Clothing Product Descriptions: A Checklist of Fabric, Fit and Care Errors

AI-written clothing descriptions most often get fibre content, fit, care instructions, length and sizing wrong, because the tool fills gaps in your input with typical wording. Check each claim against your spec sheet: percentages must total 100, the fit word must match your measurements, and care must match the sewn-in label.

The errors are small and plausible, which is why they survive. "Pure linen" for a linen-viscose blend reads fine. "Machine washable and tumble dry" reads fine. Then the customer shrinks the dress, returns it and writes a review. Five groups of checks catch them: fibre content, fit and sizing, care, model and photo details, and performance claims. Run against a spec sheet, they take about six minutes a product, and the shirt dress further down shows eight such errors in a single draft.

Follow me on Instagram@sagnikteaches

Start with the source: what the AI was given

Nearly every error traces to a missing input. If you give an AI tool a photo and a product name, it will write copy from what dresses of that type usually are. If you give it a spec sheet, it has facts to quote. So the first control is upstream: build one input block per product with these fields, and refuse to write copy without it.

Connect on LinkedInSagnik Bhattacharya

• Fibre composition exactly as it appears on the label, shell and lining separately.
• Fabric weight and construction where relevant (jersey, woven, brushed, 280 gsm).
• Care instructions as the sewn-in label states them.
• Garment measurements by size, and the body measurements each size is meant to fit.
• Length, from a stated point such as the high point of the shoulder.
• Model height and the size worn, per photo.
• Features to claim and features not to claim (pockets, lining, stretch).

If you are producing garments yourself, most of this already exists in the tech pack, as covered in AI tools for clothing brands from sketch to tech pack. If you resell supplier stock, expect the supplier's sheet to be incomplete, and ask for the label photo.

Subscribe on YouTube@sagnikteaches

Group one: fibre content

This is the group where an error does the most harm, because fibre content is both a label requirement in many markets and the first thing a buyer reads. Check these items:

1. The percentages add to 100. A draft that says "70% cotton, 25% polyester, 10% elastane" totals 105. Sum them every time; the AI will not always notice.
2. The fibres appear exactly as on the label. If the label says 55% linen, 45% viscose, the copy cannot say "pure linen" or "linen dress" without the blend.
3. Shell, lining and trims are not merged. "Cotton" for the outer fabric does not describe a polyester lining. Say both.
4. Look-alike words are not promoted. "Linen-look", "cashmere-feel" and "vegan leather" (usually a plastic such as polyurethane) describe appearance, not fibre. Keep them out of the fibre line.
5. Brand names stay the same. Some fibre names are trademarks owned by the producer. If your spec says "lyocell", do not let the copy switch to a brand name you have not licensed or confirmed.
6. Special fibre claims have evidence. Organic, recycled, wool, silk and cashmere each need a certificate, a supplier declaration or a lab test. Without it, the claim comes out.

Fibre labelling rules differ between markets, including how fibres are named and rounded, so check the rules for the places you sell, or ask your testing lab. The point for copy is simple: the website must say what the label says.

Group two: fit and sizing

Fit words are opinions, but they should be opinions your measurements support. "True to size", "oversized", "relaxed" and "slim" mean different things to different customers, which is why they cause returns. Verify these:

1. The fit word matches the ease. Ease is the difference between the garment's measurement and the body measurement. If a size S garment measures 108 cm at the hip and the body hip it is designed for is 92 cm, the ease is 16 cm, which is a relaxed fit. "Oversized" would need a stated intent in the spec.
2. "True to size" has a basis. It needs a size chart and, ideally, fit-test feedback. If you have neither, leave it out.
3. Stretch is stated only when the fabric has elastane or similar. A 100% cotton woven does not stretch much, whatever the copy says.
4. Length is measured, not assumed. "Midi" and "mid-calf" are not measurements. Give the length in centimetres or inches from the shoulder or waist, and state the point.
5. Units are consistent. Convert once, in your sheet: 38 inches is 96.5 cm, and a tool that rounds to 97 in one place and 96 in another will confuse buyers.
6. The size chart matches the copy. If the chart says "chest" and the description says "bust", pick one.

Sizing questions are also what customers ask a shop chatbot. If you automate answers, the same spec fields should feed both, so the page and the bot never disagree; ecommerce chatbot design for order status, returns and sizing covers that side.

Group three: care instructions

Care is a promise the customer acts on. Get it wrong and the garment is ruined, and the refund is yours.

