A QR code on a label is a small thing. Behind the one arriving in European product law sits a file you are expected to have written, hold, keep current and register, for every distinct product you sell. Not a marketing page. A structured record with roughly fifty fields, some of which you cannot answer today without telephoning your supplier.
- The digital product passport is established by the EU Ecodesign Regulation 2024/1781, with the registry due to become operational on 20 July 2026 and product categories phasing in from 2027.
- Batteries come first, on 18 February 2027. Textiles, aluminium and tyres are expected in the second half of 2027, furniture in 2028, mattresses and ICT in 2029.
- Responsibility sits with the economic operator, meaning the manufacturer or the importer. If you import and sell under your own name, that is you.
- Reporting on the textiles passport describes around 50 data points in four categories, created at SKU level rather than per item or as a range average.
- No small business exemption is described for apparel and footwear sold in the EU, regardless of where the brand is based or how small it is.
- A language model reached 85 percent on extracting product attributes in research conditions. For a search filter that is useful. For a declared compliance field it is not.
What is a digital product passport, and who has to make one?
It is a structured data record attached to a product and reachable by scanning something on it. The European Commission's own page on the digital product passport calls it a digital container for products, components and materials, holding information that supports sustainability, promotes circularity and facilitates legal compliance.
The legal base is the Ecodesign for Sustainable Products Regulation, numbered (EU) 2024/1781, and it is not the only route to a passport. The Commission lists several other instruments that carry their own versions, including the Batteries Regulation (EU) 2023/1542, the Packaging and Packaging Waste Regulation, the Critical Raw Materials Act (EU) 2024/1252, the Toys Safety Regulation (EU) 2025/2509, the Construction Products Regulation, and the rules on detergents and surfactants. A shop selling a battery powered toy in a box could conceivably meet three of those.
On responsibility the Commission is direct. Economic operators, meaning manufacturers and importers, must gather the required information, create the passport, and register it in the EU Registry, either storing the product information themselves or using a passport service provider. Consumers reach it by scanning a data carrier such as a QR code on the product.
That word importer is the one that catches small sellers. If you buy from a manufacturer outside the EU and sell inside it under your own brand, you are not a retailer passing on someone else's paperwork. You are the operator whose name is on the obligation, and the data you need lives in a factory you may never have visited.
When does this actually reach you?
In stages, by product category, over several years, and the first dates are close. The Commission's published timeline is the thing to plan against rather than any vendor's urgency:
| Date | What happens | Who it affects |
|---|---|---|
| July 2026 | Registry framework established | Nobody directly, but the plumbing exists |
| 20 July 2026 | The DPP Registry becomes operational | Where passports will be registered |
| 18 February 2027 | Passports mandatory for certain batteries | Anyone selling batteries or battery products |
| Q2 2027 | Construction products requirements | Building materials and trade suppliers |
| Q3 to Q4 2027 | Textiles, aluminium and tyres | Clothing, footwear and accessories sellers |
| 2028 | Furniture | Makers and importers of furniture |
| 2029 | Mattresses and ICT products | Bedding and electronics |
Two things about that table deserve emphasis. Economic operators get at least 18 months of transition after a delegated act is adopted, so the compliance date always sits later than the date the rules are written. And the categories are sequenced, which means most shops have a specific year to care about rather than a single cliff. Selling batteries makes 2027 your problem. Selling mattresses makes 2029 yours.
The trap is design lead time. Carbonfact's analysis of what the textiles passport will require makes the point that products being designed now will enter the market when the obligations already apply. For anyone with an eighteen month development cycle, a product you specify this autumn is a product that needs the data.
Is there a small business exemption?
Not one described for apparel and footwear, and that is the answer most small sellers are hoping to avoid. The same analysis states the textiles passport applies to all apparel and footwear sold on the EU market regardless of where the brand is based or its size, and notes that no small business exemptions are mentioned.
Two qualifications, offered honestly. The textile rules sit in a delegated act that is expected rather than published, so the final text can differ from what is anticipated today. And thresholds or simplified regimes for micro enterprises are a normal feature of EU product law, so their absence from current analysis is not proof of their absence from the eventual instrument.
What follows from that uncertainty is not a reason to wait, because the work that helps is the same either way. Every field a passport will ask for is a field that already improves your listings, your marketplace eligibility and your answers to customer questions. If a simplified regime arrives, you will have done work you needed anyway. If it does not, you will be ready. That asymmetry is unusual in compliance and worth exploiting.
Can a language model just fill these fields for you?
For some of them, well. For the ones that matter legally no, and the research gives a number that explains why. The paper ExtractGPT, by Alexander Brinkmann, Roee Shraga and Christian Bizer, tested large models on extracting product attribute values from unstructured text. GPT-4 reached the highest average F1-score of 85 percent using detailed attribute descriptions and demonstrations, with Llama-3-70B comparable, beating the best pre-trained language model baseline by 5 percentage points.
Eighty five percent is a genuinely strong result for that task and it is the wrong bar for a passport. Think about what the fifteen percent contains. For a search filter, a wrong material tag means one product surfaces in the wrong list and a shopper is mildly annoyed. For a declared compliance field, a wrong fibre percentage is a false statement about a product, made by you, published to consumers and registered with a regulator. Those are not the same error even though the model made them at the same rate.
The paper contains a second finding worth knowing before anyone sells you a fine tuned solution. The authors report that fine tuning GPT-3.5 reduced the model's ability to generalise to unseen attribute values. So the intuitive fix, training a model on your own catalogue, can make it worse precisely where you need it most, which is on the new product you have not sold before.
