The Flavor Deck
Every shape each engine can take, why it works, and how much evidence it has.
The one line
An engine is a force in the reader's head. A
flavor is the shape you pour it into.
Engines diagnose; flavors produce. Run all four engines at once;
run one flavor per engine, cleanly.
research A named study tests something close.
inferred Mechanism is established; the hook use isn't tested.
craft Observed practice and reasoning only.
Two laws over every gap flavor
1 · A gap needs ground
You can't be curious about something you know nothing about
(Loewenstein 1994). Land every flavor inside a world they half-know.
2 · Small gaps beat big ones: measured
Kang et al. (2009), 40 trivia questions: curiosity ran an upside-down U and
peaked when confidence of knowing was ≈ 0.50.
Give them half the answer.
Gap flavors: what feels missing
1 · Curiosity gapresearchMissing what.
Why. In Kang et al., reward-anticipation circuitry fired while reading the question. Wanting to know is processed like wanting a snack. High curiosity also improved recall 1-2 weeks later.
Fails if vague, zero confidence is the wrong end of the U.
2 · Utility gapresearchMissing how.
Why. Runs on “I'll need this,” not “I must know.” Berger & Milkman: practical usefulness predicted sharing. Limit: that measured sharing, not stopping.
Fails if they know it, or don't want the outcome.
3 · Surprising-origin gapinferredMissing the link.
Why. Loewenstein: a violated expectation causes arousal and starts a search for an interpretation. Makes gap and arousal at once. The search must be able to succeed, that's the bridge.
Fails if random. Aborted search = confusion.
4 · Emotional-contradiction gapinferredMissing how two feelings both fit.
Why. Same violation machinery, aimed at the script that says which events produce which feelings. Advantage: a contradiction inside your head can't be closed anywhere else.
Fails if manufactured, no real resolution waiting.
5 · Process gapinferredMissing the road. Show the destination, hide the route.
Why. (a) the end state hands them the knowledge Law 1 needs; (b) goal gradient (Kivetz 2006): café customers bought faster near a free coffee, and even illusionary progress worked (12-stamp card, 2 pre-filled, beat a blank 10).
Fails if nobody wants the end state, or the road never shows, that's a brag.
Arousal flavors: the ladder
Awe · Anger · Anxietyresearch
All three positively linked to sharing in Berger & Milkman (2012). Awe stretches the model and costs no trust. Anger is approach-motivated, but recruits people who disagree. Anxiety narrows attention onto the threat, and your audience already runs high on it.
Excitement / hopeinferred
Approach motivation toward a possible future self. Pairs with the process gap.
Recognitioninferred
Our build, not a named emotion in the research: self-reference effect (Rogers et al. 1977) + a privacy-violation jolt. You thought it was only yours.
Validationresearch“Finally someone said it.”
Hart et al. (2009) meta-analysis: preference for information that agrees with you, d = 0.36. Moderate, not overwhelming. The bias flips when disagreeable info is useful for a goal. So validation loses to usefulness.
Fails if the belief was already said out loud everywhere. It must have been unspoken.
Across all six: the heat must be one you actually feel.
Arousal is a body state, and you can't fake a body state in five words.
Open loop flavors: holding the tension
Softer ground than the gap flavors. Parent is Zeigarnik (1927), and the memory half largely failed to replicate, what survives is the pull to resume. Working model, not settled science.
1 · Withheld nouninferred
Name the category, hide the content, that gives the gap a shape they can hold. “Something happened” gives them no slot.
Fails if held too long. Stale debt gets written off.
2 · Numbered promiseinferred
Goal gradient: a visible finish line, and effort rises as it nears. Say the number early. They can't watch the distance shrink until they know how long it is.
Fails if item three is padding.
3 · The turninferred
Reward prediction error used mid-piece. A confirmed prediction makes almost no dopamine; a violated one spikes. A real reversal re-fires the signal your hook fired.
Fails if the reversal has no consequence.
4 · Unit cliffhangercraft
Format logic. A carousel makes them re-decide at every swipe, so tension must be renewed at each decision point, not created once.
Fails if nothing pays per unit.
5 · Loop stackingcraft
The scaling trick: a new debt starts before the old one settles, so the pull never drops to zero.
Fails if you stack more than you can pay.
Maybe flavors: honest uncertainty
The physical shape of a maybe
Fiorillo, Tobler & Schultz (2003) found two dopamine
signals: the quick prediction-error burst, and a sustained climb
that builds through the whole wait and tracks uncertainty,
maximal at P = 0.5, shrinking toward certainty either way.
Not a spike at the end. A rising line through the middle.
1 · Honest uncertaintyresearch
Two engines at once: the ramp keeps running, and admitting a limit is a pratfall that humanises you. Catch: the pratfall only helped the competent contestant. Earn it first, then admit it.
Fails if you're actually sure.
2 · The betresearch
The 50/50 finding, used literally. A bet you might really lose sits where the climb is highest; one you've already won sits near certainty, where it flattens. (Caveat: monkeys, juice.)
Fails if you can't lose, that's a brag with a countdown.
3 · The skeptic's turncraft
Puts you on their side of the table. You make their objection for them, so you're testing with them, not selling to them.
Fails if the verdict is obvious up front, then the odds aren't 50/50 and the climb never starts.
Bridge flavors: credibility (H.04)
Parent finding, and its expiry date
Hovland & Weiss (1951): identical message, credible source, more opinion
change. But the sleeper effect. Weeks later the
advantage faded as people forgot who said it. So credibility buys
the first moment, which is exactly what a hook needs and no more.
The receiptcraft
A precise number is expensive to fake and easy to check, so it proves you counted. “So much” costs nothing to say.
The tenureinferred
The expertise half of source credibility: a reason to believe you'd know.
The scarresearch
Pratfall effect (Aronson et al. 1966): a contestant who'd answered 92% correctly became more likeable after spilling coffee. But the mediocre one (30%) became less likeable. The scar only builds trust if competence is already visible, a brand built on vulnerability still has to show the competence somewhere.
The witnesscraft
Moves the claim outside you, so it stops being a thing you want and becomes a thing that happened.
Credibility = trust in the source. The Maybe = honesty
about the outcome. Use both, or credibility slides into overclaim.
The 7 field strategies, mapped
H.01 Curiosity & Open LoopsGap + Open Loop
H.02 Relatability & ContrastSelf-relevance + recognition arousal
H.03 Transformation & PayoffProcess gap + hope
H.04 CredibilityThe bridge (not a pull)
H.05 Secrets & ShortcutsUtility gap, dressed as access
H.06 Opinions & PolarizationAnger + validation arousal
H.07 Identity & SpecificitySelf-relevance, exactly
Two cautions on that list
1. Its counts are popularity, not performance.
2. It can't see the Maybe, because the Maybe has
no phrase tell.
Diagnose with the engines. Unstick with the seven.
Diagnose with the engines. Unstick with the seven.
How much to trust all this: enough to build with, not enough to argue
from. These studies used trivia questions, coffee cards, 1966 quiz tapes,
monkeys, and 2008 newspaper articles. None studied someone scrolling at 1am.
The machinery travels; the sizes and rankings don't. Full citations in
Sources.