Hold Rate: The Video Ad Metric That Decides Spend

The hold rate metric decides whether a Meta video ad earns more spend or quietly dies in the audit pile. It measures the share of viewers still watching past the 15-second mark or 75% completion, which is the threshold where intent compounds and conversions actually book. Hook rate gets you the audition. Retention gets you the role. Most accounts obsess over the first three seconds and ignore the next twelve, which is exactly why 80% of their video tests stall in learning phase. This guide walks the benchmarks, the common killers, and the adlibrary-first workflow to find and fix retention drops before the budget burns.

TL;DR: Hold rate measures viewers retained past 15 seconds or 75% completion of a video ad. A healthy cold-traffic number sits at 15-25% on Meta and 10-18% on TikTok. Below 10% means your mid-roll buildup is killing intent before the offer lands. Diagnose the cut where viewers drop using saved-ads and ad-timeline-analysis on adlibrary, fix the dead zone, then push spend.

What hold rate actually measures

Hold rate is the percentage of viewers retained from the 3-second hook past the 15-second or 75% completion mark, depending on which definition your media buyer was raised on. Meta's official Ads Manager surfaces it as Video Plays at 75% divided by 3-Second Video Plays. Most performance teams calculate it as 15-Second Video Plays / 3-Second Plays because 15 seconds is where the offer typically appears in modern UGC.

The signal is brutally specific. Hook rate tells you the thumb stopped. Retention tells you the message landed. Two ads can share an identical 30% hook rate and produce a 6x ROAS gap because one held 22% of viewers to the CTA and the other held 4%. We see this pattern weekly across cold-traffic accounts.

Three definitions you will encounter:

• Meta hold rate (75%): video_p75_watched_actions / video_3sec_watched_actions. Reported in Meta's Ads Manager metrics reference.
• 15-second hold: video_15_sec_watched_actions / video_3sec_watched_actions. The cleaner DTC metric.
• Average watch time: seconds-based, not a ratio. Useful as a sanity check but not a benchmark.

The 15-second variant is what Motion App's benchmark dataset reports across 200M+ ad impressions, and it is the version most creative strategists optimize against. Pick one definition and lock the team to it. Mixing the two during a creative review wastes 40 minutes and ends in the wrong call. The same warning applies when comparing across Instagram and Facebook placements, which surface slightly different retention buckets.

2026 hold rate benchmarks by platform and angle

Benchmarks shift by vertical, platform, and creative format, but the floor and ceiling are stable enough to call. Below cold-traffic ranges based on aggregated 2025 data from Foreplay's creative analytics report, Triple Whale's ecommerce ad benchmarks, and TikTok's official creative center.

A few field notes on reading the table:

1. Thumb-stop ratio and hook rate are the same metric. TikTok calls it thumb-stop, Meta-native buyers call it hook rate. Same numerator, same denominator, same numbers. Anyone selling you a different definition is selling you a course.
2. TikTok retention runs 5-10 points lower than Meta on identical creative because the platform autoplays the next video at 5 seconds. Compare to TikTok norms, not Meta norms.
3. Static-to-video account migrations typically post first-week numbers around 6-9%. That is fine if the team understands creative testing as a 3-week resolution problem rather than a same-day verdict.

If your account sits below the floor for three weeks running, the issue is rarely targeting. It is the middle of the video. See our breakdown of how to analyze ad performance for the diagnostic order, and the winning ad elements database for what holds attention in your specific category.

Step 0: Diagnose hold rate via adlibrary first

Before you re-edit anything, find a competitor ad in your category that is winning the retention battle and reverse-engineer the cut points. This is the moat. Most teams skip it and edit blind, which is why their second test fails for the same reason as the first.

The workflow on adlibrary's saved-ads takes about twenty minutes:

1. Filter by category and longevity using unified-ad-search. Long-running competitor ads almost always carry strong hold rates because Meta will not keep paying for low-retention creative. Surface the 30+ day runners in your vertical first.
2. Save 8-12 long-runners to a saved-ads board. Mix top-of-funnel and mid-funnel angles so you can see how each holds attention differently. The creative angle you choose determines half the result before any edit decisions are made.
3. Run ai-ad-enrichment to pull scene-level metadata: shot type, on-screen text, audio pattern, and CTA placement timestamp.
4. Open ad-timeline-analysis on each saved video. The timeline marks every cut, on-screen text change, and audio shift to the second. Cluster the cut frequency between seconds 5-15, the dead zone where most accounts lose viewers.
5. Compare to your own ad in ad-detail-view. Lay your timeline next to the winning competitor's. The gap is almost never the hook. It is a missing pattern interrupt around second 7-10 or a slow buildup before the value prop.

