From product strategy to page copy
Status: canonical method for this organization; the Research log holds its passes. It orders the work from customer evidence to a tested page and holds the reusable templates. The rules each step applies live in Principles; the sources behind each step live in the source notes cited inline. Product-specific application (audience, glossary, claims) lives with the product owner, never here; see How a product organization uses this in the organization's AGENTS.md.
The thesis
Copy is the last step, not the first. Practitioners from different fields reach the same order: understand demand, make positioning choices, tell them as a story, then adapt that story into pages. Positioning is not messaging and not a tagline (Dunford and Kramer). When copy fails repeatedly, go back to positioning or evidence, not to wording. The words themselves come from customer research, not invention (Wiebe review mining).
The eight steps
1. Collect demand evidence
Interview recent switchers and churners about specific past events: switch interviews in rounds of about ten, stepping back after each round; stop when the stories repeat (Moesta and Spiek). Use Mom Test rules and "tell me about the last time" prompts (Fitzpatrick, Torres).
Mine the words prospects already use: reviews of the alternatives and adjacent products, forums, support tickets, sales call recordings. Sort phrases into love, hate and worry and keep them verbatim (Wiebe).
Output: the four forces per segment (push, pull, anxiety, habit), the buying timeline (Moesta), and a language bank (template below).
Only past-behaviour evidence may become problem copy. "I'd totally use that" never becomes a claim (Mom Test).
2. Diagnose and set direction
Write the strategy kernel: diagnosis (what is going on, the crux), guiding policy, coherent actions (Rumelt). A vague diagnosis produces a vague headline.
Stress-test through prose before pages: a PR/FAQ for the product or launch, memos over slides (Working Backwards, six-pager). If the press release is not compelling, rethink the product, not the prose.
3. Position
Fill Dunford's five components in order: competitive alternatives (what customers would really do without you, often "do nothing" or a spreadsheet), unique attributes, value themes with proof, best-fit customers, market category (Obviously Awesome).
Pick the contrast frame by product type: vertical solution, buy versus build, new way against the status quo, or ten times better than an incumbent (MKT1).
Default to an existing category plus a subsegment qualifier. A new category name forces two extra jobs (define it, make people care) before you can differentiate; use a problem-first "new game" story only when buyers already feel the shift (Play Bigger critique).
Capture the result in the positioning statement and the onlyness line (templates below). If the onlyness line needs three qualifiers, the positioning needs work, not the copywriter (after Neumeier's onlyness test).
4. Build the narrative
For a "new way" product, use the strategic narrative: a named shift, winners and losers, the promised land (the customer's future, hard to reach without you), capabilities as answers to named obstacles, proof from a similar customer (Raskin).
For a head-to-head or niche product, use the sales pitch order: insight, alternatives and their limits, perfect world (buying criteria), introduction by category, differentiated value, proof, objections, the ask (Sales Pitch).
5. Write the messaging document
One message house per product: a roof (category plus outcome, one sentence), three pillars in customer words, and a foundation of two or three proofs under each pillar; the glossary and voice rules sit beside the house (message house).
Fill pillars only from pains and gains you demonstrably relieve; an unrelieved pain is not a claim (Value Proposition Canvas).
Keep one glossary for code, UI, docs and marketing (ubiquitous language). A mismatch between the homepage noun and the UI noun is often a sign the model itself is unclear (Evans).
Write variants per audience and awareness stage from the same house, never a second house.
6. Structure the page
Answer the reader's questions in this order: clarity (what is it), relevance (is it for me), value (do I want it), differentiation (why you) (Wynter message layers). Lay that over the narrative:
Hero: category plus outcome, audience cue. Passes the "read only this" test (Shapiro).
Problem: the push, in customer words.
Shift or insight, and where the alternatives fall short.
Three value pillars, each with its proof beside it.
How it works: the PR/FAQ solution paragraph or the job steps.
Objections and FAQ: the anxiety and habit forces, the external FAQ.
Call to action matched to the buying stage.
Match each page to the reader's awareness (Schwartz):
| Awareness | Typical entry | Lead with |
|---|---|---|
| Most aware | pricing, signup, returning visitor | the offer and terms |
| Product aware | branded search, referral | what is new or better |
| Solution aware | category search, comparison | why this approach wins, against named alternatives |
| Problem aware | use-case page, mission page, content | the problem in their words, then the solution |
| Unaware | ads, social, essays | an identification line or the shift; no product name first |
A homepage cannot serve every row. Give each awareness stage and each segment its own door (use-case or mission page) and keep the root for one stage.
7. Draft
Write the hero last. Reuse language-bank phrases verbatim. Clear before clever. One reader, one idea per page (working rules).
Ugly first draft, then edit in passes (everything, strongest points only, the reader's view, then delight), read aloud, let it rest (Handley, Graham, Bezos memos).
Run every claim through the three tests: can the reader picture it, could it be proven false, could a competitor sign it (Harry Dry). Record product claims in the product's claims register (template below).
8. Test and iterate
See the testing playbook below. When copy fails in tests repeatedly, return to steps 1 to 3 (Dunford).
Two cases the source notes do not cover
Evidence class for both: synthesis, a hypothesis until tested on the real audience.
A horizontal product with one featured mission use case. The root page addresses one audience at one awareness stage and names the category plainly; the mission page addresses a problem-aware, mission-driven reader; each use case gets its own outcome-shaped door. The glossary and the message house stay shared, so the doors never contradict each other. Decide in writing which audience the root addresses; an undecided root drifts toward whoever the last writer imagined.
A multi-stakeholder B2B sale. Buyer, champion, user and blocker (for example data protection or a works council) each need different sentences that must not contradict. Write one message house, then a sentence set per stakeholder: the outcome the buyer is measured on, the daily pain the champion carries, the friction the user feels, the objection the blocker checks. Blocker answers (data location, contracts, oversight) remove objections; they are never the headline.