The workable division looks like this:
| Field type | May a model draft it? | Why |
|---|---|---|
| Care and repair instructions | Yes, then check | Written guidance, low factual risk, easy to verify by reading |
| Product description and category | Yes | Already how most catalogues are built, and errors are visible |
| Material composition percentages | No | A fact about manufacture that only the supplier holds |
| Country of manufacture and supplier identity | No | Comes from your purchase records, never from inference |
| Chemical compliance and substances of concern | No | Requires documentation, and a guess here is a serious claim |
| Durability and recyclability scores | No | Scored against a defined method, not estimated from a description |
| Translations of approved text | Yes, with review | Source text is already verified, only the language changes |
The pattern is simple enough to remember. A model may rewrite what you already know. It may not discover what you do not. That is the same boundary we drew around moving a catalogue, where certain fields must never be left to an automated migration, and the passport raises the stakes on exactly those fields.
The granularity detail that decides the work
Passports are described as being created at SKU level, rather than for each individual item or as a single average across several products. That single sentence sets the size of the job, and it cuts both ways.
The good news is that per item passports would be unmanageable for a small maker. You are not generating a record for every unit that leaves the workshop. The bad news is the opposite end: you cannot write one passport for a product line and apply it across colours, materials or variants that genuinely differ. A shop with forty products across four colourways is closer to a hundred and sixty records than to forty, in any case where the colour changes the material or the finish.
This is where a small catalogue becomes a real advantage and a sprawling one becomes a liability. The most useful preparation available to most shops is not software. It is deleting variants that exist because somebody once thought they might sell, because every surviving SKU is a record somebody has to maintain. A range audit done for commercial reasons in 2026 is compliance work done early.
The reported textile field list includes durability and recyclability scores on a 0 to 10 scale. Scores on a fixed scale imply a defined method of arriving at them, which means those fields will be calculated against criteria in the delegated act rather than chosen by the seller. Do not plan to estimate them.
What should you start recording now?
Three things, none of which need a passport platform, and all of which are harder to reconstruct later than to capture today.
First, supplier identity and country of manufacture, per SKU, in a field rather than in an email. Most small shops know this and hold it in the form of an invoice from 2023 in a folder. When a passport asks, a folder is not an answer.
Second, material composition as your supplier stated it, with the date they stated it and who said so. Suppliers change materials without announcing it, and the version you were told two years ago may not describe what arrived last month. Recording the claim and its date turns a future problem into a dated record rather than an argument.
Third, certificates and test reports as files linked to the SKU, with their expiry dates. Compliance documentation is one of the four data categories, and the common failure is not the absence of a certificate but the inability to find it or the discovery that it lapsed.
All three belong next to your product records rather than in a separate compliance spreadsheet, because a second system is a system that falls out of date. A shop that owns the database behind its own catalogue can add three columns and a file reference this week; on a platform where the product schema is fixed you may be waiting for somebody else's roadmap, which is a reason to know now rather than in 2027.
What does the registry change in practice?
It turns a claim into a filing. Today the information about your products lives on your own pages, and if it is wrong you correct the page. Once a passport is registered, the data exists in a place a regulator and a customer can both reach independently of you, and correcting it becomes an update to a record rather than an edit to copy.
That has a quiet consequence for how you write product information. Anything you can state on a page today with a bit of generous phrasing becomes a field with a defined meaning. Handmade, natural, durable and recycled all survive as marketing words on your site, but where a passport has a numeric field behind the same idea, the number is what is checked. A shop whose environmental messaging currently rests on adjectives has some work to do on the underlying facts, not on the adjectives.
It also creates a durable record that outlives the sale, which is the intended point. A passport is meant to be readable by whoever holds the product in five years, including a repairer or a recycler. That is a genuinely different design brief from a product page written to convert a shopper this week, and it is worth knowing that the same field is serving both audiences.
Where does this leave someone who does not sell into the EU?
Watching, and not for long. The obligation follows the market rather than the seller, so a shop in Britain, Canada or Australia selling to EU consumers is inside it while one selling only domestically is not. That distinction is clean today and historically it does not hold.
The pattern with EU product rules has been that they become the format everyone uses, because a manufacturer that has built one compliant data set does not maintain a second, looser one for other markets. If your suppliers sell into Europe, the material and origin data you need will start existing whether or not you ask for it, which makes asking cheap. If your suppliers do not, you may find in 2028 that the factory cannot tell you what its product is made of, and that is a sourcing fact worth learning early rather than late.
Either way the actionable version is the same and does not depend on jurisdiction. Know your suppliers, know what the goods are made of, keep the certificates, and hold all three against the SKU rather than in your memory.
Why this is the same problem you already have
Because every recent demand on a small seller has been a demand for structured product data, and they keep asking for the same fields. Assistants and agents need machine readable attributes to recommend and buy your products. Marketplaces reject listings with missing fields. Advertising platforms disapprove feeds that contradict themselves.
And the legal layer has been converging on the same thing for a while. We looked at the fields that most product listings are quietly missing in an audit of the legal information a listing is supposed to carry, and at the specific trap of environmental language in what happens when green claims are written by AI. The passport is the same demand with a registry attached and a date on it.
Which suggests the right way to think about the cost. If you treat the passport as a compliance project, it is an expense with no return that arrives in 2027 or later. If you treat it as the product data work you have been deferring since assistants started reading catalogues, it is a thing you were going to do anyway, now with a deadline that makes it easier to justify spending a week on.
The one genuinely new burden is the registry and the data carrier, and those are mechanical. Somebody will sell you a QR code and a hosted record, and by 2027 several people will. What nobody can sell you is the knowledge of what your products are made of and who made them. That is the part to start on, and it is the part that a model, at 85 percent, cannot supply.