A new color can need a new product description

We compared colors within the same archived product family and market, using records collected on the same date. A family counted once, whether it contained two observed colors or ten. That keeps a product with many colors from dominating the result simply because it has more listings.

Among 6,136 eligible family-and-market groups, 855 had at least two different combinations of declared base materials and percentages: 13.9%. The other 5,281 groups matched on those material categories and percentages. These groups come from H&M and Uniqlo pages in the UK and Australia captured in March and April 2026.^[1][2]

The differences were not all alike. In 371 groups, at least one material appeared in one observed color’s statement and not another’s. Another 151 kept the same materials but differed by at least five percentage points in one material. The final 333 had smaller percentage differences. A change in the material list took precedence, so these categories do not overlap.

Together, the first two categories account for 522 groups, or 8.5% of the eligible total. Five percentage points is a descriptive threshold chosen for this analysis, not a rule about quality, labeling or acceptable error. Even a smaller difference matters if a caption repeats an exact percentage.

The unit also matters. These are 6,136 product-family-and-market comparisons, not 6,136 independent global styles or purchases. A family appearing in both countries contributes two groups. The 855 changed groups represent 782 distinct brand-and-family identifiers across the archive.

The archived Australian H&M records for a tailored linen-blend pants family make the distinction concrete. On March 30, 2026, its white version listed a shell of 68% viscose and 32% linen. The dark gray version listed 43% viscose, 31% linen and 26% cotton. Both appeared within the same derived product family, but cotton appeared in only the second of those two shell statements.^[3]

The other observed colors added detail to the picture. Black listed 69% viscose and 31% linen. Light beige and dark brown each listed 45% viscose, 30% linen and 25% cotton. All five separately listed a pocket lining of 100% polyester. The comparison therefore concerns the shell; it does not turn the lining into part of the shell formula.

For a creator, “linen blend” and an exact composition sentence carry different levels of detail. The broad phrase may remain relevant across these archived variants, while “68% viscose and 32% linen” belongs to the white shell statement. Copying that exact sentence into a post showing the dark gray version would lose the difference visible in the source.

The example identifies a possible handoff problem; it does not show that a campaign made that mistake. A brand preparing a brief now can pair the selected color with approved wording from its current product records. That gives the creator an exact source to use instead of relying on the shared product name or the historical numbers shown here.

A different example comes from Uniqlo’s Australian archive. A jersey-jacket family had body statements of 68% cotton, 27% polyester and 5% elastane for beige and red, compared with 74% cotton, 22% polyester and 4% elastane for blue. The material names matched, but cotton differed by six percentage points.^[4]

That distinction matters when someone checks only for familiar words. Seeing cotton, polyester and elastane in both versions is not enough to approve a sentence that includes percentages. The numbers need to follow the selected color too. Our comparison treated six percentage points as 74 minus 68, rather than describing it as a 6% relative change.

The same archive also contains a smaller boundary case: two Australian H&M suit-pants colors listed shells of 65% polyester, 33% viscose and 2% elastane, versus 70%, 28% and 2%. The largest change was exactly five percentage points. Their pocket linings were recorded separately and were not included in that shell calculation.^[3]

None of these differences tells us why a formula changed or whether one version performs better. Manufacturing, sourcing and physical testing are outside the records we analyzed. The useful conclusion for product copy is narrower: a matching list of material names does not establish a matching composition sentence.

The 5,281 matching groups are not a pool of product descriptions that can automatically be reused. We compared one clearly defined statement per color: a complete standalone material list, or a named main, body or shell component. Other parts can still differ.

For example, six colors or designs in an archived Australian H&M five-pack underwear family all listed a shell of 95% cotton and 5% elastane. Yet the accompanying lining statements differed. One front lining repeated 95% cotton and 5% elastane; another front lining listed 100% cotton; other variants named a separate crotch lining of 100% cotton.^[3]

The shell comparison correctly counts that family as matching. It would be incorrect to expand the result into “all material text is identical.” It would be equally incorrect to assume that differently named lining parts describe the same physical area. Keep the component name attached to the statement when sharing it.

Our analysis also grouped materials into fixed base categories. It did not measure every distinction in sourcing, fiber grade, finish, recycled-content wording or marketing language. A match on those base categories and percentages is a limited result. It does not approve a sustainability claim, a care instruction or a description of how the product feels.

For a business, this suggests a useful division of work. Shared copy can explain the product’s overall story when the brand has confirmed it applies. Color-specific facts should remain attached to the relevant version. A creator should be able to see which sentences are shared and which need checking when the visual changes.

The combined result comes from four different source sites, and their records do not contribute equally. The table shows both the eligible comparisons and how much of each initial same-day comparison frame survived the full text checks. That coverage is part of interpreting each rate.

