What average watch time measures
Meta divides the total time people spent watching a video ad by the number of times that video started. The result is a single figure in seconds describing how long a typical view lasted. It is reported per ad, per placement and per audience, so one piece of creative can produce very different numbers in Reels than it does in the Facebook feed.
Two things make the figure lower than people expect. Videos autoplay, so a play is counted whether or not anyone chose to watch, and every scroll past sits in the denominator. And the metric counts elapsed seconds, not attention — a video left running in a muted feed while someone reads a message still accumulates time.
Why average watch time matters
It is the plainest read on whether a video holds people once it starts. Click-through rate tells you who acted at the end; average watch time tells you how far into the message they got before deciding. When a video runs long and the average sits close to the opening, the problem is the first few seconds, not the offer.
It also settles arguments about length. If viewers consistently stop well before the end, cutting the video down to the part they actually watch costs nothing and usually lowers the cost of every result that follows. On mobile-heavy markets like Nepal, where most impressions are served on a phone in a fast-moving feed, that shortening tends to help rather than hurt.
Common mistakes with average watch time
The most common error is comparing it across videos of different lengths. A short clip watched right through can report fewer seconds than a long video abandoned early, so the long video looks stronger while performing worse. Compare the same length against the same length, or use hook rate and hold rate, which are ratios and survive the comparison.
The second is treating it as the objective. Watch time is a diagnostic, not a result. A video that holds attention beautifully and sells nothing is still a failed ad, and I have seen advertisers protect a long-watched creative for months while cost per lead drifted upward.
The third is mixing placements in one number. Feed, Reels and Stories behave differently, and an average across all of them describes no real viewing experience.
How to act on it
Split the report by placement first, then look at where the drop happens. If watch time collapses almost immediately, rebuild the opening: lead with the problem or the product on screen rather than a logo or a slow establishing shot. If it holds and then falls away in the middle, the video is padded and can be trimmed.
Once the shape of the drop is clear, test one change at a time — a new opening, a shorter cut, captions burned in for sound-off viewing — and judge each version on cost per result as well as on seconds watched. The practical playbook for building those variations sits with Meta video ad campaigns, but the reading of the metric is the same whatever the objective.
