Every sentence in a product description is one of two things. It was taken from a record somewhere in the business, or somebody made it up.
That sounds harsher than it is. Plenty of made-up sentences are harmless — a description of how a jacket might be worn is not a claim about anything. The difficulty is that on a page, the two kinds look exactly alike, and a well-written invented sentence looks more authoritative than a clumsy sourced one.
Both kinds of sentence read the same
A statement of fact has a recognizable shape: plain, declarative, specific. Anything written in that shape inherits its credibility, whether or not there is anything behind it.
The invented sentences also tend to be the better-written ones, which is worth sitting with. A sourced value is awkward — it names a fiber blend, a measurement, a wash temperature — and it sits in a sentence like a stone. An unsourced sentence is free to be graceful, because nothing is constraining its shape. So within a single paragraph, prose quality runs roughly opposite to groundedness, and a reader skimming for the parts that read well is skimming toward the parts with nothing behind them.
This is why reviewing copy by reading it does not separate the two. A reviewer moving through a paragraph encounters twelve confident sentences and has no signal distinguishing the four that came from a test report from the eight that came from somebody’s sense of what the garment is like. The mechanism, and why careful reading structurally fails against it, has its own treatment: what generated copy can and cannot claim.
What follows is the other half of that problem. If reading cannot sort the sentences, the sorting has to happen before they are written, by knowing which fields have somewhere to come from.
Every claim has a source document or it has none
The useful exercise is to take the claims your descriptions actually make and write next to each one the document it comes from and the person who owns that document. Most of them have both, and they are not the ones that cause trouble.
• Fiber content and composition come from the bill of materials, owned by whoever runs development or sourcing: where an apparel BOM actually lives.
• Care and washing instructions come from test results, owned by quality.
• Measurements and points of measure come from the size chart and the specification, owned by the pattern side.
• The color name comes from the colorway record, owned by design or merchandising.
• Country of manufacture comes from supply chain records, owned by sourcing.
• Trim and hardware details come from the trim card, owned jointly and therefore precariously.
Written out this way, a description stops being prose and becomes an assembly of fields, most of which somebody already maintains for other reasons. The copy step is then transcription rather than authorship, which is less glamorous and much more defensible.
The one field that is a decision, not a measurement
Fit intent sits apart from everything above, and mixing it in causes half the confusion in this area.
Whether a garment is meant to be close-fitting or relaxed is not a test result. Nobody measured it; somebody decided it, at the design stage, and the decision is real and worth stating. Its source is a design intent rather than a document of record, and it can be stated honestly as long as it is stated as what it is.
The failure is when intent gets written in the register of measurement. “Relaxed through the body” describes a decision. “Runs true to size” describes an outcome across many bodies that nobody has measured, phrased as though somebody had. The first is fine. The second is a claim about a population, issued without a population.
The risk is not in the hard fields — it is in the unowned ones
Here is the part worth carrying away.
Fields with a source get corrected. If a description states the wrong fiber content, someone in development notices, because it is their number and they recognize it being wrong. Ownership produces correction as a side effect, without anyone designing a review.
Unowned fields are never corrected, because no one reading them believes the sentence is theirs. Occasion suitability, styling suggestions, what the piece pairs with, how it will make the wearer feel, why the fabric was chosen — these have no document, no owner, and no mechanism that would flag them. They sit in every description and they are invisible to exactly the people capable of catching an error.
Generated copy does not fail evenly across a description. It fails precisely in the unowned fields, because those are the ones with nothing constraining them. The sourced fields are pinned by values it was given; the unowned ones are open, and open is where invention goes. What comes back is a paragraph whose factual parts are correct and whose free parts are confident fiction, arranged in the same voice.
Unowned fields also absorb most of the review attention, which completes the inversion. They are the parts everyone has an opinion about — whether the tone is right, whether the styling suggestion fits the brand — so a review meeting spends its time there and moves quickly past the fiber content, which nobody enjoys discussing and which is the part that could be checked. Attention and checkability point in opposite directions, and attention wins.
The same structure appears on the image side, where a picture settles questions it has no evidence for, and for the same reason — nothing in the artifact marks which parts are supported.
