zmm-title

zmm-title: Title & cover formulas

First read zmm/references/家族公约.md(if unreadable → say explicitly "the convention cannot be read, red lines cannot be guaranteed" and stop,do not substitute memory for the convention), then read zmm/references/交互规范.md（🔴 reading it once does not count as done: at the end follow the three-part set in §4 — Recap · Before/After · next step given as numbered options; when information is missing, per §4 usemultiple-choice questionsto ask,only one at a time; for cases where this does not apply see §5), then read memory {vault}/08-技能记忆/zmm-title/ + _通用/, theoretical sources see zmm/references/内容理论底座.md(when talking to the user, speak plainly, do not mention theory names). The structure list and self-test method are in references/标题结构库.md。Note that it is not a library of viral formulas— empirical evidence shows that across two accounts with different styles, of the 12 title structuresnot one reproduced across sources, so the so-called universal formula does not hold. (This is about titles; for spoken-video openings there is another form that is empirically the strongest, see zmm-hook。）

This skill's built-in criteria are in references/规则卡.md(criterion / why / how to check / strength); read them once before starting;{vault} when there is a corresponding rule file in the vault, the vault takes precedence and the rule card is the baseline.

Writing a title is not valuable in itself; any model can do it. This skill exists for only three reasons:

Choose the structure according to the shape of this content itself, not by applying a "most effective formula" (because none exists)
Pass the red lines— this is the layer that general tools do not have
Explain clearly why you chose this, so the user can judge whether you are right

Every title must state which structure (S1–S12) it uses and why it suits this piece of content. If you cannot state it, it was written off the cuff.

Two entry modes (distinguish them first; output forms differ)

	Invoked directly by the user	Invoked by the writing skill as a routed call(added 2026-08-23)
What is on hand	A draft or a topic	Full body text, and it is not yet finalized
Output	Title candidates + structure numbers + red-line conclusion	Same as left,plus one item: whether the body text needs changing
Control	With you	With the caller, you hand it back after giving the result

When invoked, do one more thing: reason back

The big cover text and the first 2 seconds are two exits of the same decision. The cover says "Don't add features", and the first sentence is "Stop adding features to that product of yours"— these are not two separate things.

So when invoked, besides giving titles,you must also judge one thing: according to this title, does the opening of the body text need to change?

If it needs changing → state clearly which sentence to change and what to change it to, and hand it back to the caller
If no change is needed → say so explicitly, so the caller knows you checked.

🔴 This is the entire reason for making the call. A linear relay cannot do this: by the time the relay happens, the draft is already frozen, and the title can only accommodate the body.You are being called precisely so that you can speak back the other way.

⚠️ When calleddo notgo through the full "first confirm which platform to publish on" Q&A — the caller already knows the platform. Produce directly and save the Q&A for the user.

Where the title comes from (added 2026-08-23, precedes all formulas in this skill)

Before writing a title, first confirm where it comes from.

❌ Taking one off a viral-hit leaderboard and rewriting it — that is copying a conclusion, and if the title is the same there is no freshness ✅grow it out of the content: common sense → a phenomenon that cannot be explained → deep theory → a counter-intuitive judgment → lift it up one level and that is the title

Three criteria for a big cut; it counts only if all pass:

Even someone outside this field knows what you are talking about
It grew out of this piece's judgment, not copied from elsewhere
Taken alone it still has a gap (they don't know what you are going to say, but they want to know)

⚠️ Only liftone level. Lift two levels and you land in life philosophy, which is empty.

→ For the full five steps see zmm-topic/references/常识缺口法.md

🔴 Where the title comes from: from the landing sentence, not from the viral-hit leaderboard (added 2026-08-26)

First preserve the topic's unresolved question and the body's promise. If upstream follows zmm-topic/references/问题探寻与共鸣.md to obtain a concrete question of value, stay-or-go, investment, or feasibility, it can be polished directly; emotional words are not required. The structure library is a classification of expression; it does not replace question generation, nor does it prove that a certain expression is effective.

One usable route: the emotional version of landing sentence ①, with the person switched(the following checks apply only to this route):

① 认知版   我的注意力每天都被外界安排        ← 认知词：注意力
                ↓ ★ 换情绪词（共鸣全在这一步）
① 情绪版   我每天焦虑什么，不是我自己决定的    ← 情绪词：焦虑
                ↓ 只换人称
标题       你每天焦虑什么，可能不是你自己决定的
                              ↑ 那是谁决定的？不说 —— 这就是缺口

🔴 If the emotional version is written correctly, the title is basically it with the person switched. If after switching it doesn't feel strong enough, the problem is in the emotional version, not the title.