1. Every statement matches the sewn-in label. If the label says wash at 30 degrees, gentle cycle, do not tumble dry, the page cannot say "machine washable and tumble-dryable".
2. No extra permissions. AI tends to add "dry clean" or "iron on low" because most garments allow it. If the label does not say it, the copy does not say it.
3. The symbols are read correctly. The care label code is standardised internationally as ISO 3758, whose latest edition was published in December 2023, and uses five basic symbols: the wash tub, the triangle for bleaching, the square for drying, the iron and the circle for professional care. Dots on the iron give the maximum sole-plate temperature: one dot is low heat (up to 120 degrees), two is moderate (160) and three is hot (210). Ask the AI to translate a symbol line into words, then check each against the standard, never the other way round.
4. Temperature and cycle are both given. "Wash cold" and "wash at 30 degrees, gentle cycle" are different instructions.
5. Special treatments are noted. Wash inside out, wash with similar colours, shape while damp. Include them only if the label or your spec says so.

Group four: model, photos and colours

1. Model height and size worn, per photo. "Model is 175 cm and wears size S" is useful. A missing size worn is a common omission.
2. The model in the text is the model in the shot. If you reuse copy across colours with a different model, update the height and size.
3. Colour names match the product. "Sage" and "olive" may be the same dye lot to you and different to a customer. Use the name on the swatch card.
4. Pattern and print scale are stated if they affect how it looks on different sizes.
5. Images show the claim. If the copy says "two pockets", the photos show them.

Group five: performance and feature claims

AI loves a benefit: breathable, moisture-wicking, wrinkle-free, waterproof, all-season. These claims need support.

• "Waterproof" and "water-resistant" mean different things. Use the one your test or supplier sheet supports and add a rating if you have one.
• "Breathable" for a polyester shell with a coating is a claim for the supplier to back up.
• "Wrinkle-free" must be tested on the actual fabric, not assumed from the fibre.
• "Year-round" or "all-season" is rarely provable. Say what the weight and construction suggest, such as "mid-weight" with a gsm figure.
• Sustainability words (eco, conscious, sustainably made) belong only next to a named certificate or a stated fact, such as "made with 30% recycled polyester, supplier declaration available".

The main example: a dress with eight errors

Take an illustrative independent label with a shirt dress, style SD-210. The spec sheet says: shell 55% linen, 45% viscose; no lining; plastic buttons; relaxed fit; garment hip 108 cm in size S for a body hip of 92 cm; length 98 cm from the shoulder; care: wash at 30 degrees on a gentle cycle, do not bleach, do not tumble dry, iron at medium heat; model 175 cm wearing size S.

The AI draft, given only a photo and the product name, read: "Crafted from 100% pure linen, this oversized shirt dress is true to size and perfect for year-round wear. Fully machine washable and tumble-dryable, and dry cleanable. Falls to the mid-calf. Our model is 175 cm and wears a small. Sustainably made."

The corrected copy reads: "A relaxed shirt dress in a linen-viscose blend (55% linen, 45% viscose), unlined, with plastic buttons. 98 cm long from the shoulder in size S. Wash at 30 degrees on a gentle cycle; do not bleach; do not tumble dry; iron at medium heat. Model is 175 cm and wears size S." It is shorter, plainer and every word is traceable, so a customer who reads it knows what will arrive and how to wash it.

The prompt that makes the AI check its own draft

You can use the tool as the checker, provided it has the spec. Give it both texts and ask for a claim table, which makes it do the comparison line by line.

Below are (A) the spec sheet for a garment and (B) a draft product description.
Extract every factual claim in B (fibre, fit, length, care, performance, model,
sustainability) and compare it with A. For each claim return: the claim, the
matching line in A (quote it), and a verdict: SUPPORTED, CONTRADICTED or NOT IN
SPEC. Do not correct the copy yet. Then check that the fibre percentages total
100 and that care words match the care line exactly.

(A) SPEC:
[paste]

(B) DRAFT:
[paste]

Keeping the correction separate from the check matters. When a tool is asked to do both at once, it tends to polish the wording and quietly keep a claim it should have flagged. Run the check first, fix the contradicted and unsupported claims yourself, then ask the AI to rewrite using only the supported ones. For the wider job of writing store copy that persuades once it is accurate, writing Shopify product descriptions with AI picks up from there.

