://mhigrowthengine.com/blog/hold-rate-video-ads-dtc/) (15s mark on cold 15–30s) | Below ~10% = mid-roll problem before scaling |
| Reels (any length) | 18% – 30% | ~23% | [adlibrary.com 2026](https://adlibrary.com/posts/hold-rate) (+5–10pp vs Feed at 15s hold) | Full-screen immersion holds hooked viewers longer |
| Longer video (30–60s), prospecting | 8% – 18% | ~12% | [Triple Whale 2025](https://www.triplewhale.com/blog/facebook-ads-benchmarks); [MHI 2026](https://mhigrowthengine.com/blog/hold-rate-video-ads-dtc/) (30s watch-depth 28–40%) | Harder on long creative; do not compare to sub-15s mechanical hold |
| Retargeting (warm) | 25% – 40% | ~30% | [adlibrary.com 2026](https://adlibrary.com/posts/hold-rate) (30%–50% higher than cold at 15s) | Warm audiences tolerate slower pacing |
| Sub-15s video (mechanical) | 50% – 80% | ~65% | [Meta Business Help Center — ThruPlay](https://www.facebook.com/business/help/2051461368219124) | 15s threshold resolves on completion — weak standalone quality signal |
| Cold traffic floor (kill signal) | n/a | n/a | [adlibrary.com 2026](https://adlibrary.com/posts/hold-rate) | Below **10%** at 15s÷3s → structural mid-roll problem before scaling |

Every row uses the **15s ÷ 3s** definition unless noted.

**Alternate hold definition — ThruPlays ÷ 3-second views** (Vaizle / Billo / Motion via Sepia Lab). Same creative often reads ~2× higher here because ThruPlay also counts sub-15-second completions, not just raw 15-second views:

| Tier | Range | Source |
| --- | --- | --- |
| Needs improvement | below 30% | Motion (2025, via [Sepia Lab](https://sepia-lab.com/en/blog/hook-rate-benchmarks)) |
| Average | 40–50% | [Vaizle 2026](https://insights.vaizle.com/hook-rate-hold-rate/); Motion (2025) |
| Strong | above 60% | Motion (2025) |
| Billo Meta target band | 25%+ minimum; 45% ideal paired with 35% hook | [Billo 2026](https://billo.app/blog/hook-rate-to-hold-rate/) |

**Watch-depth hold (avg play time ÷ length)** — [MHI Growth Engine 2026](https://mhigrowthengine.com/blog/hold-rate-video-ads-dtc/) benchmarks by video length:

| Video length | Average hold | Strong hold |
| --- | --- | --- |
| 15 seconds | 40–55% | 55%+ |
| 30 seconds | 28–40% | 40%+ |
| 60 seconds | 18–30% | 30%+ |

A 35%+ hold on a 30-second DTC ad means the average viewer reaches ~10.5 seconds — deep enough to hear the product pitch. This is **not** the same as 35% ThruPlay rate or 35% 15s÷3s hold.

Hold rate isolates **body** performance: if thumbstop is strong but hold is weak, re-cut seconds 3–15 before touching the opener. Practitioner framing: [Billo — Hook Rate to Hold Rate](https://billo.app/blog/hook-rate-to-hold-rate/) · [Glued.me — Hook vs Hold guide](https://www.glued.me/blog/hook-rate-vs-hold-rate-guide).

## ThruPlay Rate Benchmarks (Meta)

**Meta ThruPlay rate (ThruPlays ÷ video plays) by segment**

| Segment | Typical range | Median | Source | Note |
| --- | --- | --- | --- | --- |
| 15–30s, cold prospecting (Reels/Stories) | 18% – 28% | ~22% | [Affect Group 2025](https://affectgroup.com/blog/understanding-meta-ads-video-metrics-how-to-measure-analyze-and-improve-performance/); [Dynamoi 2026](https://dynamoi.com/learn/instagram-ads/meta-ads-manager-metrics-music-campaigns) (25%+ good, 35%+ great) | Primary creative-quality band for standard performance video |
| 15–30s, cold prospecting (consensus band) | 15% – 25% | ~20% | [Affect Group 2025](https://affectgroup.com/blog/understanding-meta-ads-video-metrics-how-to-measure-analyze-and-improve-performance/); [adlibrary.com 2026](https://adlibrary.com/posts/hold-rate) | Below ~10% ThruPlay on cold 15–30s usually points to hook/early retention failure |
| 30–60s, cold prospecting | 8% – 18% | ~12% | [Triple Whale 2025](https://www.triplewhale.com/blog/facebook-ads-benchmarks) | Longer videos lose roughly half of viewers before 15s on cold traffic — no public URL publishes finer 30–60s ThruPlay segmentation |
| Sub-15s (any vertical) | 70% – 95% | ~85% | [Meta Business Help Center — ThruPlay](https://www.facebook.com/business/help/2051461368219124); Nielsen via [Amra & Elma](https://www.amraandelma.com/best-video-ad-completion-rate-statistics/) (79.4% sub-15s completion) | Mechanically inflated — compare sub-15s creatives to each other, not to 30s+ ThruPlay |
| Retargeting (warm) | 25% – 45% | ~32% | [Affect Group 2025](https://affectgroup.com/blog/understanding-meta-ads-video-metrics-how-to-measure-analyze-and-improve-performance/) | +30%–50% over cold per [adlibrary.com](https://adlibrary.com/posts/hold-rate) |

