Hook Rate, Retention, Watch Time: Reading Short-Video KPIs

Most social media reports lead with views, likes and follower growth. None of those numbers explains why one video got distributed and the next one stalled. The metrics that steer sit one level below: how many people stay after the video starts, how long they stay, how many watch to the end and how many pass the video on.

This is about exactly that level: which metrics you can influence per video through editing, structure and length, what the platforms actually report, and when a difference between two videos is solid enough to act on. The formats these metrics refer to are placed in the pillar Social Media Content Creation & Formats. How platforms turn these signals into reach belongs to Social Media Algorithms & Distribution. Cross-channel attribution and proxy KPIs are covered in Social Media Analytics, KPIs & Measurement.

Hook rate, hold rate, completion: what the four core metrics measure

The metric family for short video is manageable: Watch Time, Average View Duration, retention or completion rate, hook rate, hold rate, saves, shares, sends per reach, profile visits and follows per reach. Four of them steer production, the rest describe the result.

Hook rate: the share of impressions where someone stays beyond the first few seconds. The term comes from the advertising context, where hook rate and hold rate are reported as creative metrics. Organically you usually calculate it yourself or read the opening section of the retention curve as an approximation.

Hold rate: usually calculated as the share of those who keep watching after the hook, based on the group that survived the hook, not on all impressions. It separates a video with a strong opening and an empty middle from one that carries all the way through.

Completion rate and retention: the share that watches the video to the end, and the path there. The curve matters more than the final value because it shows the drop-off point.

Watch Time and Average View Duration: the total viewing time generated and the average viewing duration per person. Both depend on length, which is why they only carry meaning within comparable formats.

Sends per reach and saves are strictly speaking not video KPIs, but they belong in the same evaluation. They answer whether the content was worth enough to someone to forward it or save it. A video with a mid-range completion rate and a high send rate is usually worth more than the reverse case.

Why the first few seconds decide the rest

Meta measured back in 2016 that people spend an average of 1.7 seconds with a single piece of content on mobile devices and 2.5 seconds on desktop. The figure is now readily quoted as a stay-or-scroll decision made in 1.7 seconds. That is interpretation: what was measured is average dwell time, not a decision point. For production it changes little. Waste the opening shot and you lose distribution before the actual content begins.

As a rule of thumb: the first three seconds set the topic, the benefit or the tension. Whatever is good after that, most people never see. Felix Beilharz put the shift in a sentence in the DIM podcast on 29 July 2025: “Creative is the new targeting.” Delivery runs automated, the difference comes from the material.

Part of what circulates as hook craft (pattern interrupts, fixed storytelling formulas, a ban on greetings in the first sentence) is proven in practice but not properly evidenced. Treat it as a hypothesis for your channel, not as law. The difference is practical: anyone who adopts hook rules as settled never tests them and does not notice when they fail to work in their niche.

What the platforms actually count

The definitions are not compatible, and that is the most common source of false conclusions.

Instagram: According to the Instagram Help Center, Watch Time counts the total play time of a reel including the time spent on replays, while views count every start and every restart. A 3-second view as a denominator does not exist there. Your hook rate on Instagram is therefore a derived figure, not a reported one, and Watch Time rewards videos people watch twice.

TikTok: In its own newsroom, TikTok has described since 2020 that a strong indicator of interest, such as a longer video that someone watches from beginning to end, is weighted more heavily than a weak signal. TikTok names shares and likes as interaction signals without explicitly describing them as heavily weighted. Completion rate is therefore the steering metric with the best evidence base on TikTok.

YouTube: Average View Duration is the retention metric YouTube reports in Analytics, defined as the average minutes watched among those who watched at all. That Average View Duration and session watch time dominate ranking is widespread industry consensus from tool and creator analyses, not a platform statement. Mark it as industry consensus in your own reporting, not as a platform statement.

One hard rule follows from this: never compare these metrics across platforms. A completion rate on TikTok and one on Instagram have different denominators. Which formats, lengths and specs sit behind the respective feeds is covered in Instagram Reels format vs. Shorts vs. TikTok.

Testing hooks without fooling yourself

With organic social there is no stable traffic split. The platform decides who gets shown which video, and delivery depends on time of day, topic affinity and chance. Statistical significance in the classic sense is out of reach. Anyone who still argues with confidence intervals is selling noise as insight.

What remains is sequential testing of large effect sizes. That works if you stick to a few rules:

One variable per test series: either you test the first three seconds or the topic or the length. Change two things at once and the result cannot be attributed.

Several videos per variant: a single video proves nothing. Run each variant across a series and compare the series, not the individual values.

Median instead of average: one viral outlier pulls every mean upwards and makes the analysis worthless. The median of your videos per format is the more stable reference.

Only take large differences seriously: small deviations between two series cannot be interpreted under organic distribution. If a hook variant is not clearly ahead, the result counts as open.

Keep conditions constant: same publishing windows, same cover logic, same subtitle treatment. Otherwise you are testing the posting time along with it.

Document every test series with hypothesis, variant and median. Without a record you will run the same test again in six months.

Reading retention curves: three patterns

The curve tells you which part of the production you have to touch.

Steep drop in the first seconds: an entry problem. Change the opening shot, the first sentence and the text overlay, not the topic. If several videos on the same topic show the same drop, it is the topic after all.

Clean start, drop-off in the middle: the promise from the hook is delivered too late. Cut it down, pull the payoff forward, drop the intermediate steps.

Flat curve with a rise at the end: the video is being rewatched, usually with dense information or a clean loop. That is the state you aim for on Instagram, because Watch Time counts replays there.

Benchmarks: international, not DACH

Published comparison figures help with orientation but do not work as a target. As of August 2026, Socialinsider measures an engagement rate by followers of 3.70% on TikTok, 0.48% on Instagram and 0.15% on Facebook in an analysis of 70 million posts from international brand accounts (January 2024 to December 2025). Quid arrives at 0.30% for Instagram. Neither figure is wrong, they work with different denominators. Writing such numbers into a report as a target without supplying the methodology produces arguments instead of steering.

Two developments from the same analysis support the shift away from likes: shares per post rose by 45% on TikTok and by 12% on Instagram year on year, while comments declined (TikTok by 24%, Instagram by 16%). Organic reach rate fell by 18% to 3.50% according to Socialinsider. All of these figures are international, and DACH-specific engagement benchmarks are largely missing. Do not transfer them quietly to the Austrian or German market.

The usable reference is the median of your own channel, separated by format and length, across a rolling time window.

What belongs in weekly reporting

Run sends per reach, saves, completion rate and Watch Time as primary steering metrics. Likes and follower growth stay in the report, but explicitly declared as secondary. That declaration is not a formality: it decides what gets discussed in the weekly meeting and which videos count as the model for the next production.

The same split between steering and reporting metrics applies in search, and the logic behind it is described in SEO KPIs: The Most Important Key Performance Indicators.

Typical mistakes in this routine:

Judging single videos: the spread of organic reach is too large. What gets judged is the series.

Comparing Watch Time across formats: a longer video generates more viewing time and is not better for it.

Reading views as success: on Instagram every start and every restart counts, which limits the value for comparisons.

Raising frequency when the numbers fall: more weak material improves no median.

Adopting hook rules from other channels: what works in a consumer niche often does not carry in a B2B channel.

The decision rule

If the organic reach of a format stays below your own format median despite consistent hook optimisation, the problem no longer sits in the first three seconds. Then you change format or topic. Not frequency.