Research report · 12 October 2026 · Run 22
Copywriting can structure a draft; it cannot approve a claim
Retain and narrow the Copywriting card for one owner-approved local landing-page draft. Start with an owner-maintained brief/template and reviewed prompt. The source provides a page-copy scaffold, not verified conversion effects, an AI detector, installation clearance or publication authority.
What the mutable source says—and does not prove
| Subject | Inspected source | Use limit |
|---|---|---|
| Draft format | The selected file asks for a page goal, one action, audience, offer, proof, traffic source, awareness and voice. It describes sectioned copy, annotations and headline/CTA alternatives. | Drafting instructions are not evidence of quality, persuasion or host behavior. The owner supplies and authorizes every product fact and local-context read. |
| Performance figures | The selected file asserts +81% conversions, a 38% shorter sales cycle, 28% lower CAC and 175% more referrals. Copy frameworks repeats them and describes a roughly 3× SavvyCal rewrite. | The inspected passages give no checkable study, baseline, denominator, population, method or confounder control. Do not use these numbers as proof or imply expected customer results. |
| Proof and persuasion | The selected file warns against fabricated statistics/testimonials and uses [NEED: …] for missing proof. The framework suggests metrics, comparisons, testimonials and guarantees; its confidence rule may remove hedges. | These are not a claim-to-source register, rights check, terms approval or comparison audit. Keep necessary qualifications; hold unsupported claims rather than rephrase them confidently. |
| “AI-tells” | The selected file directs a style self-check and asserts that buyers reject AI-looking copy. The reference cites Kobak et al., Juzek & Ward and Paech et al. | The cited work does not validate this checklist as an AI-authorship detector or test buyer rejection, conversions or this draft’s quality. Use it, if useful, only for optional style editing from approved facts. |
| Reference wording | The natural-transitions reference includes generic attribution phrases such as “Research shows…”. The selected file gives no publish path in the inspected text. | An attribution phrase is not a citation. The absence of a publish instruction in selected text is not a package or host permission guarantee. These references were reviewed for this claim audit, not as a full inventory. |
The selected file displayed 2.1.0 on 12 October in a mutable main view. The 2 October change commit is attributable to this file, not a verified latest selected-file SHA/date.
Pool-proposed claim-to-proof contract—not enforced by the skill
For one approved, non-sensitive local brief, record one row per material proposed claim: exact draft wording and section; owner-supplied fact or dated source; evidence owner and use rights; audience, product and market scope; numeric calculation, denominator and qualifiers; status approved / revise / hold; named reviewer and decision date. Keep a separate list of missing or contradictory facts. This is our review template, not a source requirement.
A missing claim may become a narrower verified fact or a visible [NEED: proof/terms] hold, not a confident paraphrase.
Why the maintained template is first
| Review question | Source-described Copywriting draft | Owner template + reviewed prompt |
|---|---|---|
| Coverage | Requests purpose, audience, offer/proof, traffic, voice and sectioned alternatives. | Can use the same fields plus evidence, rights, qualifiers, status and reviewer columns. |
| Missing proof | Has honesty instruction and [NEED: …], but no required claim register, rights field or named sign-off. | Can require explicit holds and row review. A prompt can still omit or invent, so human verification remains required. |
| Data and handoff | Directs reading possible local context files if present; selected text does not ask for authorization first. | Scope deliberately submitted approved input and keep draft outside the CMS until a separate release. |
This preference is a process inference, not a head-to-head result. A consented same-brief comparison with a fixed claim/proof rubric could test whether the skill prevents recurring omissions beyond the template. No such comparison exists.
Limits and reconsideration
The run inspected current returned selected text, three directly needed references and three cited research abstracts. The mutable views were marked crawled today, not pinned. It did not establish a full present inventory, reference-loading behavior, authorized-data scope, host behavior or safety. No real brief, customer facts, page, CMS, conversion data or host was supplied; no skill or prompt was installed, run, quality-compared or tested for AI detection or conversion.
Reconsider or remove the card if a verified full revision requires unauthorized reads or actions, or if missing proof repeatedly becomes confident promises. Documented skill-only improvements across consented cases could justify a controlled later evaluation—not immediate installation or publication. Stop before install advice until the intended revision, full resources and host/data scope are reviewed; stop before external release until named reviewers approve it.
Current catalogue · Browse by job · Research library · Run 14 first route