Chapter 2. Audience Is Not an Age Bracket — It Is Desire and Viewing Context#

2.1 Age, gender and city only describe the entrance#

"Target audience: women 25–44, lower-tier cities" looks like a definite answer. In practice it has completed a small part of the media-buying job and nothing else. Two women of the same age may watch the same revenge microdrama for entirely different reasons. One wants to watch someone who was underestimated prove her competence. The other wants to watch a traitor lose face in public. Both might describe themselves as liking stories about strong women — but the first cares whether the professional process is credible, and the second cares whether the punishment is satisfying.

Demographics help you find a person. They do not help you write the next second she is willing to watch. What creative work needs to study is not what people like this usually enjoy, but what emotional task the viewer is completing through this show, in a particular viewing situation.

That task might be compensation: she cannot fight back in life, so the story does it for her. It might be confirmation: she needs to see that her own pain is not self-indulgence. It might be rehearsal: she wants to experience a dangerous relationship safely. Or it might be the restoration of order: the guilty must be identified, the evidence must be seen, and a public humiliation must be repaid in public.

The final product of audience research is not an attractive persona portrait. It is a set of hypotheses capable of changing a story decision. If the research cannot tell you which opening to use, which passage of explanation to cut, or which piece of proof to strengthen, it is still decoration.

Figure 2-1 The same "segment" pursues different desires in different viewing situations, accepts different story promises, and produces different behavior — continuing, paying, or watching an ad. Age and gender are only the entrance. Context, desire and promise are the actual creative chain.

Write a falsifiable hypothesis for each path through this diagram. For example: viewers watching alone at night are more willing to follow a restrained investigation, while fragmented commuting sessions need an early public reversal. Campaign results can only validate a specific path; they cannot be generalized into "all women like this genre."

2.2 The desire stack: nobody wants only one thing#

Viewers rarely sustain sixty episodes on a single desire. A practical method is to build a desire stack with four layers.

The first is immediate stimulus: what happens in the first three seconds that is enough to stop the scroll. The second is emotional process: how much oppression the viewer will tolerate, in exchange for what release. The third is identity fantasy: who she would like to be, temporarily. The fourth is moral order: what this world should ultimately prove is right.

The desire stack for Backlit Takeover is not a vague preference for a strong female lead:

audience_desire_stack:
  immediate_stimulus:
    - a woman fired in front of everyone produces acquisition authority
    - every judgment in the room about who she is collapses at once
  emotional_process:
    - hold the anger down
    - find verifiable evidence
    - make it impossible for the aggressor to keep rewriting the facts
  identity_fantasy:
    - keep your judgment intact while a group is denying you
    - take power back through professional skill, not a lucky protector
  moral_order:
    - power can control the narrative for a while, but it cannot erase the record
    - humiliation created in public must be corrected in public

The stack constrains the writing. If the script later has Shen Yan repeatedly producing evidence on Lin Yun's behalf, the show still delivers a kind of wealthy-protector comfort — but it breaks the identity fantasy of taking power back through competence. The audience may not articulate the reason. They will feel the protagonist turn from an agent into a recipient of rewards.

2.3 A pain point is not a list of misfortunes#

Beginners tend to read audience pain as a checklist — unemployment, betrayal in marriage, a damaging family, financial pressure — and then load all of it into the protagonist's backstory. The problem is that more suffering does not equal more resonance. Resonance comes from stating the mechanism by which the suffering happens.

"Suppressed by her boss" is too broad. "She discovers the project data is falsified, and her boss gets there first and defines her as emotionally unstable and not a team player" is far more specific. What the audience recognizes is not a label but a familiar technique of power: first strip the speaker of credibility, then avoid the fact.

When researching a pain point, break it into at least five parts: what happened; who holds the right to interpret it; why the victim cannot retaliate immediately; which detail is the most enraging; and what correction would actually satisfy. Only then does a pain point become a scene rather than a sad item in a character biography.

pain_mechanism:
  event: Lin Yun files a lead on an anomalous payment
  power_holder: Zhou Lan controls the board and the internal investigation
  silencing_method: reframe a professional objection as revenge against the family
  humiliating_detail: the resignation statement was drafted before she arrived
  desired_correction: retake the right to interpret, in the same meeting, with an undeniable audit record

2.4 Viewing context changes the shape of the work#

The same person is not the same viewer on the commute, at lunch, before sleep, and casting to a television. Her attention, her sound environment, the length she can accept and her emotional need all change.