A *cost-per-ThruPlay* efficiency band ($0.007–$0.024, Strike Social / Varos 2025) circulates in vendor decks but has **no public URL**, so it is excluded here per this page’s every-cell-cites-a-source rule. Read cost-per-ThruPlay from your own account, not an unsourced benchmark.

> **Placement cells dropped** (info)

Industry commentary consistently notes Reels runs ~5–10 percentage points above Feed at the 15-second mark (adlibrary.com; Triple Whale 2025 aggregate ~29% with sub-15s at 53.7%). We cite the Reels delta directionally but do not invent Feed-specific ThruPlay ranges — filter Ads Manager by placement and compare to your account’s Reels-stripped baseline when public segmentation is missing.

**ThruPlay Rate Benchmark Analyzer** — Enter your ThruPlay rate segmented by length and audience temperature for an instant range-relative verdict.
Tool: /resources/tools/analyzers/thruplay-rate-benchmark

Full segmented analysis: [2026 ThruPlay benchmarks blog](/blog/topics/ad-performance/what-is-a-good-thruplay-rate-2026-benchmarks) · [ThruPlay interpretation guide](/resources/guides/thruplay-rate-interpretation-guide) · [Hold rate glossary](/resources/glossary/metrics/hold-rate)

## Sources & Further Reading

**Platform definitions**

- [Meta Business Help Center — About ThruPlay](https://www.facebook.com/business/help/2051461368219124)

**Independent research roundup (June 2026)**

- [Sepia Lab — Hook Rate Benchmarks 2026](https://sepia-lab.com/en/blog/hook-rate-benchmarks) — cross-vendor Meta hook tiers (Motion, Vaizle, AdManage.ai, Billo); placement table (Feed/Reels/Stories); notes no Meta-official hook benchmark
- [adlibrary.com — Hook Rate (2026)](https://adlibrary.com/posts/hook-rate) — tier table (15–45%+); Reels 30–40% good band; kill floor below 15%
- [adlibrary.com — Hold Rate (2026)](https://adlibrary.com/posts/hold-rate) — 15s÷3s cold band 15–25%; p75÷3s band 8–15%; kill floor &lt;10% at 15s
- [MHI Growth Engine — Hook Rate on Meta Ads (2026)](https://mhigrowthengine.com/blog/hook-rate-meta-ads-guide/) — placement bands (Feed 20–30%, Reels 15–25%, Stories 12–22%); Reels vs Feed heuristic
- [get-ryze.ai — Meta Video Ad Benchmarks 2026](https://www.get-ryze.ai/blog/meta-video-ad-benchmarks-thruplay-hook-and-hold-rate-2026) — cold hook 25–35%; kill if &lt;15%
- [Vaizle — Hook Rate & Hold Rate (Jan 2026)](https://insights.vaizle.com/hook-rate-hold-rate/) — hook ~20–25%; hold = ThruPlays÷3s at ~40–50%
- [Billo — Hook Rate to Hold Rate (Jun 2026)](https://billo.app/blog/hook-rate-to-hold-rate/) — Meta avg hook ~24%; Tuff 11-account TikTok avg 30.7%; ThruPlay÷3s hold reporting
- [MHI Growth Engine — Hold Rate for DTC Video Ads (Feb 2026)](https://mhigrowthengine.com/blog/hold-rate-video-ads-dtc/) — watch-depth hold (avg play time ÷ length) by video length
- [Zeely — Boost Meta Hook Rate](https://zeely.ai/blog/boost-meta-ads-hook-rate/) — practitioner tiers (&lt;15% weak, 20–25% solid, 30%+ strong)
- [Madgicx — Custom Metrics in Meta Ads](https://madgicx.com/blog/meta-ads-custom-metrics) — hook/hold custom column formulas; Number vs Percentage format pitfalls
- [Benly.ai — Ad Creative Benchmarks 2026](https://benly.ai/learn/ad-creative/ad-creative-benchmarks-2026) — cross-platform hook medians (12k+ ad sample); no Meta vertical placement splits

