Nobody sets out to write a prohibited claim. What happens instead is that a description is assembled from three sources, two of them describing a neighbouring model, and the resulting paragraph asserts something no document about this device ever said. It reads well. That is the problem.
The Labeling Is the Ceiling
A regulated device is cleared or approved for stated intended uses. The labeling that came out of that process defines what the device is for, who it is for, and under what conditions. Promotional material is measured against that document. It can say less. It cannot say more, and it cannot say something adjacent that a reader would take as more.
This is why an accurate sentence can still be a problem. A specification that is genuinely true, placed beside an implication about clinical outcome that the labeling does not support, produces a claim the labeling does not carry. The individual facts survive scrutiny. The paragraph does not.
Three Things That Happen to a Claim
The distinction between the second and third columns is the one that matters most and is understood least. An unsupported claim is a gap: find the document and it may become publishable. A prohibited claim is not a gap. Producing better evidence for it does not help, because the objection was never evidential.
The Categories That Need Care
- Intended use and indications
- Stay inside what the labeling states. A use that is common in practice but absent from the labeling is off-label, and promoting it is off-label promotion whether or not the practice itself is lawful.
- Comparative superiority
- Claiming a device is better, faster, safer or more accurate than a named or implied competitor calls for head to head testing. An inference drawn from two specification sheets is not head to head testing.
- Safety and outcome
- Statements about clinical outcomes, patient benefit or safety performance belong to the labeling and the evidence behind it. They are not available to be improved upon in a product description.
- Regulatory status
- Cleared, approved, registered and listed mean specific and different things. Using them loosely, or applying one device's status to another, is a misstatement about the device rather than a wording preference.
- Authorisation and endorsement
- Implying an OEM relationship, authorisation, endorsement or certification you do not hold is a claim about your business rather than the device, and it is checked the same way.
- Certification and compliance
- A certification is either held or not. A statement that a device or a company complies with a standard is a factual assertion with a document behind it, or it is nothing.
Resellers Are Not Outside This
Reviewing Claims Without Reviewing Prose
Reviewing a product description by reading it is the wrong shape of work. Prose hides claims inside sentences that flow, and a reviewer reading for quality will approve a paragraph whose third clause asserts something unsupported, because the paragraph reads well and reading well is what they were checking.
Split the Draft into Claims
One assertion per line. A sentence with three assertions in it becomes three lines, because they will not all have the same answer.
Ask Each Line for Its Source
Not a general source for the section. The specific passage that states this specific thing. If nobody can produce it, the line is unsupported.
Check the Source Describes This Device
This is where sibling products do their damage. A datasheet for the adjacent model in the range is a real document that supports a real claim about a different product.
Screen for the Prohibited Categories
Separately from evidence, and after it. A well sourced comparative superiority claim is still a comparative superiority claim.
Record the Decision, Not the Rewrite
Rejecting with a reason teaches the next draft something. Silently rewriting teaches it nothing, and you will review the same defect again next month.
Why Generative Tools Make This Harder
A language model asked to describe a product will produce a complete, confident, well structured description whether or not it has anything to describe it from. Fluency is what these systems are good at, and fluency is precisely the quality that makes an unsupported claim hard to spot. The failure mode is not obvious nonsense. It is a plausible specification, in the right units, in the right place, that no document anywhere supports.
This is why GearDock checks each drafted claim against the attached documents before a person sees it, and withholds the ones nothing supports rather than softening them into vaguer language. A hedged claim is still a claim. The reviewer's job then becomes judgment about the claims that survived, rather than a search for the ones that should never have been written.
One Thing This Page Is Not