Testing playbook for low traffic
Do the arithmetic first. Per variant n is about 16 p(1 minus p) divided by the squared absolute lift. At a 3 percent signup rate, detecting a 20 percent relative lift needs about 13,000 visitors per variant. If that takes longer than four to six weeks, do not A/B test conversion; use the methods below and ship on judgment (Kohavi, Evan Miller).
Test comprehension, not confidence. Five-second tests with 5 to 10 target-like people: "What does this company do? Who is it for? What would you do next?" Grade answers as understood or not; Likert confidence scores missed real comprehension failures in the CXL value proposition study (CXL, Lyssna).
Blind tagline test. Your tagline and three competitors' without names; can outsiders tell who does what (NN/g).
Message testing panel with verified target buyers for clarity, relevance, value and differentiation; about 15 respondents per narrow audience reach saturation, Wynter's default test size (Wynter; https://wynter.com/research-playbook, read 29 October 2026).
First-click test on homepage and pricing: a correct first click predicts task success (87 percent against 46 percent) (Sauro, Bailey).
Moderated usability: three rounds of five people with fixes between rounds (Nielsen).
Preference tests are tie-breakers only: stated preference often differs from behaviour, as the CXL pricing study says itself (web reading note).
Instrument big effects only: scroll depth, CTA clicks, signup drop-off, with a written prediction before each change and a before/after read that names its confounds (season, channel mix) (web reading note).
If you A/B test: pre-register metric, minimum detectable effect and sample size; run full weeks; do not peek (Evan Miller, Kohavi); check sample ratio; distrust any lift far above the 0 to 4 percent that real tests average (Twyman's law, Kohavi et al.); test big swings (value proposition, pricing display, proof block), never button colour (Qubit, Berman et al., Kohavi).
Keep a decision log: hypothesis, evidence, change, observed result.
Simulated persona panels are an exploration tool, not evidence of reader response. Every simulated round names the real-reader check it will be compared against (open question, see Conflicts, simulated persona panels).
Templates
Paraphrased from their authors; each author and book is in the positioning source note.
Positioning statement (Moore). For [target customer] who [need], [product] is a [category] that [key benefit]. Unlike [primary alternative], it [main differentiator].
Positioning canvas (Dunford). Competitive alternatives / unique attributes / value themes with proof / best-fit customer traits / market category / optional trend.
Three blanks (Kramer). Who is it for? What is it? Why is it better, and better than what?
Onlyness (Neumeier). Our [offering] is the only [category] that [difference], for [who], in [where], who need [why], at a time of [trend].
PR/FAQ (Bryar and Carr). Heading (name plus one plain sentence); subheading (who and benefit); summary paragraph; problem paragraph (quantified pain); solution paragraphs (how it solves it, how it differs); spokesperson quote; customer quote; getting started; external FAQ (price, how it works, support); internal FAQ (alternatives, market, economics, risks, what must be true).
Strategic narrative (Raskin). Big relevant shift; winners and losers; promised land; capabilities against named obstacles; evidence from a similar customer.
Sales pitch (Dunford). Insight; alternatives and limits; perfect world; introduction via category; differentiated value; proof; objections; ask.
Pitch (Singer, Shape Up). Problem as one specific story; appetite; solution at sketch level; rabbit holes; no-gos. No-gos keep copy honest about what the product does not do.
Job story (Paul Adams's team at Intercom; named by Alan Klement). When [situation], I want to [motivation], so I can [outcome]. The "when" becomes the audience cue, "I want to" the promise, "so I can" the payoff line.
Outcome statement (Ulwick). [Minimize or maximize] + [metric] + [object] + [context]. High-importance, low-satisfaction outcomes become metric-shaped proof claims.
Strategy kernel (Rumelt). Diagnosis; guiding policy; coherent actions.
Switch interview (Moesta and Spiek, question wording paraphrased). When did you buy, where were you, who was with you? When did you first think the old way was not working, and what happened that day? What did you start noticing? What made you start actively looking, what else did you consider, what did you rule out and why? What was getting worse (push)? What did you picture afterwards (pull)? What almost stopped you (anxiety)? What did you have to give up or learn (habit)? What made that day the day, who else had a say? What happened the first time you used it?
Mom Test guardrails (Fitzpatrick). Their life, not your idea; specific past events, not future opinions; listen more; deflect compliments, anchor fluff, dig under requests; end with a commitment or next step.
Message house. Roof: one sentence of category plus outcome. Pillars: three value themes in customer words. Foundation: two or three proofs under each pillar (metric, quote, logo, case). Beside the house: glossary and voice rules.
Audience brief (this organization). Reader and role (buyer, champion, user, blocker, agent); language and register; awareness stage and entry point; the job in their words; push, pull, anxiety, habit; accepted proof types; words that mark us as outsiders; evidence held and evidence missing, each with its source. Agent readers: writing for an agent reader.
Language bank (this organization). Per audience: verbatim phrase; speaker role (never a name in a shared organization); date; source; category (job, pain, alternative, fear, success, outsider word); evidence class. Language banks with customer speech live in the product's organization.
Claims register (this organization). Claim as published; surface and location; evidence and its date; live-check owner; legal basis checked (UWG, PAngV, AI Act, data protection); status. Legal floors: UWG and PAngV, AI Act, price display.
Copy teardown (this organization). Page and date captured; hero sentence; category named or not; audience cue; awareness stage targeted; proof types and placement; contrast frame; CTA; what a visitor would believe after five seconds; what we take and what we avoid. Teardowns are dated product work and live in the product's organization.