Key takeaways
- Treat the claim—not the page—as the governed unit so evidence follows every reuse.
- Record source, period, population, method, qualifier, approved wording, owner, expiry, and publication surfaces.
- Require stronger review for performance, comparative, safety, financial, and customer-result claims.
- Withdraw or correct controlled copies when the evidence or offer changes.
01
The dangerous sentence is often the one that sounds completely ordinary
A model turns “18 of 31 respondents in one pilot” into “most customers prefer.” A writer removes a date to make the copy smoother. A sales deck borrows the sentence and loses the footnote. Nobody typed a deliberate lie, but a narrow observation has become a portable fact.
Create a claim record when a material factual statement first enters production. Give it an ID, exact wording, claim type, product or service, market, audience, evidence, period, population, method, denominator, limitations, owner, reviewer, approved variants, expiry, and every controlled place it appears.
Swipe to compare every column
| Claim type | Evidence question | Review trigger |
|---|---|---|
| Performance | Measured against what baseline and period? | Method, product, or conditions change |
| Comparison | Is the competitor set and basis current? | Market or compared offer changes |
| Customer result | Typical, exceptional, or selected case? | Permission, context, or data changes |
| Forecast | Which assumptions drive the range? | Input, model, or decision horizon changes |
02
Separate a source from support
A link is not proof that the page supports the wording. Reviewers should be able to locate the exact table, passage, dataset, interview segment, or calculation. Record what transformation connected the source to the claim and who approved that interpretation.
W3C provenance guidance is useful here: keep the entities, activities, derivations, and responsible agents in the chain. Preserve the dated evidence snapshot behind a published claim so a later source update does not silently rewrite history.
03
Constrain generation before review becomes a bottleneck
Expose approved claims and their conditions as structured inputs. Tell the writing workflow that it may simplify syntax but not widen population, geography, certainty, causation, comparison, or time. Block expired records and unsupported numerical claims from automatic publication.
Route higher-risk statements to someone qualified to judge the underlying evidence. The reviewer should see the claim and support together, not hunt across folders while the campaign waits. Legal requirements vary by market and subject; this workflow supports review but does not replace it.
04
Make correction part of distribution
Track pages, ads, emails, decks, feeds, partner kits, and generated-answer sources that reuse the claim. When a source is corrected, an offer changes, or approval expires, open correction tasks for those surfaces. Preserve what changed and why.
Measure unsupported-claim rate, expired claims in production, review turnaround, correction propagation time, and repeated failure patterns. The goal is not to sterilize the copy. It is to let a human voice carry evidence without quietly upgrading it.
Primary sources and further reading
Use the source material to validate details against your own context and current platform configuration.
This field note follows the XenGrowth editorial policy: primary sources where available, visible limitations, material review dates, and no invented first-hand experience.
Stay with the problem