We have looked at thousands of in-market ads through this lens and the pattern is depressingly consistent. High-retention creative cuts every 1.5 to 2.5 seconds through the middle. Low-retention creative lets a single shot run 4-6 seconds. The platform doesn't care about your storyboard. It cares whether the viewer keeps watching.

Skip this step and you are guessing. Run it once and the rest of this article costs you nothing. The full creative strategist research workflow closes the loop on how teams operationalize this.

The four hold rate killers and how to fix them

Across audited accounts, four patterns account for roughly 80% of low-retention creative. Each has a specific fix that costs a re-edit, not a re-shoot.

A note on the slow buildup fix. UGC that opens with "I tried this product" and spends five seconds explaining context does not survive cold traffic. The body of the ad creative has to start delivering by second 5 or it dies. Foreplay's UGC analysis consistently shows the highest-retention UGC ads compress the "what is this product" answer into the second-3 to second-5 window. The reasons-to-care list comes after.

The mid-roll dead zone is the hardest to spot internally. Your team has watched the ad fifty times. They can't feel the boredom anymore. The fix is mechanical: timeline the cuts, count the seconds between them, and force a change every 2 seconds in the second 5-15 window. Doesn't matter if it is a zoom, a text reveal, a B-roll cut, or a sound effect. The viewer's brain just needs new information. Spokesperson video ads particularly suffer from this because the talking-head shot rarely changes on its own.

For the CTA framing problem, study cold-audience hook patterns. The hooks that work share one trait with the CTAs that hold viewers to the end. They make inaction feel costly. "Shop now" doesn't. "Stop overpaying for X" does. The same logic applies to video ads for ecommerce stores where price reveal timing is decisive.

Hold rate vs hook rate vs thumb-stop ratio

Three metrics, three different decisions. Mixing them is why creative reviews degenerate into vibes.

• Hook rate / thumb-stop ratio answers: did the scroll stop? It is a signal about thumbnail, opening frame, and audio attack. Treated as the gate metric. Below 15% on cold traffic and the ad doesn't get a second look. Read our hook rate breakdown for the diagnostic.
• Hold rate answers: did the message land? It is a signal about pacing, narrative compression, and offer clarity. Treated as the scaling metric. Below 10% and you cannot scale even if hook rate is healthy.
• CTR (link) answers: did the offer convert intent? It is a signal about CTA framing and price-perceived-value alignment. Treated as the conversion metric. The basic math sits in our CTR calculator.

A useful frame: hook rate is the audition, hold is the performance, CTR is the close. An ad needs all three. Most accounts only optimize the first.

The order of operations matters in creative testing. Fix hook rate first because nothing else matters if no one is watching. Then fix retention, because hook rate without hold is just expensive impressions. Then optimize CTR, because hold without click is engagement theater. Skip a level and you will mis-diagnose for weeks.

For the underlying math on what each metric tells you about ad fatigue, the rule is: hook rate degrades first as audiences saturate, retention degrades second as creative becomes familiar, CTR degrades last as the offer itself wears out. That sequence is also the order in which you should refresh creative.

Reading the retention curve, not just the headline number

Hold rate is a single number. The retention curve is the truth. Pull the per-second video play data from Meta Ads Manager and chart drop-off second-by-second. The curve will tell you exactly which cut killed the ad.

Three retention curve shapes you will see repeatedly:

• Cliff at sec 3-4: Hook rate ok but viewers leave the moment the hook resolves. The pivot from hook to body is too sharp. The fix is to bridge with a one-second continuation of the hook's tension before introducing the product.
• Slow bleed sec 5-15: A linear drop with no clear cliff. Pacing problem. Cuts are too slow. Insert pattern interrupts at second 7 and second 11.
• Cliff at sec 14-15: Strong retention to the offer reveal, then a wall. The offer itself or the framing is wrong. Could be a price reveal that breaks trust, a CTA that feels mismatched to the build, or an "ask" that comes before earning the right to ask.