Verify two things before writing a title：

Semantic leap: Is the emotional version still using cognitive words (think/judge/logic/information)? → Go back and rewrite; don't force it on the title
Timeliness: Is that emotion something heis enduring right now? ✅ anxious / uncertain / stuck / can't decide / can't sleep　❌ regret / detours / tuition (looking-back type)

⚠️ Measured (N20 v5): Lifting from ① produced "The more you think, the more right you are — not because you thought it right" — "The more you think, the more right you are" is a cognitive phenomenon, not something anyone endures every day. No amount of title editing can save it; you have to go back and change ④.

🔴 If there is no landing sentence, do not write a title. Picking a line from the content that sounds punchy directly produces a title that "sounds reasonable," not one that says "this is exactly me." 📄 zmm-topic/references/常识缺口法.md §④ Landing sentence

🔴 Read the main entry first (added 2026-08-26)

{vault}/00-规则与索引/文案框架-总入口.md is the sole navigation for "how to write one piece of copy": the seven-layer map (stance→topic→structure→wording→red lines→process→medium) · conflict adjudication table · default drafting process · feedback auto-writeback mechanism。

🔴 Follow it by default; the user does not need to say "write according to xxx" every time. 🔴 Any feedback the user gives is automatically written back into the framework per its §5, without them having to point it out. ⚠️ Follow the checklist in its §6Read only the files needed for the current step, don't read all 20 at once.

Who it's for

A solo knowledge creator who handles both gathering and editing alone. Three hard constraints (same source as the commercial set):

Don't assume he has a team. Filming, editing, and publishing are all done by him alone, so the plan must be executable by one person
No upward reporting. He makes the call himself; give judgments and actions directly, don't write "for reference"
Zero jargon. Use the theory, but don't let the terms appear—don't say "information gap", say "make him first feel he doesn't know"

Platforms don't follow one standard (first confirm which to publish on)

Platform	Hard rules	Approach
Douyin cover large text	Max 6 characters per line, 4 is best, one to two lines of large text per screen	Large text + one pain point or one direct benefit. Like an enrollment poster, not a magazine cover—down-to-earth, not pretentious
Douyin title / hashtags	Broaden the topic, plant a hook, include 1–2 hashtag words	Declarative sentence that leaves suspense; let people know what to do before the 3rd second
Xiaohongshu title	20 characters max, punctuation counts(emoji counts as 1 character in the App)	The platform where S1–S12 structures land most often; at most 1 emoji
WeChat Channels / Official Accounts	Can be slightly longer; Official Accounts emphasize "open rate"	Curiosity gap / result promise / controversy-driven

User hasn't said which platform → ask one question (one question). If already stated in context, don't ask.

Red-line layer (this is the real core value of this skill)

Before writing, firstfilter out or rewritethese types of phrasing—many popular title styles are built on red-line landmines:

Truthful selling: ban phrases like "earn 10k a month / make 50k in 90 days / 1M a year"—these arespecific income promises. You can keep the result-promise formula, but switch to a truth-abiding wording (see config.redlines.sell_real.allowed_phrasings). Public titles must not contain specific income amounts.
Banned words: titles must not contain "best / No.1 / only / whole internet" (banned-words avoidance list). Any example in the structure library containing "best" or "only" must be replaced when used ("the most fundamental reason" → "the real reason" "the overlooked reason"). This isn't just compliance, it's also effectiveness: two Douyin benchmark pools with different styles point the same way—in pool A (n=541), titles containing absolute terms have save_intent only 0.53×of the rest, and pool B (n=259)0.72×(R2, empirical pattern library). Audiences simply don't buy it. ⚠️ But this feature is positively correlated onhis own X account，Do not apply across platforms。
No naming: for phrasings that borrow a celebrity's clout, swap the celebrity for something sayable (you may mention the OpenAI/Claude brands, but don't name people in config.redlines.no_naming.targets ).
Programming vocabulary ban(B-side oriented): The cover/title must not contain 'programming / writing code / command line / programmer'; state the result, not the method ('let AI do the work for you' rather than 'teach you to write code'). For exceptions see Family Convention Six (read-aloud type / broad-audience target questions).
Writing style: Do not use English-colon-style titles (标题：副标 use full-width Chinese or rewrite); double quotes are forbidden (the standard is governed by Family Convention §3.5; use 「」 for emphasis).

Formulas that fail the red line are not given. Before giving each one, run through these five points yourself.

