fusengine/ux-copy

UX Copy — Self-Audit + Ban-Lists, One Register Per Page

When

Any time visible copy is written or edited — headlines, subheads, eyebrows, button labels, body, captions, alt text, error messages, empty states — inside generate/ critique/audit/polish (design-method) or as a standalone copy pass. Register is resolved once, at Gate 0 (design-method/SKILL.md), before this runs — never re-guessed here.

Input

• Register (brand/product) from Gate 0.
• Sector, if resolved, for references/voice-tone-sectors.md.
• Every visible string in the deliverable.

One register per page — hard rule

A single copy register per page (references/copy-self-audit.md §7): don't mix technical mono, editorial prose, and marketing punch on one surface unless the brand voice explicitly calls for it. Register direction is conditional exactly like layout and motion (design-method/references/register/copy.md §3): brand copy must trace to the one-sentence POV — run the Competitor Lift Test on the paragraph itself, could it run unchanged on a competitor's site; product copy stays precise and domain-specific, naming the real object and the real action, never marketing voice bleeding into the app shell.

Ban-lists — mechanical (grep), not vibes

1. Filler verbs / hype adjectives — elevate, seamless(ly), unleash, next-gen, revolutionize, empower, supercharge, game-changer, cutting-edge, unlock, robust, effortless, frictionless, plus delve and tapestry. Full grep pattern and flag-not-fail nuance (read against the POV or the domain-specificity floor before rewriting): design-method/references/register/copy.md §1.
2. Slop placeholder names — Acme, Nexus, SmartFlow, Cloudly, Initech, Globex, Contoso, Umbrella Corp, Zenith (Labs/Tech). Blocks unless the brief explicitly names a fictional example on purpose. Grep + rule: design-method/references/register/copy.md §2.
3. Fake-precise numbers — 92%, 4.1x, 48k, 99.99%, or any figure not traced to real data, brief/brand guidelines, or an explicitly labelled mock (<!-- mock -->/"example"). Real data reads organic and messy (47.2%), not fake-perfect. references/copy-self-audit.md §3.
4. Title Case headlines — sentence case is the default register for headings and body copy; reserve Title Case for UI labels, buttons, and eyebrows only, never a body headline, unless the brand style guide explicitly specifies Title Case. Inconsistent or unjustified Title Case on headlines is a recurring AI-slop tell (grounding corpus; general consistency principle also flagged in the reference clone's capitalization check — apply it here as a default ban, not a mere consistency note).
5. Humor in error copy ("Oops!"/"Whoops!") — banned. Errors use the what-happened + why + how-to-fix formula (references/templates/error-messages.md), never cute framing — a frustrated user doesn't want jokes. Bad-example precedent already in this skill: references/voice-tone-sectors.md error row flags exactly "Whoops! Something went wrong." against the correct "Connection lost. Try again or continue offline."
6. AI production-tell catalogue — version-label eyebrows (v0.6/BETA/EARLY ACCESS), numbered section eyebrows (00 / INDEX), scroll cues, weather/locale strips, "quietly trusted by", decorative photo-credit captions, fabricated live counters, poetic craftsman labels, generic step labels (Stage 1/2/3). Full table: references/copy-self-audit.md §2.
7. Em-dash as a crutch — a single — isn't a hard fail; 2+ occurrences reading as a repeated tic is. references/copy-self-audit.md §1 (shared gate with design-review).
8. "Not X. Y." manufactured-contrast fragments — one instance is fine, 2+ on one page is a blocking tic. references/copy-self-audit.md §4.
9. "theater"/"théâtre" keyword — any hit blocks, near-universal AI copy tell. references/copy-self-audit.md §5.
10. All-caps body copy — 3+ consecutive all-caps words outside a recognized UI label blocks; fine inside a button/eyebrow/badge. references/copy-self-audit.md §6.

Steps

1. Confirm register (inherited from Gate 0) — never re-resolve it here.
2. Write/review against the sector voice profile (references/voice-tone-sectors.md, NNG 4 dimensions) and the matching sector guide if one applies (references/copy-saas.md, references/copy-fintech.md, references/copy-ecommerce.md).
3. Apply the formula references for the surface being written: buttons/forms/toasts (references/microcopy-patterns.md), CTAs (references/templates/cta-patterns.md), errors (references/templates/error-messages.md), empty states (references/empty-states-copy.md), onboarding (references/templates/onboarding-copy.md). Project-level output uses references/templates/copy-guide-template.md.
4. Run every visible string through the ban-lists above.
5. Final read (references/copy-self-audit.md §7): grammatically broken, unclear referent, hallucination-sounding, "LLM trying to sound thoughtful" (forced metaphor, mock-poetic aside, cute-but-empty wordplay). Rewrite; when unsure, replace with the plain functional sentence — boring-correct beats clever-wrong.
6. Mechanical checks (1-10 above) resolve on the spot. Judgment calls — does this line actually trace to the POV, does it sound like forced profundity, does the aphoristic cadence read as punchy ad copy pretending to be product copy — are named specifically (quote the line, name the section) and routed to the fuse-ai-pilot:challenger gate inside design-review, never self-scored (design-method/references/register/copy.md §4; references/copy-self-audit.md, "Not self-audited — routed to challenger").

Failure Handling

• Sector not resolved / no sector guide fits → default to references/microcopy-patterns.md's generic patterns, flag the gap instead of inventing sector voice.
• A ban-list hit is contractually required by the brief (brand literally named "Acme", a launch brief that genuinely wants a BETA eyebrow) → note the override explicitly in the deliverable rather than silently keeping it unflagged.

Output

• Every visible string checked against the 10 ban-lists above, register-consistent, one voice per page.
• Judgment-call flags (if any), quoted and located, handed to design-review's challenger gate — not resolved solo.

Next → design-review — Part 1's em-dash/production-tell checks are the same gates

enforced here; Part 2's challenger judges any remaining judgment call (aphoristic cadence, POV-tracing) this skill flagged but didn't self-score.

References

| File | Purpose | |------|---------| | references/copy-self-audit.md | Canonical self-audit — em-dash, AI tells, fake numbers, final read | | design-method/references/register/copy.md | Filler-verb + slop-name ban-lists, register direction (external, canonical there) | | references/voice-tone-sectors.md | Sector voice profiles, NNG 4 dimensions, do/don't examples | | references/microcopy-patterns.md | CTA/form-label/validation/toast/button-state patterns | | references/empty-states-copy.md | Empty-state copy formulas per type and sector | | references/copy-saas.md | SaaS sector copy examples | | references/copy-fintech.md | Fintech sector copy examples | | references/copy-ecommerce.md | E-commerce sector copy examples | | references/templates/cta-patterns.md | CTA label examples by context | | references/templates/error-messages.md | Error message formula + good/bad examples | | references/templates/onboarding-copy.md | Welcome-flow / first-run copy patterns | | references/templates/copy-guide-template.md | Project-level copy guide output template |