Across all four sites, 7,324 groups passed the initial identity, color and date checks. Of those, 6,136, or 83.8%, met the complete material-text rules. We excluded the other 1,188 rather than repair ambiguous wording, ignore an alternative formula or treat an unresolved group as a match.

Coverage was lower for the Uniqlo records. Their source text could contain shared statements, defaults and separate color or product branches. An additional check limited Uniqlo to explicitly assigned colors and retained only 39 groups across both markets, below our chosen minimum of 50 for interpreting that sensitivity. It cannot establish a stable Uniqlo rate.

That is why the four percentages should not be read as a quality league table. Source structure, retained products and disclosure forms differ. The archive is not a representative sample of all retailers or a measurement of current catalogs. Its contribution is the observed within-family comparison and the concrete situations it reveals.

Imagine a brand briefing a creator on a cream top, then replacing it with the navy version shortly before filming. The photo and product link get updated. The caption draft still contains the cream version’s exact material sentence. This is an illustrative workflow, not an error observed in the dataset.

The first useful check is identity: which product, market and color will actually appear? The next is the statement itself. Open the current source for that version, retain the component name, and compare every material and percentage used in the draft. If the source is incomplete or conflicts with the supplied brief, ask the product owner to resolve that point.

The brand can make this easier with a short source note beside the approved copy. A product link alone may open a default color or change after the brief is shared. Record the product identifier, selected color, market, source date and relevant material statement. That lets the creator understand exactly what the sentence refers to.

For a video, check spoken language, on-screen text and the caption together. For a carousel, check the text beside each color. For a paid ad, the same discipline applies to the asset and the copy approved for that variant. These are recommended review steps; the study does not measure how often those channels contain errors or what a correction does to sales.

For the material-copy check, revisit the claims that depend on the selected color. A confirmed sentence about the overall design may remain usable. An exact material formula needs its own comparison. Keeping those decisions separate makes the review more specific than asking someone to approve the entire brief again without saying what changed.

Once the wording is approved, keep the source reference with the finished content. If a later edit substitutes another color, the next person can see what needs to be revisited. This makes the handoff useful to a small business, an in-house content team and an independent creator working from the same product information.

Use this sheet for the exact variant shown in a post or video. It separates a source fact from the sentence you intend to publish. The download includes a blank version and a worked example based on the archived linen-blend records.^[5]

Product: brand ______; product identifier ______; selected color or design ______; market ______; source link or document ______; checked on ______.

Material statement: component name exactly as supplied ______; material names and percentages ______; other parts mentioned separately ______; person responsible for the current product information ______.

Proposed content: asset showing this variant ______; exact material sentence ______; where it appears, including spoken or on-screen wording ______; source reference ______.

If switching variants: previous color ______; new color ______; materials added or absent ______; percentages changed ______; component names changed ______; revised sentence ______.

Decision: wording matches the approved current source / clarification needed / remove the unsupported material sentence. Reviewer ______; date ______; final content reference ______.

In the historical linen-blend example, a white-to-dark-gray change would replace the shell sentence “68% viscose and 32% linen” with “43% viscose, 31% linen and 26% cotton.” That is an illustration of matching copy to the archived source, not approval to advertise a current product with those numbers.

If the campaign only needs a broad description, ask the brand which wording applies across the selected variants. Do not introduce a precise percentage merely because it appears in an old caption. The sheet should preserve useful specificity while making the source of each specific claim clear.

We used the complete 47,522-record public release created by Kai Li and Grit Walther. The archive contains product identifiers, markets, colors, collection timestamps, preserved material text and structured component records. We checked the published file checksum before analysis.^[1]

This is an exploratory analysis. An initial source audit exposed ambiguity in the producer-parsed text and a product-category mismatch. We then fixed stricter complete-text rules before recalculating the revised results. We kept the source-defined product scope, which can include sets, multipacks and misclassified non-apparel objects. We do not describe the denominator as individual garments.

For each brand, market and derived family identifier, all retained colors had to share a recorded collection date and a comparable disclosure form. We required a fully read standalone material list or one eligible named main, body or shell statement, totaling exactly 100%. Every original record and every preserved branch had to pass the source checks. Missing colors were not invented.

The matching comparison used fixed material categories and exact decimal percentages. It ignored list order, but it did not silently discard extra text, add a missing component heading or combine different component bases within a family. Preserved text and the producer’s structured components had to agree. The package documents each exclusion stage and the four disclosure bases separately.^[5]

Two independently written calculations agreed on all included groups and results. A deterministic audit then reviewed 285 families covering 866 original records, including changed, matching and excluded cases. That checks the analysis against the retained source text; it does not verify the original web extraction or the physical products.

Keeping identical literal component headings left the main result unchanged. Allowing different recorded dates increased the comparison to 8,634 groups, of which 1,397 differed, or 16.2%. That broader result mixes date and color variation, so the same-day comparison remains primary. No confidence interval or market-wide estimate is implied by these descriptive archive counts.