The methods behind each tool

The Angle Engine

Most idea generators fail the same way: they produce five costumes on one idea. The output reads fine and impresses nobody. The fix is not more ideas — it is a set of deliberate moves that each produce a structurally different concept rather than a synonym. Two rules sit above all of them: packaging is the idea, so an idea you cannot picture a clickable package for gets killed at the idea stage rather than after filming; and no idea may repeat a title the channel has already published.

• Person / POVTell it through a specific someone. People outperform abstractions — a face and a name carry the click.
• Time / constraintAdd a clock or a hard limit. A constraint is a built-in curiosity loop: will they pull it off?
• Stakes / consequenceName what is lost by getting it wrong. Loss is one of the three core click emotions.
• Contrast / comparisonPit two things against each other. Contrast is the engine of hooks.
• Curiosity gapWithhold the payoff. Open the question; do not answer it in the idea.
• Scale / extremityBiggest, smallest, first, worst. Extremes force a glance.
• Insider / revealPull back a curtain. Insider knowledge is social credit for sharing.
• ContrarianState the common belief, then deny it. The strongest single packaging device.
• TransformationShow the journey. The before is relatable, the after aspirational, and the gap is the hook.
• Format transplantBorrow a structure proven in another niche. Proven elsewhere, novel here.

Seven opening mechanisms, labelled on every line

The hook is the highest-leverage part of a video: however good the body is, none of it is watched if the viewer leaves in the first five seconds. Rather than generating lines that merely sound good, the hook writer labels every line it returns with the mechanism it used — one of seven — so you can judge the mechanism rather than the vibe, and deliberately test a different one next time.

• Open loopStart something the viewer needs the rest of the video to finish.
• ContradictionState the opposite of what the audience expects, then owe them the reason.
• StakesName what is lost by getting this wrong, before explaining anything.
• DemonstrationShow the result first. The claim is the thing on screen, not a sentence about it.
• Direct promiseSay plainly what the viewer will be able to do by the end.
• SpecificityLead with the one concrete number or detail that proves this is not generic.
• QuestionAsk the thing the viewer already half-wonders, so the video becomes the answer.

Traffic source, then emotion, then model an outlier

Five to ten words decide whether anything else gets seen, and clicking is a split-second emotional decision rather than a logical one. The method runs in a fixed order: establish where the video will be found, choose the emotion that suits it, then model the structure of something that already worked instead of inventing from a blank page.

• 1. Pick the traffic sourceBrowse and search are different jobs. A browsing viewer is scrolling past and must be interrupted; a searching viewer is hunting and must be reassured they have found the right thing. Negative angles tend to win in browse, clear positive outcomes in search.
• 2. Name the emotionCuriosity, desire or fear — and curiosity generally works as the top layer over one of the other two rather than on its own. If the title makes nobody feel anything, it gets rewritten.
• 3. Model, never inventBreak a proven title down to its underlying structure — the mechanism, not the subject — and apply that structure to your topic. The generator will not claim to have modelled a specific video it was not given, because it has no swipe file unless you supply one.

The Modern Story Arc

The traditional arc — slow rising action, climax, resolution — assumes a captive audience, and a scrolling viewer is not one. The script writer starts in the middle of something already in motion at roughly seventy per cent context, spikes hard within the first ten seconds, then stacks peaks and releases while re-earning attention every ten to twenty seconds.

• Click confirmationThe first line confirms the promise the title and thumbnail made. A mismatch here is an instant bounce, and it is the most common structural fault in a script that tests well and retains badly.
• The five-step introHook, confirm the promise, establish stakes, preview the payoff, open the loop that keeps them watching.
• Rolling peaks in the bodyPeak, release, repeat, rehooking constantly. Open loops planted early and paid off later.
• A resolved endingClose the loop, deliver the payoff, and carry the call to action inside the content rather than bolting it on.
• Shorts are a different shapeA Short is not a shorter long-form video. No intro, no outro, no rehooking cadence — one continuous idea whose first sentence carries the whole tension, because there is no title or thumbnail doing that work.

Psychology first, design second

A thumbnail is mostly the psychology of the click and only a little the craft of the image, which is why it should be designed before the video is made rather than in the hour before upload. The viewer reads packaging in one to two seconds in a fixed order, and the method is built around that order rather than around visual taste.

• The three-step readThe image stops the scroll; the eye drops to the title to hunt for value; then it returns to the image to check that it supports the title's promise. All three have to hold — a thumbnail that pops but does not validate the title loses the click at the last step.
• At most three attention elementsThere are seven that work — colour contrast, a face carrying a readable emotion, a compelling graphic, large text or numbers, arrows or circles that aim the eye, aesthetic imagery, and a collage arrangement. Use no more than three. Thumbnails render small, and a fourth element makes the other three unreadable.
• Composition that survives the interfacePick one layout — centred, offset on a third, or split for before-and-after. Keep anything that matters out of the bottom-right corner, where the timestamp sits. Most viewers are in dark mode, so check the image against a dark interface, not a white canvas.
• Emotion has to matchThe facial emotion must match the video's emotion; a default smile on a negative video kills it. And an unfamiliar face can cost more than no face at all.

One pass over the finished video

Descriptions, chapters, tags, hashtags, the pinned comment and the community post are each a small method rather than a filled-in box, and each runs its own skill. Bench Optimize runs them together over one video so the set is coherent — the chapters match the script, the description opens on the same promise as the title.

• DescriptionFront-load the promise the title made; the first two lines are all most viewers see before the fold.
• ChaptersDerived from the actual script or transcript, not invented, so timestamps land where the content changes.
• Tags and hashtagsModest, relevant sets. These are a minor ranking signal and are treated as one rather than stuffed.
• Pinned comment and community postWritten to open a conversation rather than to repeat the description.