| name | gsj-product-accuracy |
| description | Fact-checks Groundswell (GSJ) product and marketing copy against the approved/banned claims tracker before it ships. Use whenever GSJ copy makes a factual, health, sourcing, sustainability, packaging/compostability, or community-impact claim — health effects of ingredients (tart cherry, beet, ginger, turmeric), "clinically proven" or "guaranteed"-style wording, compostable/recyclable/eco/sustainable claims, city counts or pounds-diverted figures, farm-sourcing claims, or competitor comparisons — in emails, social captions, packaging, web, ads, or customer-service replies. Run before publishing, in addition to gsj-brand-voice. This skill governs whether a claim is substantiated; brand-voice governs how it is phrased. |
Groundswell Juice Co. — product accuracy
Groundswell copy has shipped unsupported claims before — the "clinically proven" line in the cup
email that legal flagged. This skill is the safety net that runs before copy goes out: it checks
every factual claim against the claims tracker, so nothing ships that the brand can't back up.
The rule that makes this work: a claim is either on the approved list, on the banned list, or
unverified. There is no fourth category called "sounds fine." Plausibility is never a reason to
approve. If a claim isn't in the tracker, it is blocked until someone adds it with a proof source.
What this skill does, and its boundary with brand-voice
This skill answers one question about each claim: is it substantiated? — meaning it matches an
approved row in the tracker, stays within that row's limits, and carries its required caveat.
It is the factual counterpart to [[gsj-brand-voice]], which governs how copy is phrased (tone,
register, banned hype words, punctuation). The two overlap on a few terms — "clinically proven,"
"eco-friendly," "detox" are called out in both — but they judge different things: brand-voice asks
does this sound like Groundswell, this skill asks can we prove it. When copy is going out,
run both. A line can be perfectly on-voice and still make an unverified claim, and vice versa.
The tracker is the only authority
At the start of every review, read references/gsj-claims-tracker.md. That file (a mirror of
the marketing/legal-owned gsj-claims-tracker.xlsx) is the source of truth for what is approved and
what is banned. It changes over time.
- Never approve a claim from memory, or from the examples in this file. The examples below are
illustrations of the process, not a substitute for reading the current tracker.
- Use
references/gsj-product-facts.pdf only as supporting evidence for why a tracker row is
approved (formulation, sourcing, packaging, community facts). It is not an independent approval
source — a fact appearing there does not make a claim shippable unless the tracker approves it.
Process — check the copy step by step
Work through every claim in the copy in this order.
-
Extract every claim. Read the copy and list each factual assertion separately — ingredient/
health effects, sourcing, sustainability and packaging (compostable/recyclable), community-impact
figures (pounds, cities, events), and any competitor comparison. Quote each one exactly. If a
sentence bundles two claims, split them.
-
Match against the Approved list by meaning, not wording. Copy paraphrases; the tracker won't
contain the exact sentence. Match on intent. "Beets help your blood flow" maps to A2.
-
Check for banned wording. Scan for the banned terms and close variants (B1–B5): "clinically
proven," unqualified "eco-friendly"/"100% sustainable," "detox"/"cleanse," "guaranteed," "all
cities compost." If matched → BANNED, and cite the reason on file from the tracker (see the
next section — every ban must carry its reason).
-
Check for upgrades / overreach. An approved claim pushed past its stated limit becomes a
violation. Use the tracker's "Limits" column: A1 "supports recovery" upgraded to "cures soreness"
or "clinically proven" → violation; A5 "compostable in 47 of 55 cities" inflated to "compostable
everywhere" or "most cities" → violation. Flag as UPGRADE/OVERREACH with the limit it broke.
-
Check required caveats. Some approvals are conditional, not standalone:
- A6 recycling claim must be paired with the compost-coverage caveat (47 of 55).
- A5 composting must state the specific number, never "most"/"all."
- A4 sourcing must name the farm and location (Traverse City family farm).
- A7's ~900 lbs figure is Riverside-only — do not apply it to any other city.
A claim missing its required caveat is NEEDS CAVEAT, not approved as written.
-
Anything unmatched → UNVERIFIED (hard block). If a claim matches no approved row, it cannot
ship — regardless of how reasonable it sounds. Do not approve because it "seems fine" or "is
probably true." Emit the exact tracker row that would need to be added: the claim, and what proof
source would be required to approve it. Approving it is the job of marketing/legal updating the
tracker, not of this review.
-
Confirm every ban carries its reason. Each BANNED finding must cite the reason-on-file from
the tracker. If you believe a claim should be banned but no reason is recorded in the tracker,
flag the tracker as incomplete rather than asserting the ban unsupported — the requirement is
symmetric: no ban without a reason, no approval without a proof source.
Verdict for each claim
Assign exactly one:
- APPROVED — matches an approved row, within limits, caveat present (or none required).
- APPROVED — WITH REQUIRED CAVEAT — approved only once the missing caveat is added; state which.
- NEEDS CAVEAT — same as above when the caveat is simply absent from the copy.
- UPGRADE/OVERREACH — based on an approved claim but pushed past its limit; state the limit.
- BANNED — matches a banned row; cite the reason on file. Give the in-voice replacement.
- UNVERIFIED — cannot ship — on neither list; emit the tracker row that would need to be added.
Output format
Return a per-claim table:
| Quoted claim | Verdict | Tracker row / proof source | Required fix |
|---|
Then a one-line gate: SHIPS (all claims approved) or DOES NOT SHIP, followed by the list of
blockers (every BANNED, UPGRADE/OVERREACH, NEEDS CAVEAT, and UNVERIFIED item).
Do not invent proof
Never fabricate a proof source, a number, a farm, or a city count to make a claim passable. This
mirrors the brand-voice invented-fact rule: a fabricated specific is worse than an honest gap,
because the brand's credibility rests on its specifics being real. An unknown specific is
UNVERIFIED, not a guess.
Known gaps to watch (blind spots)
- Golden Hour has no tracker rows. Any health claim about it (turmeric, mango, pineapple) is
automatically UNVERIFIED until marketing adds approved rows. This will recur — flag it every time
rather than reasoning from general knowledge about turmeric.
- Facts are time-bound. "47 of 55 cities, as of this spring" and "~900 lbs last year" drift. If
copy restates a figure, confirm it against the current tracker/product-facts dates; a number that
was right last quarter can be stale now.
- Location generalization. Riverside's partner (Riverside Conservation Group) and its figure are
city-specific. Do not let copy apply either to another location — confirm per city.
- Approved ≠ unconditional. Several approvals only hold with their caveat (steps 4–5). A bare
version of an approved claim is not automatically shippable.