Facebook Ads Hold Rate: Formula, Benchmarks, Audit Sheet

Facebook ads hold rate measures how many people who watched the first 3 seconds of your video kept watching to a ThruPlay. The formula is ThruPlays ÷ 3-second video plays × 100. You build it from two counts Ads Manager already reports.

Three terms, as Meta defines them:

Worked example, an illustrative ad we built from the sample medians in the next section: 20,000 impressions, 4,180 3-second plays and 920 ThruPlays.

• Hook rate (3-second plays ÷ impressions): 4,180 ÷ 20,000 = 20.9%. Our guide to calculating hook rate covers that metric in full.
• Hold rate (ThruPlays ÷ 3-second plays): 920 ÷ 4,180 = 22.0%.
• ThruPlays ÷ impressions: 920 ÷ 20,000 = 4.6%.

Hook rate is the share of impressions that stopped the scroll. Hold rate only looks at the people the opening already won.

Use ThruPlays ÷ 3-second plays, and write the denominator into the column name. It separates the body of the video from the opening. Dividing by impressions bakes hook rate into the result, so a weak opening makes a good body look bad.

The versions are linked by one identity: ThruPlays ÷ impressions = hook rate × hold rate. The sample below fits it. We call it Biddy's sample: 479 video ads from 17 US ad accounts, each account spending $100K or more over June to August 2026, with $10.1M combined spend on the ads kept. Its median hook rate was 20.9%. Its median "3-second-to-ThruPlay retention," the same ratio as hold rate here, was 22%. Its own "hold rate," ThruPlays ÷ impressions, had a median of 4.6% (Biddy, 2026). We multiplied the first two: 0.209 × 0.22 = 4.6%. Medians do not always multiply exactly, but here they line up.

Some teams put 15-second plays or 75%-watched plays in the numerator. Those are different metrics from ThruPlays, so do not mix them in one column.

A good hold rate is one that beats the median of your own comparable ads. For reference, the median was 22% on ThruPlays ÷ 3-second plays and 4.6% on ThruPlays ÷ impressions, where the middle half of ads fell between 3.0% and 7.3% (Biddy, 2026).

That is one agency's sample of top-spending ads from 17 accounts, with each ad spending at least $1,000. Treat it as a reference point, not a target; your account will differ.

Three things follow:

• If 22% looks low next to a 40% or 50% band you have seen elsewhere, check the formula, the video length and how many ads sit behind that band before you worry. A band is usable only when it states all three.
• High scores did not mark the big spenders. On ThruPlays ÷ impressions, the ads Meta spent most on did not have higher hook or hold rates than ads that spent less, and ads with higher hook rates did not have a lower cost per purchase (Biddy, 2026). That result was measured on the impressions formula; do not carry it over to ThruPlays ÷ 3-second plays. Our guide to hook signals versus conversion data explains why attention metrics and purchase results need to be read separately.
• The median ad's average watch time was about 4 seconds (Biddy, 2026). Our inference is that most plays end well before 15 seconds, which makes a ThruPlay a high bar.

Our recommendation: set your target from the median of ads in the same length bucket, placement and audience temperature, where cold means people who have not interacted with your brand.

Yes. A ThruPlay is a play to completion or for at least 15 seconds, whichever comes first, so on a video shorter than 15 seconds a ThruPlay is a completion (Meta). Hold rate then measures finishing, not holding.

The same 15-second checkpoint also asks different things at different lengths. On a 45-second video it asks for a third of the runtime. On a 16-second video it asks for almost all of it. This is our inference from the definition.

So put every ad into a length bucket, for example under 15 seconds, 15 to 30 seconds and over 30 seconds, and compare only inside a bucket. A 10-second ad scoring higher than a 40-second ad tells you nothing about which creative is better.

The count depends on your bucket median and the gap you want to catch. From a 22% median, spotting a drop to 16% takes about 669 3-second plays per ad; 18% takes about 1,567; 12% takes about 220. Below the count, treat the gap as a direction, not a result.

This is our model, the standard two-proportion sample size for comparing two ads:

n = (1.96 · √(2 · p̄ · (1 − p̄)) + 0.84 · √(p1 · (1 − p1) + p2 · (1 − p2)))² ÷ (p1 − p2)²

Here p1 is your bucket's median hold rate, p2 is the hold rate you decide in advance is worth catching (the gap is p1 − p2), and p̄ = (p1 + p2) ÷ 2. The 1.96 gives 95% confidence, meaning a 5% chance of calling a gap real when none exists. The 0.84 gives 80% power, meaning an 80% chance of catching a real gap of that size. It assumes viewers are independent. As a check, p1 = 0.25 and p2 = 0.30 gives about 1,250, the figure in our hook rate guide.

Hold rate's denominator is 3-second plays, so count plays, not impressions. The impressions column divides by the sample's median hook rate of 20.9%, rounded to the nearest 10.

For example, 1,567 ÷ 0.209 = 7,497, which rounds to about 7,500.

Fill in this audit sheet from an Ads Manager export of your video ads. Each step has a pass or fail, and an ad only earns a diagnosis if it passes steps 1 to 5.

For step 7, read hook rate and hold rate together. These readings are hypotheses to test, not proof of cause:

Hold rate tells you where viewers leave. It does not tell you what to put there instead, and that is the decision that matters.

At Deepsolv we combine ad-performance data, competitor activity, customer signals and past ad learnings into a ranked weekly test plan and execution-ready creative briefs. The audit sheet's step 7 gives you one candidate test. Our job is to rank candidate tests against each other so your team starts each week with the one most worth running. Our guide to analyzing Facebook ads and choosing your next test goes deeper on that call.

To check an opening before you spend on it, export its first frame as a still image and upload it to our free Attention Heatmap. It shows predicted attention on a frame, not your hook rate or hold rate, and our guide on reading an attention heatmap explains when to trust it.

If you want a ranked plan built from your own account, book a demo with our founding team, and ask us about pricing there.