A size chart before and after the AI touched it

Size charts are where unit slips hide. An illustrative chart in inches, as supplied by a factory, gives chest measurements of S 36, M 38, L 40, XL 42. Asked to "convert to centimetres and add to the page", a tool might answer S 91, M 97, L 102, XL 107. Check them: 36 x 2.54 = 91.4, 38 x 2.54 = 96.5, 40 x 2.54 = 101.6, 42 x 2.54 = 106.7. Every one is within a centimetre, which looks fine until you notice that the steps are 6, 5 and 5 cm in the AI's chart and 5.1 cm each in the true one, and that M has been rounded up from 96.5 to 97 while S has been rounded down. A shopper between sizes is exactly the person who reads that chart carefully.

Do the conversion in a spreadsheet with =ROUND(B2*2.54,1), publish one decimal place or none consistently, and keep the inches column only if you sell to people who use it. More important, state what the figure measures. A chest figure for the body and the same figure for the garment differ by the ease, and a chart that does not say which it is will mislead whichever way the customer reads it. Add the line "garment measurements, laid flat and doubled" or "body measurements" to the chart, whichever your factory supplied, and make sure the AI copies that line instead of dropping it.

Before and after: a performance top

An illustrative running top shows how claims creep. Spec: 88% recycled polyester, 12% elastane, 140 gsm, flatlock seams, supplier declaration for recycled content on file, reflective logo. The AI first draft: "Engineered with breathable, moisture-wicking technology, this top keeps you dry, fresh and odour-free on every run. Made from sustainable recycled materials. Perfect for all conditions."

The claim table finds four problems. Breathable and moisture-wicking are not in the spec, odour-free is an antimicrobial claim that needs a treatment and a test, "sustainable" is vaguer than the declaration supports, and "all conditions" is untestable. The supported rewrite is: "A lightweight running top (140 gsm) in 88% recycled polyester and 12% elastane for stretch, with flatlock seams to reduce rubbing and a reflective logo. The recycled polyester is covered by a supplier declaration." If the supplier can supply a wicking test, add it by name. If not, leave it out and let your photos and reviews carry the story. Customers can tell the difference between a spec and a slogan, and the spec is the one that does not come back as a complaint.

Using photos as input without trusting them

Many tools now write copy from an uploaded product photo. They are good at describing what is visible: colour, neckline, button count, pattern. They are poor at everything that is not visible: fibre, weight, stretch, drape, lining, fit, and the true colour under your shop's lighting. A photo cannot tell a viscose from a polyester satin. Treat a photo-based draft as a description of the picture, and put every invisible fact in from the spec before anything is published. A quick test is to cover the spec and ask whether a stranger could verify each sentence from the photo alone. Sentences that fail the test either come from the spec, with a source, or come out.

A batch routine that takes six minutes a product

1. Minute 1: confirm the input block is complete. If fibre, care or measurements are missing, stop and get them.
2. Minutes 2 and 3: run the claim-table prompt and read every CONTRADICTED and NOT IN SPEC line.
3. Minute 4: check the fibre arithmetic and the unit conversions yourself.
4. Minute 5: compare the care sentence with a photo of the sewn-in label.
5. Minute 6: check the model line against the photos, then publish.

For 40 products that is about four hours, a figure worth setting against the hours your team spends each month on returns and refund emails. The pay-off shows in the numbers you already hold: if you track why garments come back, reading return reasons with AI will show whether "not as described" and "shrank in the wash" fall once the copy is checked.

How one wool claim turned into an honesty complaint

An illustrative online shop sold 120 knitted cardigans, and one description said "100% wool". The supplier's invoice said "wool blend", and the label said 70% wool, 30% acrylic. The AI had read "knit cardigan, warm, soft" and filled in the most common material for the word warm. Eleven customers wrote that it was not pure wool, three returned it, and the shop's wording became a complaint about honesty, not about the product. The fix cost nothing: the input block now has a field for the label text, and the copy is blocked until it is filled in. A second control is a short rule in the prompt: "If a material, measurement or care instruction is not in the spec, write nothing about it."

Three more kinds of business, same checklist

• A children's clothing label adds age ranges and the height or weight a size is intended for. The check is that each age band matches the measurement chart, and that no safety claim, such as flame resistance, appears without the test report.
• A workwear supplier sells to firms, so fibre and performance claims such as high visibility or flame retardance must match the certification on the product, quoted by reference.
• A resale shop describes one-off items, so condition, measurements taken by hand and flaws replace the supplier spec, and the AI is only allowed to rephrase what the person measuring wrote.