Commuting often means muted playback, so the first shot and the subtitles must carry the conflict on their own. Watching with headphones before sleep lets narration and music carry more intimacy. Scrolling episodes back to back in fragments demands an extremely short recap. Watching a compiled batch in one sitting requires cutting repeated information and excessive fades between episodes.

A viewing context sheet should record at least: device, orientation, whether sound is on, usable session length, probability of interruption, entry source, prior knowledge, and current mood. These answers belong in shot, sound and edit decisions — not only in a marketing document.

viewing_context_hypothesis:
  device: mobile_phone
  orientation: vertical
  entry: paid_social_creative
  sound: often_low_or_muted_at_entry
  session: 3_to_8_minutes
  interruption_risk: high
  prior_knowledge: none
  production_implications:
    - the first shot must be legible without dialogue
    - do not introduce too many names and relationships at once
    - key evidence needs a visual close-up and a subtitle that explains it
    - music must not be the only carrier of an emotional turn

2.5 Where audience evidence comes from#

Audience research can draw on four kinds of material. First, behavioral data: completion, drop-off, return rate, payment, creative click-through. Second, natural language: comments, search terms, Q&A, complaints, titles of derivative works. Third, content supply: charts, trending subjects, ad creatives, platform recommendations. Fourth, direct research: interviews, accompanied viewing, concept tests.

All four are biased. Clicks tell you what attracted someone, not whether she was satisfied. Comments come from people willing to speak and cannot represent the silent majority. Charts tell you what has already succeeded, not whether there is still room to copy it. In interviews people rationalize their own choices, and may behave completely differently while actually scrolling.

So you cannot declare "users just love this" from one hit or a few dozen upvoted comments. The safer method is to build a chain of evidence: several works exhibit the same emotional mechanism; comments show how viewers describe that satisfaction; and a small test shows that the corresponding hook changes behavior.

Code the comments; do not collect quotable lines#

When collecting comments, tag each one with a function: agreement, anger, prediction, taking sides, demanding the next episode, questioning the logic, complaining about stalling, discussing the cast or the visuals, asking about the schedule, expressing willingness to pay. Then tag the plot point and timecode it refers to.

A hundred comments saying "so satisfying" is worth little. Twenty comments saying someone finally put the chat log on the big screen tells you the satisfaction comes from making proof public — not merely from the antagonist losing.

comment_id,episode,timecode,function,target,signal
C018,E003,00:48,demand_next,truth_about_access,viewers care who deleted the ledger
C031,E003,00:52,logic_question,system_permissions,the rules of evidence were not made clear
C044,E003,00:55,taking_sides,Lin Yun,the professional counterattack earns agreement

2.6 Do not ask interviewees what they like#

A useful interview is built around the most recent real behavior, not an invitation to invent an idealized self. Do not ask whether she likes stories about strong women. Ask: what was the last microdrama you watched ten episodes of in a row? When were you planning to stop? Which moment made you keep going? Which part did you skip? Was there a moment when you felt the show was cheating you?

Concept tests should also not open with one complete, beautiful proposal. Test interest with a one-sentence premise first. Then test three different opening shots for what stops the scroll. Then test comprehension and return intent with a one-minute rough cut. Showing highly finished images too early makes people report their liking for the art direction as liking for the story.

Watching how someone retells it matters most. If, after the rough cut of Backlit Takeover, she says "a real heiress comes back for revenge," while our core promise is "an auditor retakes the right to interpret using professional evidence," then the first episode's identity trope has buried the professional differentiator.

2.7 Why character-style personas so often fail#

"Sister Wang, 36, second-tier city, one child, enjoys yoga and shopping" cannot guide an opening. A persona is only useful when it contains context, trigger, desire, aversion, trust condition, and reason for stopping.

persona_id: AUD_REVENGE_PROOF_01
context: alone with her phone in the evening, usually entering muted
trigger: someone with power publicly distorts the facts
fantasy: reverse status with a record the other side cannot deny
accepts:
  - brief humiliation early on
  - professional terminology when it is visually explained
rejects:
  - a protagonist who does not fight back for a long time
  - a male lead who unconditionally solves things for her
  - the same misunderstanding stretched over several episodes
trust_condition: every episode repays at least one earlier question
stop_reason: the rules of evidence change whenever the writer needs them to

Every line here can change a production decision. Because she usually enters muted, the first shot needs the visible action of a badge being cut. Because she accepts limited jargon, the anomalous payment can serve as proof — but it must be explained through a screen close-up and a character reaction. Because she hates rules that shift arbitrarily, the evidence ledger has to be maintained across the whole season.

