positioning-and-messaging：先想清楚你是誰，再寫落地頁

Positioning and messaging

A position is a choice a reader can repeat: who this is for, what they would use otherwise, what is provably different, and what that lets them do. Messaging is that choice worded once and reused everywhere. Most weak copy has no position underneath it, so this skill settles the choice first and then words it in sentences a customer could say.

Do not produce

• Category soup ("a smart, unified solution that helps businesses thrive"): it names no one and no alternative.
• A fill-in-the-blank positioning sentence as the deliverable.
• "Easy to use", "great support", "powerful" or "innovative" as a difference. They are claims until measured.
• "The X of Y" in place of a category, or an invented category word with no plain descriptor beside it.
• A voice guide made only of adjectives ("friendly, professional, bold").
• A customer quote, pain, number, rival fact, market size or "teams like yours" the user did not give.

When not to use

• A full page or its sections: landing-page-copy if installed. This skill hands it a brief. Single headlines or sets of them: headlines-and-leads if installed.
• Price, tiers, trials and guarantees: offers-and-value-propositions if installed. Whether each difference is true and provable in print: honest-claims if installed.
• Logos, colors and visual identity: brand-identity if installed. Strings inside the interface: ux-microcopy if installed.
• Whether to enter a market, or research the agent cannot run (interviews, surveys, search volume). Ask for existing customer words and data, label anything inferred, and recommend the real check.

If a neighbor is not installed, use ordinary judgment. An existing positioning, tagline or voice guide the user wants kept wins: work inside it and improve within it.

Method

A one-liner or a naming question can skip straight to steps 3, 4 and 6.

Work toward the one-sentence idea, then test it and name its category.

1. Know who it is for. If the user has customer data, look at who bought fastest, stayed and referred others, and what they share. If not, describe the likely best fit by situation and need ("sends dozens of invoices a month from a spreadsheet"), label it an assumption, and say who it is not for. Company size alone is not a segment.
2. Draft the one sentence first. The single idea people should repeat. A teammate should be able to use it to reject a feature request or a headline; if not, it is too vague.
3. Test it against "what would they do without us?" Name two to five real alternatives: a rival, a spreadsheet, an assistant, a manual routine, doing nothing. Use the one the buyer actually considers, including the unglamorous one, not the rivals the user admires. Then list the differences those alternatives lack and say what each lets the customer do, in the customer's words. One theme is a fine result.
4. Test it against "what proof do we have?" Proof is a figure the user gave, a feature anyone can check, or a third-party statement. If a difference has no proof, keep it but mark it [unproven: how it could be shown] so the user sees the gap.
5. Then choose the category word a stranger already knows. Compete inside an existing category, or a narrower slice of it with a plain modifier. A new label only when every existing category would make buyers misjudge the product; then pair it with a plain descriptor every time and explain the problem before the solution. Introduce only one unfamiliar thing at a time: either the problem or the solution, with a familiar anchor for the other. A trend ("AI-native") is at most a real why-now line beside the category, never the category.
6. Check the wording against four questions. Could a rival put their name on it and stay truthful? Would someone in the group recognize their own situation, and place the product in a known category within seconds? Can every clause be backed by what the user gave? Does it use the customer's words rather than internal ones? Fix what fails; do not report passes.

Voice and naming follow from the position, not before it.

• Voice. Three or four traits, each written as a habit with a boundary (this, but not that) and a say / don't-say pair. Task text defaults to clear first, brief second, human third. Voice stays constant while tone shifts: plain and direct in errors, billing, security and anything irreversible, lighter only in low-stakes moments. Show one example per surface the user names. If the user has no voice yet, derive it from their own writing, not a famous brand's.
• Naming. A literal name, or a plain modifier on a familiar word, beats a coined one. A coined name carries a one-line plain definition wherever it appears. Say it aloud, check it means one thing, and reject names that need decoding, names that only add a suffix to an earlier one and internal codenames. You cannot search the web or a trademark register for the user, so never say a name is available or clear; tell them to check.

Judgment calls

• Narrow or broad. Default to the narrowest group that meets the near-term goal; broaden later with its own message. Change when that group is too small for the goal the user stated.
• Segment by situation or by company type. Default to need and situation. Use size or industry only as a filter on top.
• One message or several. Default to one controlling idea. Role and stage variants change emphasis and proof, never the claim.
• Plain or clever. Default to plain and specific; a reader who has to decode a line has already left.
• Naming rivals. Default to naming the alternative ("spreadsheets", "general project tools"). Name a specific rival only when the user does, or on a comparison page with dated facts the user supplied that credit the rival's strengths.
• Audience language. Default to the customer's words. Words written to impress investors ("disrupt", "platform play") tell a customer nothing about their own day.
• Repetition. Say the controlling idea the same way across pages, emails and sales calls. Do not give a number of repeats.

Missing facts

Tasks often arrive with a product name and little else. Never fill the gap with invented customers, quotes, pains, numbers, rival weaknesses or market claims. Write the position the facts support and mark the rest: [assumed: who], [unproven: difference], [add: a real customer sentence]. If two positions are plausible, give both with the trade-off and the evidence that would decide, and still deliver the stronger one. Ask one question only when the session is interactive and the answer changes the position; otherwise deliver with the brackets. Never reply with only questions.

If the text already works

If the user's one-liner or description is already specific, names who it is for or what it replaces, and could not sit on a rival's site, say so in one line and change at most a word. Do not rewrite for taste or flatten a deliberate voice.

Output

Deliver the requested thing first, with nothing before it. A one-liner request gets the line, or exactly the number of options asked for, not a brief. A brief fits one page and is written in sentences people say. A critique quotes the failing phrase, says why in a clause, and gives the fix. Add at most two one-line notes, and only when they change what the user does. Honor any stated word or character cap and count rather than assert it. The reason: the user will paste the result somewhere, and every unrequested section is something they must delete or read past.

Quick checks

• The category word is one a stranger knows; the alternative is named; who it is not for is stated.
• Each difference has proof from the user or is marked unproven; no invented evidence anywhere.
• One controlling idea, in the customer's words, that a rival could not truthfully sign.
• Voice guidance has say / don't-say examples; names are plain or defined.

References

• brief-and-voice.md: load when producing a full message brief or a voice guide; has the field list, the voice-habit format and a short worked pair.