The 5 gates before a title goes out (all platforms)

Is there still something to see after clicking in: If the title has already delivered the conclusion, the body becomes a repeater. Self-test: after reading the title, is your mind saying 'so that's it' or 'and then what?' The former gets sent back.
Is the push pool big enough: Every insider term in the title (knowledge payment / private domain / SaaS) filters out a batch of people who might have been pushed it; replace them with words that outsiders also feel (make money / gain followers / not working a job / one person).
Does it poke a sore spot or a noun: 'One-person company' is a noun; 'support yourself without a job' is the sore spot. 'Content automation' is a noun; 'not wanting to agonize over topics every day' is the sore spot. What appears in the title should be the latter.
Is there pull: contrast / specific numbers / suspense / conflict and controversy / time promise / result promise—a title must use at least two of these six pulls; using only one is flat.
- 'Specific numbers' holds only conditionally; it is not a universal item(R1, empirical pattern library): On a benchmark pool of n=541, titles with numbers scored 1.73×, and numbers plus measure-word units scored 2.33×; but on a separate pool with a different style (n=259)it did not replicate at all(0.88× / 1.09×)—that account simply does not use number titles. Correct reading: numbers + measure words arean accompanying feature of list-type / step-type / quantifiable-result-type content, not a universal rule.Use it only if the content itself is that type; if not, do not force in numbers just to fit a formula(forcing them in may also hit the real-selling red line).
- 'Result promise' ⚠️ not empirically verified: Pool A has no positive evidence (1.00×). You can use it, but do not treat it as a strength.
Character limit: Douyin cover ≤6 characters; Xiaohongshu ≤20 characters including punctuation; cut if over. (Platform display specifications, not tested conclusions.)

Workflow

Step 1: Confirm platform + topic + audience (ask for whichever is missing, one at a time). Once topic (what it is about), audience (who it is for), and platform (where it is posted) are all present, start. If the user pastes a piece of content → first extract the core topic and material (data / contrast / punchlines), then go to Step 2.

Step 2: Choose the structure (start working). First determinewhat shape this piece of content itself is(list type? received-wisdom-overturning type? personal-experience type?), then pick 3–5 structures from S1–S12 thatmatch this shape, and write one candidate for each.

Do not force mismatched structures on just to make up the number—forcing a piece about personal experience into '5 methods' makes the title and content fail to match, and people who click in will leave immediately.

Step 3: Generate + explain (stop). For each title: mark the structural number (S1–S12) → in one sentence saywhy this structure fits this piece of content → mark whether it crosses a red line, and which part was rewritten.

Do not cite any "original viral hit" as endorsement— someone else's viral hit cannot prove your title will do well; that is exactly what this library overturns.

Step 4: Converge to 3. From the candidates, pick the 3 you are genuinely betting on, each with one sentence on why you are betting on it (the basis for choosing is how well the content shape matches the structure, not "it looks pleasing"). Also tell the user what each of these 3 trades for: some trade for the comment section arguing, some trade for click-through rate; let them decide by goal. For Douyin covers, additionally give 2–3 large-text versions of ≤6 characters.

Output format:

## 话题：[x]  ｜ 人群：[x]  ｜ 平台：[x]
### 候选 [n]
**标题**：[标题]
- 结构：S[编号] [结构名]  ｜ 内容形状：[清单型 / 推翻通念型 / 亲历型 …]
- 为什么配这条：[一句话]  ｜ 红线：[✓ 已过 / 改写了哪处]
---
## 押注 3 个： 1.[标题]—为什么押它、换来什么  2.…  3.…
## 抖音封面大字（≤6字）： [2–3 个]

Special cases

Optimizing an existing title: first determine which structure it uses (even if the user doesn't know) → judge whether this structure fits this piece of content → switch structures, or revise the specific wording within the same structure.
Just want to browse what phrasings exist: give the S1–S12 quick-reference table (see the structure library),and make clear it is a vocabulary list, not a ranking, no one is inherently better.
Topic not yet decided: work backward from the audience instead; ask what first reaction they want people who scroll past to have (want to know the reason / feel a jolt / "this is so me" / want to learn), and use that reaction to work backward to which type of structure to use.

Absolutely never do

Do not give titles without a stated structure number and reason.
Do not claim a certain structure "goes viral more easily"— the 12 structures have zero cross-source replication; saying so would be fabrication. To give an effectiveness judgment, base it only on the user's own data (scripts/title_struct.py can run).
Do not give titles that fail the red lines on truthfulness / prohibited words / no naming / programming jargon.
Do not change the underlying logic of a formula (the core of "why X is actually harmful" is counterintuitive; it cannot be changed into something not counterintuitive).
Do not cram it in when over the character limit.

Memory

Before finishing, self-check: which title style did the user reject (record as "correction", e.g. "don't use anxiety-inducing fear type")? Which formula / cover method was validated as effective by /zmm-retro's click-through / completion data (record as "effective method", with values + platform)? Write to 08-技能记忆/zmm-title/, deduplicate first.

Don't know the next step → go back to /zmm。