Skip to content

Press-Release Claim Review

Review — instantiates Summary-Substance Alignment Audit

Reads a promotional summary — a press release or announcement — against the study or report it publicizes, grading each headline claim as supported, overstated, or unsupported with the author's incentive to amplify held in view.

A Press-Release Claim Review takes the promotional summary of a piece of work — a press release, launch announcement, or "key findings" blurb — and reads it against the study, report, or dataset it publicizes, grading every headline claim as supported, overstated, or unsupported by the source. What makes it this review and not a generic alignment check is that it treats the summary's author as an interested party: a communications office, marketing team, or principal investigator is rewarded for reach and impact, so the review reads the release the way one reads an advocate's brief — expecting amplification exactly where a stronger claim would travel further — and maps that amplification claim by claim.

Example

A university communications office drafts a release headlined "New study shows daily walking cuts dementia risk by a third." The Press-Release Claim Review pulls the underlying paper and lays the two side by side. The paper reports an association in an observational cohort, strongest in a subgroup over 70, with the authors explicitly cautioning against a causal reading. The review grades each claim: "shows … cuts" → overstated (an association reported as causation); "by a third" → supported but stripped of scope (true only in one subgroup, presented as general); "daily walking" → unsupported specificity (the paper measured total activity, not a walking protocol). It also records the incentive — the release is timed to a fundraising campaign, which is exactly the pressure that turns "associated with" into "cuts." The output is a marked-up release with each claim tagged and a required rewrite for the two that cross the line, delivered before distribution rather than after a correction.

How it works

The distinguishing move is that the review does not merely compare wording — it extracts the source's actual claim set and forces every promotional claim to map onto one of them, returning one of three verdicts (supported / overstated / unsupported), and it annotates each overstatement with the incentive that likely produced it. A promotional claim with no anchor in the source is the review's primary catch. Detecting one specific distortion — certainty and causal inflation — is deep enough to be a sibling specialty of its own; this review is the whole-release, incentive-aware pass that decides which claims ship.

Tuning parameters

  • Verdict granularity — three coarse buckets or a finer scale (e.g. "directionally true, scope-inflated"). Finer catches soft overstatement but slows the review and invites argument.
  • Incentive weighting — how hard to scrutinise claims that happen to serve the author's interest. Turning it up catches self-serving spin but risks treating every strong-but-true claim as suspect.
  • Source of truth — full paper, author interview, or raw dataset. The stronger the source, the more fidelity (not just consistency) can be checked.
  • Reviewer independence — self-review, peer, or an independent party with no stake in reach; the further from the author's incentive, the more trustworthy the verdicts.
  • Scope — headline only, or every sentence and the inserted quotes; wider scope catches buried overstatement at higher cost.

When it helps, and when it misleads

Its strength is catching amplification at the moment of publication, when a fix is cheapest, and naming the incentive so the fix reads as structural rather than personal. Its failure modes: a reviewer without the source can only check internal consistency, not fidelity; and a captured reviewer — same team, same incentive — rubber-stamps. The classic misuse is running it backwards: assembling the review after the release is drafted to manufacture a defence of wording already chosen, rather than to test it. The discipline that guards against this is an independent reviewer holding the source, with the authority to block, acting before distribution. Skipping the review altogether feeds churnalism — the reproduction of a press release as news with no independent check.[1]

How it implements the components

Press-Release Claim Review fills the extract-and-grade side of the archetype, weighted for a promotional author:

  • substantive_surface_claim_set — extracts what the source actually claims, the yardstick every promotional claim is graded against.
  • claim_alignment_matrix — maps each release claim onto a source claim (or onto nothing) and records the supported / overstated / unsupported verdict.
  • author_incentive_and_review_map — records who wrote the release and what they are rewarded for, so amplification is read as expected rather than incidental.

It does not scan for the qualifiers a claim quietly sheds — that granular omission check is Qualifier-Drop Scan — nor build the standalone claim-by-claim provenance grid (Summary-Claim Traceability Matrix), nor test what a release-only reader concludes (Summary-Only Reader Test).

Notes

The review checks the release's fidelity to its source, not the source's own validity — a faithful summary of a weak study is still faithful. Keeping "is the release true to the paper" separate from "is the paper any good" is what stops the review from sprawling into peer review; the second question belongs to a different process entirely.

References

[1] Churnalism — the practice of publishing press-release material as journalism with little or no independent verification. It is the downstream cost of an absent claim review, and the reason the review's authority to block before distribution matters. ↩