2.8 Anti-audience and the aversion map#

A product does not need to please everyone. State plainly which viewers are not the current target, and what most easily offends the ones who are.

An anti-audience is not a value judgment. Viewers who prefer slow realism and want audit procedure reproduced rigorously are probably not the primary target for a 75-second paid pilot. That does not license incoherence. It means professional authenticity serves emotional credibility rather than complete instruction.

The aversion map records exit factors: excessive humiliation, repeated misunderstandings, a protagonist made stupid, an antagonist who never pays, cheap coincidence, too much narration, AI face drift, harsh audio, subtitles covering faces. It belongs to the story, the production and the commercial systems at once.

2.9 Hypotheses must be allowed to lose#

An audience profile is not a religion. Every important judgment should be written as a testable hypothesis with a disconfirming condition attached.

hypothesis_id: H_AUD_004
statement: the target audience is more drawn to "counterattack by professional evidence" than to "wealthy-family bloodline identity"
test:
  variant_a: opening frame on the anomalous payment record
  variant_b: opening frame on the DNA report
controlled_elements: same character, duration, subtitle weight, audience targeting
supporting_signal: A shows higher effective plays and higher return rate
disconfirming_signal: B leads consistently across multiple creative sets, and comments focus on parentage
decision_if_disconfirmed: move the bloodline mystery earlier, but keep professional competence as the payoff mechanism

Note that one test never proves a permanent truth. Creative execution, traffic source, sample size and timing all move the result. What you are looking for is the direction most worth betting on next, for this project.

2.10 SOP for audience research#

First, write down the business model and the viewing entrance. Second, choose five to ten works that genuinely compete for the same attention and the same emotion — not merely the same genre. Third, collect behavioral data, comment language and the promises made by ad creatives. Fourth, code the comments into needs and exit reasons. Fifth, interview or watch alongside a small number of target viewers, observing where they actually stop and how they retell it. Sixth, build the desire stack, pain mechanism, viewing context, persona and aversion map. Seventh, rewrite the most important judgments as falsifiable hypotheses. Eighth, test with low-cost story cards, opening shots or rough cuts. Ninth, write the results back into the opportunity brief.

The point of all this is not the satisfying feeling of "understanding users." It is reducing the uncertainty of one specific decision. If the team is deciding whether to open on the badge, the DNA report or the photo of the abduction, the research should produce evidence for that decision.

2.11 Fault tree#

Symptom: high creative click-through, but comments are all about the cast or the art style. Visual novelty may be masking story demand. Check effective plays, return rate and how viewers retell it; do not mistake aesthetic attention for genre demand.

Symptom: interviewees say they like it, and real playback drops off fast. Suspect polite answers, finish-quality bias, or a different viewing environment. Go back to testing at the real entrance, shorten the questionnaire, observe behavior instead of attitude.

Symptom: the definition of the target audience keeps widening. Usually the team is afraid of excluding market. Require that every added segment be justified by the story or production decisions it changes. If nothing needs to change, it is probably just a statistical label.

Symptom: the team chases every comment with a revision. Coding and prioritization are missing. First decide whether a comment reflects a general mechanism, a niche preference, a production defect or an individual opinion — then check it against behavioral data.

Symptom: the profile is highly detailed but cannot choose a hook. The profile describes a life instead of a viewing task. Delete fields that cannot influence the work, and add trigger, desire, trust and exit reason.

2.12 Checklist, exercises and deliverables#

Check whether the audience document answers: what environment she enters in; what the first visible trigger is; how much oppression she will tolerate; which payoff she expects; who she wants to be temporarily; what would make her feel cheated; whether it is still comprehensible with the sound off; which judgments are only guesses; and what evidence would change our minds.

Exercise one: rewrite "likes female revenge stories" into at least two mutually conflicting desire stacks, and design a different first shot for each. Exercise two: take one episode and code fifty comments, with positive/negative banned as the only classification. Exercise three: have three people who know nothing about the project read the text shot list for episode one of Backlit Takeover, then record how they retell the story.

Deliverables for this chapter: the audience desire stack, the pain mechanism map, the viewing context sheet, a behavioral persona, the aversion map, the comment coding table, and the audience hypothesis register.

A note on sources#

The method here combines the recurring lessons about openings, emotion, paid acquisition and comments found in this project's Xiaohongshu corpus with hypothesis-driven methods from product research. Social platform posts are good for discovering creator language and candidate mechanisms; they cannot on their own prove the preferences of a whole population. Any real project still has to validate with its own creatives, its own entrance and its own data.