The principle across all of them is the one at the top: a description can only be as accurate as its inputs, and an AI tool will fill silence with confident wording. Feed it facts, ask it to separate checking from writing, and keep the sewn-in label and the spec sheet as the final authority. If you want help setting that routine up across a whole catalogue, that is the kind of work my AI implementation consultation covers.

Checking AI clothing copy: common questions

Which details does AI most often get wrong in clothing copy?

Fibre content, care instructions and fit claims. The tool fills gaps with typical wording, so a blend becomes pure cotton, a hand-wash garment becomes machine washable, and a relaxed cut becomes true to size. Anything not stated in your spec sheet is a guess.

Do I need to check every product?

Check every product the first time, because errors come from missing inputs and repeat across a range. After a few batches, spot-check a sample of each batch and always check any product with a special fibre, a performance claim or a new supplier.

Should the description repeat the care label exactly?

It should never contradict it. Use the same temperatures and restrictions as the sewn-in label, in plain words, and keep the symbols themselves for the label or a care image. If label and page disagree, customers and returns teams will notice.

How should I describe model height and size?

State the model's height and the size worn, in the same units as your size chart, for example 175 cm wearing size S. If the shoot used several models or more than one size, say which photo shows which.

Is it safe to let AI write sustainability claims?

No. Claims such as organic, recycled or sustainably made need evidence such as a certificate or a supplier statement, and rules on environmental claims vary by market. Include only what you can document, using the same words as the document.

Further reads

• Taking a Local Shop Online With AI: Product Listings in a Weekend — Get a whole range listed in a weekend, then check it.
• Dropshipping Product Pages: Rewrite Supplier Copy With AI — Rewrite supplier copy without inheriting its errors.
• Ecommerce AI Chatbot Examples: 7 Store Conversations to Copy — Store conversations to copy, including sizing questions.
• AI in Fashion Photography: Lookbook Images for a Small Brand — Keep lookbook images consistent with what you describe.
• Online Shop AI Mistakes That Hurt Trust and Conversions — Other AI slips that cost an online shop its customers' trust.
• AI Marketing for a Clothing Brand: One Drop Planned Start to Finish — Plan a drop around copy that has been checked.
• AI for Marketing Content: A Vintage Clothing Seller's Weekly Drop — One seller's Thursday drop, from a garment sheet to honest condition wording, a scheduled countdown and a photo check before anything goes live.
• ChatGPT Add-On for Boutiques: Which Apps Are Worth Adding? — A verdict table for boutique owners: which ChatGPT apps to add, which to skip, what replaced the stylist GPT, and what shoppers can do in ChatGPT already.
• Mattress Shop Marketing: Comparison Pages, Offers and Reviews With AI — How an independent mattress shop can use AI for a firmness comparison page, trial wording, offers worked through in numbers, and review requests after delivery.
• AI for Fabric Stores: Cut-Length Orders and Fabric Descriptions — A fabric shop's workflow: supplier sheet to checked description, then a metre-based order form and emailed requests turned into cut-length lines.
• Can a Chatbot for a Clothing Store Answer Size and Fit Questions? — The size chart as source, a six-step answer flow, two worked chats, return-policy limits, hand-over triggers and 30 test questions for a clothing store.
• AI Model Photos for a Clothing Brand: Cost per Look and Disclosure — How much do AI fashion model photos really cost per look? Compare studio shoot expenses against synthetic models, fit accuracy, and disclosure rules.
• AI for eBay Descriptions: A Condition, Flaws and Measurements Template — Draft accurate eBay descriptions with AI using an objective condition, flaws, and measurements template that stops buyer returns.
• Bulk Product Description Generator: From CSV Rows to Unique Copy — How to turn a 500-row product CSV into unique descriptions with AI, check for duplicates and invented specs, and import the result into Shopify safely.
• Walmart Marketplace Listing: Draft Titles and Attributes With AI — From a facts sheet to a filled item spec template: prompts for titles, features and attributes, plus the checks that stop upload errors.
• WooCommerce Product CSV Import: Clean Supplier Data With AI First — AI cleanup rules for supplier spreadsheets, the exact columns WooCommerce's CSV importer expects, and a 10-row dry run on staging.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Ginetex care-labelling pages (ISO 3758:2023, the care labelling code using symbols) and general garment-labelling practice. Check your own markets' fibre-labelling and environmental-claims rules.