A sourced field is only correct on the day it was taken
Pinning a claim to a document solves half of the problem. The other half is that documents change and descriptions do not.
A fabric is substituted mid-season and the composition line is now wrong. A colorway is dropped and the color name refers to something nobody can buy. A supplier changes and the country of manufacture changes with it. In each case the description was correct when it was written, nothing about it was careless, and it is now false — with no event anywhere marking the transition.
The reason is that copy is written once, at launch, by someone assembling values for that moment. Nothing connects the published sentence back to the record it came from, so a change in the record has no path to the page. This is the same structure that leaves images serving after the garment they show has changed, and it has the same remedy: the value should carry a reference to its source rather than a copy of it, so that a change in one place can reach the other.
Where that link cannot be built, the fallback is a review triggered by the change rather than by a date. A fabric substitution should reach whoever owns the description, in the same message that reaches production, because a calendar-based copy review will not fire at the moment that matters.
Turning the list into a brief
The practical consequence is that a copy brief should carry values, not adjectives.
A brief listing the sourced fields and their current values gives a generation step something to be constrained by. A brief describing the tone gives it nothing to be constrained by, so it fills the whole description with the only thing it can produce unaided, which is plausible language. The difference in output is not subtle and it costs nothing except assembling the values once — which is normally a lookup rather than work, since the fields already exist somewhere.
For fields with no source, the brief should say so explicitly and state what they are: this is a design intent, this is a styling suggestion, this is not a measurement. Marked that way they can be written without pretending to be facts, and a reviewer reading the output knows which sentences they are entitled to check. That marked brief is what should be handed to the step that drafts the copy, rather than a garment name and a tone of voice.
Anything touching labeling, origin, or sustainability claims should go through whoever owns compliance before publication, against current requirements for the markets you sell in. Those fields have sources, and they also have consequences this article does not attempt to assess.
Questions teams ask about product copy claims
Which fields can a description safely state?
The ones where somebody can point to the document the value came from. That test is quick and it does not require judgment about wording. If nobody can name the source in a few seconds, the sentence is either an intent that should be labeled as one or an invention that should be removed.
Is a styling suggestion a claim?
Not in the same way, since it proposes rather than asserts, and it is a normal part of a description. What makes it a problem is writing it in the grammar of a fact, at which point a reader cannot tell it from a specification. Keeping the two registers distinct solves this without removing anything useful.
Our copy says “runs true to size.” Is that a problem?
It is a statement about how a garment behaves across many bodies, and it is usually made without anyone having examined many bodies. If you have fit data supporting it, it has a source and is fine. If it was written because the garment feels normal, it is a claim about a population issued without one.
How do we stop generated copy from inventing things?
Give it the values rather than a description of the tone, because invention happens where nothing is specified. A brief carrying fiber content, measurements, care values, and the color name constrains most of a description. What remains unconstrained will still be invented, which is why the brief should also mark which parts are intent.
Who should own the description?
Somebody has to own the assembly of fields, which is a different job from writing. The field values are owned by the departments that generate them and should not be re-entered by anyone else. The common failure is treating the whole description as a copywriting deliverable, which detaches every value from the person who could verify it.
What about claims like sustainable or ethically made?
Those have sources — certifications, audits, supply chain records — and they also carry consequences beyond accuracy. Route them through whoever owns compliance for your markets rather than treating them as copy decisions. This article deliberately does not assess what any particular claim requires.
Where this leaves you
A product description is an assembly of fields, most of which somebody in the business already maintains. Written that way, the copy step becomes transcription, and the sentences that matter are pinned to documents with owners who will notice if they go wrong. What remains is a small set of fields with no source and no owner — styling, occasion, intent — and those are exactly where invention concentrates, because nothing there is holding it. Mark them as what they are, give the sourced values to whoever drafts, and stop asking a writing step to settle facts it has no access to.
Write the source next to each claim
Take one product description and break it into individual claims. Next to each, write the document the value came from and the person who owns that document. Most lines will fill in quickly. The lines that stay empty are not writing problems — they are claims with nothing behind them, and they are where generated copy will drift every time. Decide for each whether it should be removed, or restated plainly as an intent rather than a fact.
Written by