**Practitioner benchmark reports (2024–2026)**

- [Affect Group — Key Meta Ads Video Metrics Explained](https://affectgroup.com/blog/understanding-meta-ads-video-metrics-how-to-measure-analyze-and-improve-performance/) — 15–30s cold prospecting ThruPlay 18%–28%; cold band 15%–25%
- [Dynamoi — Meta Ads Manager Metrics for Music Campaigns](https://dynamoi.com/learn/instagram-ads/meta-ads-manager-metrics-music-campaigns) — vertical-specific ThruPlay tiers (25%+ good, 35%+ great)
- [Triple Whale — Facebook Ads Benchmarks 2025](https://www.triplewhale.com/blog/facebook-ads-benchmarks) — aggregate ThruPlay/VTR ~29%; sub-15s vs longer splits
- [Amra & Elma — Video ad completion rate statistics](https://www.amraandelma.com/best-video-ad-completion-rate-statistics/) — Nielsen-derived completion data (79.4% sub-15s)

**Hook → hold funnel framing**

- [Glued.me — Hook Rate vs Hold Rate](https://www.glued.me/blog/hook-rate-vs-hold-rate-guide)
- [FunnelInsiders — Thumbstop Ratio analysis](https://funnelinsiders.com/blog/a-full-spectrum-analysis-of-thumbstop-ratio-tsr-with-data/) — TSR alone weakly predicts ROAS (R²≈0.003 in cited cross-account study)

**Internal cross-links**

- [Thumbstop rate glossary](/resources/glossary/metrics/thumbstop-rate-tsr) · [Hold rate glossary](/resources/glossary/metrics/hold-rate) · [ThruPlay rate glossary](/resources/glossary/metrics/thruplay-rate)
- [How to read video metrics in Meta Ads Manager](/blog/topics/ad-performance/how-to-read-video-metrics-in-meta-ads-manager)

## Retention Curves Beat Headline Hold %

[adlibrary.com’s 2026 hold guide](https://adlibrary.com/posts/hold-rate) argues the **shape** of the retention curve matters more than a single hold percentage. Pull quartile columns from Ads Manager (`video_p25`, `p50`, `p75`, `p100` watched actions) or use Motion / Triple Whale frame analytics when available.

| Curve shape | Where viewers leave | Likely fix |
| --- | --- | --- |
| Cliff at 3–4s | Hook resolves; body never starts | Bridge hook tension one more second before product intro |
| Slow bleed 5–15s | Pacing — no pattern interrupts | Force a cut, text reveal, or audio shift every ~2s in the mid-roll |
| Cliff at 14–15s | Offer/CTA framing breaks trust | Reframe CTA as consequence (“without this, you keep paying X”) |

**Mid-roll dead zone (seconds 8–15)** is the most common cold-traffic failure mode in practitioner audits: high-retention creative changes visual information every 1.5–2.5 seconds through the middle; low-retention creative holds one shot 4–6 seconds. See [Madgicx custom metrics guide](https://madgicx.com/blog/meta-ads-custom-metrics) for hook/hold column setup in Ads Manager.

## The Three-Metric Diagnostic Stack

![The video engagement funnel — 100% impressions → 20–40% reach 15 seconds (hold) → 10–30% ThruPlay, with drop-off points at the hook and the body.](/images/resources/guides/video-creative-metrics-benchmarks-2026/video-engagement-funnel.webp)
*Two drop-off points, two different fixes: a weak hook bleeds viewers at 3s; weak hold loses them in the body. Diagnose which before you re-cut.*

1. **Thumbstop weak, hold weak** — Rewrite the first 1–3 seconds — motion, promise, and audience qualification. Do not edit the body until the hook clears ~20% on cold prospecting Feed.
2. **Thumbstop strong, hold weak** — Hook works; body loses viewers. Tighten pacing, front-load proof, cut dead air between 3–15 seconds.
3. **Hold strong, ThruPlay weak vs peers** — Unusual — check whether sub-15-second completions are inflating ThruPlay or whether placement mix skews the aggregate. Segment Reels vs Feed.
4. **All three strong, ROAS weak** — Engagement depth is not purchase intent. Shift optimization to conversion events or tighten offer/landing alignment.

**Creative Quality Grader** — Score hook, body, and CTA strength against benchmark rubrics in one pass.
Tool: /resources/tools/analyzers/creative-quality-grader

**Key takeaways:**
- Never compare metrics across different length buckets
- Thumbstop diagnoses the hook; hold diagnoses the body
- ThruPlay counts sub-15s completions + 15s-plus plays (not clicks) — pair with hold rate for watch depth
- Drop benchmark cells when the source lacks segmentation — do not average mismatched cohorts
- Your account’s top-quartile history beats generic industry medians once you have volume

## Landing page

https://www.adsights.ai/resources/guides/video-creative-metrics-benchmarks-2026-guide