Conflicts and open questions: copy and messaging
Status: canonical list of genuine source tensions, scope limits and uncovered evidence. Each entry says when judgment or testing on the real audience is needed. Do not silently erase a disagreement: add to it, resolve it with evidence and date, or move a resolved rule into Principles.
Resolved by context
Long copy against short copy. Hopkins, Ogilvy, Caples and Sugarman argue for telling the full story; NN/g (concise copy measured 58 percent more usable) and Unbounce (word count correlates negatively with conversion) argue for brevity. Resolution: length follows awareness and the size of the decision. Write a short, complete hero, then as much as a hesitant buyer needs, with every section passing the concise test (disagreements, Morkes and Nielsen, Unbounce).
Clarity against curiosity. Caples ranks curiosity third; 37signals and Wynter put clarity first; Halbert and Schwartz's identification headlines use intrigue. Resolution: homepages, product pages and pricing need clarity; curiosity belongs in ads, emails and content for readers who are not yet looking (disagreements).
Hype against understatement. Direct-response agitation (Kennedy, Halbert) against Hopkins on superlatives, NN/g's marketese penalty (objective copy 27 percent more usable, Morkes and Nielsen 1997) and Dry's "honesty feels warmer". Resolution: for software buyers and saturated markets, factual understatement wins; agitate only in the reader's words and only as far as the real cost justifies (disagreements).
Facts against story and identity. Unknown products must first make a clear, specific claim; identity selling works for known brands in saturated markets (Schwartz's fifth stage of market sophistication, identification). Stories work inside pages as proof, and a manifesto belongs on its own page (Schwartz, exemplars).
Formulas against research. Formulas give order; content comes from customer research (Wiebe).
Personas against one real reader. Segments decide whom to target; each page is written to one real customer whose words you hold (Shleyner, Torres).
Jobs as progress against jobs as activities. Progress-school interviews (Christensen, Moesta, Klement) feed the hero, story and objections; activity-school outcomes (Ulwick) feed proof points and feature sections (JTBD schools).
Category creation against an existing category. Default to an existing category plus a subsegment; use a new-game narrative only when buyers already feel the shift (Play Bigger critique).
Mandatory AI disclosure against not leading with AI. The law requires disclosure in some places (in a chat at first interaction, on deep fake images, audio or video, and near some published text); disclosing AI costs the person who discloses trust under every framing tested (pooled d = 0.81). Resolution: the law sets where disclosure appears, not where copy leads. Marketing leads with the outcome; the chat, the article and a methods page carry the AI facts as a statement of roles, with the reader's control inside it. That trust cost on disclosed surfaces remains and is accepted (mentioning AI, conflicts; principles 28, 29 and 32).
Plan and limit copy against a real capacity model. Several limits can bind at once, the unit shown can be approximate, and a reliable retry time may not exist. Resolution: copy names the binding limit and who can raise it, keeps "about" on approximate units, and states the honest variant when no time is known. Narrowing a broad access grant is a product decision, not a copy fix (activation, tensions).
Open, needs judgment or testing on the real audience
Loss framing against gain framing. Loss aversion averages about 2 in risky choice but is weaker for small stakes and experienced buyers and remains contested. Test both framings where it matters; do not default to loss (loss aversion).
Social proof effect size. Descriptive norms work but are smaller and more context-bound than popular books claim; in two German hotels, norm messages beat no message but not the standard appeal (Bohner and Schlüter 2014). Specific, close proof is the safer bet; measure where possible (Cialdini).
Simulated persona panels against real readers. Some copy methods use LLM persona panels. Whether simulated readers predict real reader response for a product's real audiences is unmeasured. Until measured, a panel round is exploration, and each round names the real-reader check it will be compared against. Queued evidence: silicon-sample literature and its critiques (see Research log, queued persona-panel pass).
Writing for AI answer engines. Evidence is thin, mostly observational and vendor-published; one lab paper (GEO) suggests quotations, statistics and cited sources raise visibility and keyword stuffing lowers it. Google says no special optimization is needed. Treat any tactic as a hypothesis (SEO and AI answer engines).
A/B testing on low traffic. Below tens of thousands of visitors a month, conversion A/B tests on copy are mostly noise. Qualitative comprehension methods replace them; the decision then rests on judgment plus a decision log (testing).
Uncovered evidence (queued passes)
These gaps are owned by queued research passes in the Research log: multilingual and low-literacy end-user copy (A1/A2 readers, translations on a small budget, Leichte against Einfache Sprache and the dignity tension); German B2B trust and claims law; emails and sales material, including the consent rule for advertising email (§ 7 UWG, https://www.gesetze-im-internet.de/uwg_2004/__7.html, read 29 October 2026); recipe and content page copy, creator voice, search visibility and when a translation earns its cost; primary voice-of-customer research for each audience. Already covered in part: the basic distinction between Leichte and Einfache Sprache and generic trust and privacy copy.
Where a product's messaging conflicts with itself
Conflicts between a product's own canonical documents and its shipped copy are product decisions, not research questions. They live in the product's organization with the ruling owner, never here. This organization supplies the method (claims register, glossary, message house); see How a product organization uses this in the organization's AGENTS.md.