Triple Whale's video creative dashboard and Motion App's per-frame analytics both surface retention curves at the second level. If you can't get them through your stack, pull video_p25_watched_actions, video_p50_watched_actions, video_p75_watched_actions, video_p100_watched_actions from the Meta Marketing API and chart them yourself.

The shape of the curve is the diagnostic. The headline percentage is just an aggregate. Two ads with identical 18% retention can have wildly different curves and require completely different fixes. Optimize the curve, not the headline metric. The same applies to your creative iteration loop, which should ingest the curve, not the average.

When to act on hold rate signal

Retention is not a same-day metric. Acting on it after 24 hours of data is the second-most common mistake in performance creative. The first is acting on ROAS in 24 hours. Both are noise.

The thresholds for acting:

• Day 1-2 (200-500 impressions): Don't read hold rate at all. Statistical floor too low. The number is gossip.
• Day 3-5 (1500-3000 impressions): Read it as a directional signal. If at floor (<8%), kill the ad. The structural problems won't fix themselves with more spend. If healthy, let it run.
• Day 7+ (5000+ impressions): Now actionable. Compare to your account benchmark, not the universal one. A 14% hold in a long-form coaching offer is healthy. The same number in a 7-day product DTC test is weak.
• Day 14+: If retention has dropped 30% from week 1, you are watching creative fatigue emerge. Not a creative problem, an audience saturation problem. Refresh the angle, don't refresh the edit.

The other rule: this metric matters most on cold traffic. Warm and retargeting audiences self-select for relevance, so retention runs 30-50% higher there and the number is less diagnostic. Track it, but don't kill warm ads on hold rate alone. They will outperform on CTR even with a 9% number.

For the underlying creative testing framework on how long to run before reading retention, the short version is: 3 days minimum, 7 days for confidence, 14 days before declaring a winner. Andromeda makes the spend efficient inside that window. Your job is to give it enough signal.

Frequently asked questions

What is a good hold rate for Meta video ads?

A healthy cold-traffic number sits between 15-25% measured at 15 seconds, or 8-15% measured at 75% completion. Anything below 10% (15-second) signals a structural retention problem in the middle of the ad and should be diagnosed before scaling. Best-in-class DTC accounts hit 25-30% on long-running creative. See the benchmarks table for category-specific norms.

How is hold rate different from hook rate?

Hook rate measures whether the scroll stopped, calculated as 3-second video plays / impressions. Hold rate measures whether the message landed, calculated as 15-second plays / 3-second plays. Hook rate is the audition. Retention is the performance. Both matter for scaling, but they diagnose entirely different creative problems.

Why is my hold rate dropping after 7 days?

A 30%+ drop after week one usually signals creative fatigue, not a creative defect. The audience has saturated and the angle has lost novelty. Refresh the angle and the hook, not just the edit. Run the diagnosing ad fatigue workflow to confirm before re-shooting.

Can I improve hold rate without re-shooting the ad?

Yes, in most cases. Roughly 70% of retention problems are pacing problems, slow cuts in seconds 5-15, and they are fixable in the edit. Tighten cuts to a maximum of 2-second intervals through the middle of the ad, add a pattern interrupt at second 9-10, and preview the value prop by second 5. Re-edit, re-test, re-read after 5 days.

Does hold rate matter on TikTok the same way?

The principle is identical, the numbers are different. TikTok retention runs 5-10 points lower than Meta because the platform autoplays the next video at 5 seconds, so the effective denominator changes. Use TikTok creative center benchmarks for that platform, not Meta benchmarks. The diagnostic logic, find the cut where viewers leave and insert pattern interrupts, is the same.

Bottom line

Hook rate buys the audition. Hold rate decides whether you get the role and how much you get paid for it. The accounts that scale on Meta in 2026 do not have better hooks because everyone has decent hooks now. They have better second 5-15s, because they diagnose the dead zone using competitor benchmarks before they ever press record. Find the cut, fix the cut, then ship the spend.