Editorial Policy

Who writes GLP Watchdog

Two editors, two beats. Ruth Calder covers enforcement: FDA warning letters, state pharmacy-board licensing and discipline, 503A and 503B registration, and the corporate records that connect a brand to whoever is actually behind it. Nora Bissett covers pricing — what a month of treatment costs once the billing cycle is applied, which is frequently not the number on the seller’s page.

Nothing about either byline is embellished. Neither editor holds a medical, pharmacy, academic, or pharmaceutical-industry credential, and we make no claim that they do. No award, no post, no patent, no professional body. What each byline stands behind is the documentary record cited in the piece — a letter, a registration, a filing, a dated price capture — which you can pull yourself without taking our word for any of it.

Neither editor is a licensed clinician, and none of this is medical advice. That disclosure sits on every clinical page rather than only here, because the place a reader needs to know what the writing rests on is the page they landed on.

We are actively seeking a credentialed medical reviewer (MD, PharmD, NP, RN, or RD) to join the editorial team on a freelance / consultant basis. When that role is filled, the byline on every clinical article will switch from “Editorially reviewed (not clinically reviewed)” to “Medically reviewed by [Name, Credentials]” automatically — the infrastructure is already in place. If you are a credentialed clinician interested in this role, see the open posting on our Careers page for details and how to apply.

The 125% standard

One source is not enough to publish a claim on. A clinical or pricing statement has to hold up against a second, independent one, and where the two disagree we either work the discrepancy out on the page or hold the claim back. We call it the 125% standard, which is a deliberately awkward name for aiming past sufficient.

Two places where it costs us something, which is the only kind of standard worth describing. Citations: a build gate asks PubMed whether every ID we print really belongs to the paper we printed it under, and the build fails until each one comes back matched. That check exists because the classic failure of written-at-volume health copy is a real identifier sitting under a real but different study, which looks more trustworthy than no citation at all.

Prices: when a sweep finds a seller’s live figure but nothing independent corroborates it, we leave the old figure in place and flag the row for a person, even where the stored number is almost certainly wrong. A conspicuously stale price we can date beats a fresh one we cannot stand behind.

How we source clinical claims

The acceptable source hierarchy for clinical claims:

1. FDA prescribing information (the official drug label) — the highest-trust source for approved indications, dosing, contraindications, and warnings.
2. Peer-reviewed primary research indexed in PubMed, cited by PMID, wherever the trial evidence runs ahead of what the label says.
3. Published regulatory filings — FDA warning letters (with the actual letter URL on fda.gov), DOJ press releases, FTC enforcement orders, court records (PACER, CourtListener) for litigation claims.
4. Clinicaltrials.gov — for trial design, enrollment, and primary endpoints.

Not acceptable, for a clinical claim: a company press release, a seller’s own blog, another comparison site, a listicle, a social post, an AI summary, or an earlier article of ours. Citing yourself is not citing anything.

Where the seller data comes from

What a company charges, which states it serves, which drugs it lists, what it claims to be accredited for, and which pharmacy fills the order all come off that company’s own live pages on a recorded date. Around that:

• The URLs we read are kept on the record itself, in verification.source_urls, so a claim can be traced back to the page it came off
• Each review page shows the month its record was last checked, alongside a list of what that check actually covered
• A confidence tier of high, medium or low says how much of a company’s account of itself we could corroborate. A seller that will not publish its state list gets marked low rather than filled in from guesswork, and that mark drags its score down
• An automated pricing sweep runs monthly and a person merges the result. It stamps a date only where the live page agreed with the stored figure

Where AI sits in the work

It is in the workflow, and pretending otherwise would be a poor start for a page about honesty. Drafting, formatting, structured data, and a good deal of the checking machinery run with AI assistance. Nothing reaches a reader that one of the two editors above has not read line by line and signed off on under their own byline.

What it is put to work on:

• A first draft from an outline, which the editor then rewrites and verifies. The draft is raw material, never the published article
• Checking citations at scale — that a PMID resolves to the paper named beside it, that an fda.gov link returns the letter we said it returns, that a docket exists
• Keeping structured data in step with the seller and author registries, so the schema a search engine reads matches the page a person reads
• Catching contradictions between one page and another, which is the failure a human editor is worst at spotting across hundreds of documents

What it is never put to work on:

• Producing a clinical fact, a trial figure, a price, or a state list. Those are read off a source or they do not exist
• Writing the final wording of a clinical claim that no human has checked against the source
• Producing a citation of any kind. A PMID or a letter URL is looked up live and copied, never recalled

Google’s own position is that AI-assisted work is fine when it is good and genuinely checked, and is spam when it is mass-produced without either. We agree with that, and the rules above are how we intend to stay on the right side of it.

The rubric behind a score

Six dimensions carry fixed weights. A quarter each for value and effectiveness, 15% each for user experience and trust & safety, a tenth each for accessibility and ongoing support. The per-dimension breakdown is printed on every review page rather than kept internal, so you can see which of the six is holding a company up or dragging it down. The weights and the caps applied to them are set out at /methodology.

Independence, and the exact shape of the conflict

Commissions from a subset of outbound links are what pay for this. A score is not for sale and never has been. Order sometimes is, and we would rather write down which is which than let you find out.

What the money does not touch:

• The rubric. Whether a company pays us is not an input to any of the six dimensions, and the uniform evidence caps run across the whole register at once, blind to commercial status
• What we publish about a partner. FDA enforcement, live litigation and a bad complaint record are written up on partner and non-partner alike. The Direct Meds and Zealthy reviews are the ones to check us on
• Whether a company gets covered. Sellers we earn nothing from are researched and reviewed at the same length, and the A–Z index is alphabetical

What it does touch: the order of the reviews directory and the comparison tool, the head of the high-intent boards, and the partner cards on the homepage. In the directory and on the boards that marker sits on the listing itself rather than buried in the link; the comparison tool discloses once at the page level. The full accounting is at /disclosure, and the mechanics are at /methodology. When we get something wrong, the fix goes out with a visible stamp on the page it was wrong on, under the policy at /corrections, whether or not the company involved is paying us.

Lines we will not cross

• No clinical statement goes out without a document behind it
• Nothing publishes that a human editor has not read and verified
• A wrong claim gets a correction stamped on the page it appeared on, in the format set out at /corrections, rather than being quietly edited out of existence
• No payment moves a score or a rating, in either direction
• Our own pages are never the source for a fact they were never the source of
• No credential gets implied that neither editor holds

Editorial contact

Disputes, source requests and corrections go to editorial@glpwatchdog.org. Send the specific claim and where you found the better evidence. That gets a faster answer than anything else will.