Zavis Video OS
← All skills
Skill

zavis-engagement-engineer

Engineer videos for retention, not just to exist. Load this skill whenever you need to ensure a video EARNS its viewer second-by-second. The intent is never just "make a video" — it's "make a video people watch through to the end and share." This skill is the strategic mind behind every pacing, structure, and content decision. Pairs with zavis-hook-engineer (for the opening) and zavis-taste-dna (for the look).

skills/zavis-engagement-engineer/SKILL.md

Zavis Engagement Engineer

You are responsible for the single hardest job in content: making someone who didn't ask to watch your video keep watching anyway.

The brutal truth about retention

  • 45% of viewers drop in the first 3 seconds
  • 65% of those who survive the first 3 seconds drop by 15 seconds
  • The remaining viewers drop steadily until ~70% are gone by the end
  • Your job is to identify EVERY moment where someone is about to drop, and put a reason to stay there

The retention curve mental model

For every 5-second slice of a video, ask: why does the viewer keep watching here?

If the answer is "because they have to," you've already lost. The answer must be one of:

  • Curiosity loop is still open — a question was asked, the answer hasn't come yet
  • Pattern hasn't completed — they're tracking a sequence (3 of 5 founders shown, etc.)
  • Escalation in progress — each beat is more interesting than the last; they're waiting to see how far it goes
  • Setup for payoff — they sense a punchline coming
  • Visual surprise — what they're seeing is something they've never seen
  • Voice / sound is doing work — the audio is too good to skip

If none of those apply at a moment, rewrite that moment.

The retention budget

Every second has a retention cost. You spend retention to do things; you earn retention by surprising people.

| Action | Costs retention | Earns retention | |---|---|---| | Hook with unexpected visual | | +++ | | Slow buildup with no payoff promise | --- | | | Fast pace with no anchor | -- | | | Specific number / unfamiliar fact | | ++ | | Generic statement | -- | | | Surprising connection between two ideas | | +++ | | Saying what's already on screen | --- | | | Hard cut on a beat | | + | | Long static shot with talking | --- | | | A turn / reveal | | ++++ | | The viewer learning something they wanted to know | | ++ | | Bullet list of features | ----- | | | A specific person's name | | + | | "Here's what you need to know" | -- | | | A question the viewer is now thinking | | ++ |

Net positive every 5 seconds, or rewrite.

The 3-second retention check

Every 3 seconds in your script, ask:

  1. Did something change?
  2. Is curiosity still high?
  3. Is there a reason to stay another 3 seconds?

If you have 3 consecutive 3-second blocks where nothing changed, that's a 9-second snore. Cut.

Hooks vs anti-hooks

A hook is anything that makes someone stay. An anti-hook is anything that gives them permission to leave.

Anti-hooks (forbidden in the first 5 seconds):

  • Logo animations
  • "Welcome to" / "Today we'll talk about"
  • Slow cinematic music fade-in
  • Generic statements ("AI is changing everything")
  • The narrator clearing their throat
  • Two seconds of nothing

Hooks (use one in the first 3 seconds):

  • A specific number ("In 5 days, ChatGPT had a million users")
  • A bold contrarian claim ("The robots aren't coming. They're already here.")
  • A visual you've never seen ("[archival photo of a 1956 conference room with handwritten algorithms on a chalkboard]")
  • A direct question whose answer is the video ("What if the entire AI revolution started with eight pages?")
  • A pattern interrupt: 3 quick visual cuts of unrelated things that resolve to one theme

The escalation rule

The middle of the video must escalate. If beat 6 is less interesting than beat 5, you've lost retention. Each beat must do at least ONE of:

  • Raise the stakes (more money, more people, more time)
  • Get more specific (general → personal)
  • Reveal something hidden
  • Connect to something the viewer already knows
  • Show consequence

If none of those apply, the beat is filler. Cut it.

The turn

Every video over 30 seconds needs ONE turn — a moment where the viewer's mental model shifts. This is the moment they decide to share it.

The turn is usually:

  • A reveal ("...and the founder was 19 years old")
  • A flip ("It turns out the opposite is true")
  • A connection to something familiar ("This is what happened in 2008")
  • A confession ("We tried this. It failed. Here's why.")

Without a turn, the video is information delivery. With a turn, it's a story. Information delivery doesn't get shared.

The landing

The last 5-10 seconds of the video are THE most important seconds for sharing. The hook earns the watch. The landing earns the share.

The landing must:

  • Pay off the curiosity loop the hook opened
  • Use the shortest words in the entire script
  • Land on a noun, not a verb
  • Imply more than it says
  • Connect back to the opening (loop closure)

Bad landing: "Follow Zavis for more videos like this!" Good landing: "We're not at the end of this story. We're barely past the beginning."

The integrated CTA

A CTA is NOT "go to zavis.ai." That's a banner. A CTA is the natural extension of the video's argument that puts Zavis inside the story, not outside.

If the video is about "the evolution of AI," the CTA is NOT "use Zavis." The CTA is something like:

"What happens next is being written right now. By people who build with AI. Zavis is one of them."

The CTA is the narrator's voice saying "and that's why we exist" without saying those words.

How to design CTAs that match the video's argument

  1. State the video's central claim in one sentence
  2. Identify what action that claim implies a viewer might take
  3. Identify what role Zavis plays in that action
  4. Phrase the CTA as the next sentence after the climax — not as a sales pitch

Example for "Evolution of AI":

  • Claim: "AI evolved from theory to infrastructure in 70 years."
  • Implied action: "Build with it before it builds without you."
  • Zavis's role: "We help businesses use AI to engage their customers."
  • CTA: "The technology is here. The question is what you'll build with it. — Zavis."

Engagement-first script audit

When reviewing or writing a script, run this audit:

| Check | Pass condition | |---|---| | First 3 seconds | At least one hook pattern present | | First 10 seconds | Curiosity loop opened | | Every 5-second block | Net positive on retention budget | | Middle | Escalation present | | ~70% mark | One clear turn | | Final 10 seconds | Landing rhymes with hook | | CTA | Integrated, not appended | | Total length | No filler — could you cut 10% and lose nothing? |

If any check fails, rewrite. Don't compromise.

On narration speed and silences

Silence is engagement. Long static narration is not.

  • Periods over commas (gives the narrator a beat)
  • Pauses after key words (let them land)
  • Short sentences (most under 9 words)
  • Front-load the surprise
  • One idea per sentence

On visuals serving narration

Every visual in a video should be doing one of three things:

  1. Showing what the narrator can't say (a face, a graph, a photo)
  2. Reinforcing what the narrator IS saying (a key word lights up)
  3. Counterpointing the narrator (irony, juxtaposition)

If a visual is doing none of these, it's wallpaper. Cut it.

What to do when you're not sure

If you're not sure whether a beat earns its retention cost, cut it. A 60-second video that earns every second is infinitely better than a 90-second video with 30 seconds of filler.

Zavis videos are short for a reason. The reason is that retention is the only metric that matters.