The Retention Curve Reads As 3 Doubts

The drop is not a mood. It is a question the viewer asked and the video did not answer. The expected-payoff model is the whole mechanism. The title and thumbnail make 1 promise, that promise sets an expected payoff in the viewer's mind, and while the payoff still looks real they keep watching. The moment they doubt it, they leave. The 3 doubts arrive in a fixed order. Is this the video I clicked on, which lands right after the click. Is there anything unique here, which lands during the opening. Is there anything left to watch, which lands in the mid and late drop. This is why a retention number on its own says nothing. 2 videos at 62 per cent are not doing the same job; 1 is answering the 3 doubts on time and the other is losing people at a cliff 40 seconds in. Read the curve and the drop point names the doubt that needs answering. Then answer it in the edit rather than in a comment pinned under the video.

70 per cent average retention is the bar. Below it, the video is engineered higher before anything else is touched. Top creators hold 60 to 80 per cent overall, and the sharper tell sits at 75 to 85 per cent at the 30-second mark. Both Mr Beast and Logan Paul read in that band, which makes the 30-second reading the 1 number worth checking first. Retention and click-through rate are the 2 signals that get a video promoted. Click is the title and the thumbnail; watch is the average view duration and the relative retention. The 2 together are what a platform reads as worth showing to a bigger room. Averages hide the shape, so read the curve as well. A flat 65 that holds to the end beats a 90 that falls off a cliff at 40 seconds, because the second curve tells the platform the promise broke. 1 rule covers the whole check: if the 30-second reading is under 75 per cent, the opening is the problem, not the topic.

This doubt fires first, inside the first 5 seconds, and it is the cheapest 1 to fix. The answer is visual continuity. Open from the thumbnail: the same scene, the same frame, the same subject the viewer just clicked on. On the home screen the viewer has already seen a silent sneak peek of that shot, so continuity is the proof that the click was honest. Title, thumbnail and first 5 seconds have to agree. When they do, the expected payoff is set correctly and the first doubt never forms. 1 more move belongs in the same window. Overlay the keywords on screen for the first 5 seconds, because the sneak peek plays muted and a promise the viewer cannot hear was never made. No greeting, no logo, no warm-up. The opening frame is already the answer.

The second doubt lands during the opening and it asks what this video has that the other 40 in the feed do not. The fix has 2 parts and they take 1 sentence each. State the base concept out loud in the first sentence, tied back to the promise already made, and then show a trailer of the high notes without emptying it. The opener reads like this: I am going to spend the next 7 days buried alive under 20,000 lbs of dirt. Or: you have been hearing about F12, right, so let us talk about it. Or: I am about to spend 30,000 dollars in the next 5 minutes on the most futuristic tech on the internet, and then we test whether it is the future or marketing nonsense. The trailer changes by genre. A challenge shows what is at stake. An explainer shows a glimpse of the insight the whole video was built on. A product review adds 1 relatively unknown feature on top of the review. 1 trap sits here. A trailer that gives the payoff away closes the loop, and a closed loop is a reason to leave. Show the shape of the high note, never the note itself.

A retention curve is not a score. It is a list of 3 doubts, in order.

The third doubt is the mid and late drop, and it is answered with volume rather than with a better sentence. Over-stimulate the first 30 seconds. Across challenge, edutainment and productivity creators the numbers are consistent: 1 shot change every 1.77 seconds on average, about 19 shots inside those first 30 seconds, and fewer than 1 in 5 of them a talking head or base-concept b-roll. Read that as 3 targets. Average shot length sits at 2 seconds or under. Face-on-camera holds under 20 per cent of the opening shots. The barrage is the point, because 1 of the hooks lands even when 6 do not. Then the turnaround. Past 30 seconds the sunk cost starts doing the work for the video; the longer someone has watched, the more resistance there is to quitting. Get them past that line and half the argument is won. Past the line, break sensory monotony on purpose. Change the background, the camera angle or the b-roll, or move a text overlay that reinforces the concept; on the numbers, something should change every 5 to 10 seconds. Hollywood already ran this experiment. The average shot length of English-language films fell from about 12 seconds to 2.5 seconds over 8 years, for attention reasons rather than artistic ones. Then have 1 other person watch the rough cut and mark the 2 or 3 places their focus went. That mark is the interrupt, or the trim.

1 storyline carries about 3 minutes. Past that the curve needs a subplot running alongside the main story, planned at concept stage rather than patched in the edit. The reference pair is well known. A heist video runs a main line - an FBI agent has to steal 100,000 dollars by midnight - and a subplot of a giant telephone pole blocking the exit while a teammate pops the tyres. An AI explainer runs a main line on what the image model is, with a race between a human-drawn skyline and the machine as the subplot. The subplot is not a second topic. It is the same promise, still being paid, from a second angle. So the working checklist is 7 lines long: 1 clear promise on the title and thumbnail; the first 5 seconds opening from the thumbnail with the keywords overlaid; the base concept stated in the first sentence; 19 shots in the first 30 seconds at 2 seconds or under with under 20 per cent of them a talking head; a change every 5 to 10 seconds after that; a subplot decided at concept stage for anything past 3 minutes; then a rough cut and 1 other pair of eyes on it. The same discipline built the physical things first. Companies built across 12 countries. 210 energy systems across Africa and Asia. 200 wood-gasification machines across the UK and Europe. 16 years of field work from rural Nigeria to Fukushima, and not 1 of those projects measured its own curve by the average. The probability that a strong middle rescues a weak opening is not 0. It is just low enough that the middle never gets watched. Read the curve, find the drop, name the doubt, fix the edit. That order never reverses.