The Legal Fields Missing From Most Product Listings

A candle maker in Lyon selling to Germany, a UK potter shipping to Dublin, an American seller listing on a European marketplace: all three are subject to a rule that arrived at the end of 2024 and that almost nobody in that group has heard of. It does not care how small you are. It applies per offer.

What follows is the field level version of that obligation, the equivalent rules on the American side, and an honest account of which parts of the audit an AI tool can run for you and which parts it will get wrong in ways that cost you.

What has to appear on a listing sold into the EU?

Four things, in the offer itself. The General Product Safety Regulation replaced a directive that had stood for over twenty years, and the European Commission's own summary confirms it became applicable on 13 December 2024 with no transition period. Its distance selling provisions target the online offer specifically, because that is where the buying decision now happens.

The four items are the manufacturer's name and postal and electronic address, the details of the responsible economic operator inside the EU when the manufacturer sits outside it, information identifying the product including an image, and any warnings or safety information in a language the consumer can readily understand. Guidance published for businesses selling into the EU states plainly that online listings must include key safety information and that non compliant goods can be pulled from marketplaces.

The responsible person requirement is the one that catches small sellers hardest, because it is structural rather than editorial. Every product needs a designated operator established in the EU or Northern Ireland who holds the compliance file. That can be an authorised representative appointed by the manufacturer, an EU based importer under a written agreement, or a fulfilment provider under the same. Without one, the same guidance warns that goods risk being stopped at the border. No amount of careful listing copy substitutes for the appointment.

Why is the American version so different?

Because it was written to solve a different problem: counterfeit and stolen goods moving through marketplaces, rather than unsafe products reaching consumers. The FTC's guidance on how the INFORM Consumers Act affects online sellers describes an obligation that lands on identity. Marketplaces collect and verify bank, contact and tax information from high volume third party sellers, and disclose seller identifying information to consumers.

The thresholds are worth writing on a sticky note if you sell on any marketplace. A high volume third party seller is one that, in any continuous twelve month period within the past twenty four months, made 200 or more separate sales of new or unused consumer products and took 5,000 dollars or more in gross revenue. A further disclosure tier applies above 20,000 dollars in annual gross revenue, at which point name, address and contact information must appear clearly on listings or order confirmations. The Act took effect on 27 June 2023, and the FTC notes that it and state enforcers can seek substantial penalties.

The practical consequence for a seller who crosses both regimes is that your compliance data has two shapes. One set of fields describes the product and its maker. The other describes you and how a buyer reaches you. Storing them in the same place is the only way this stays manageable.

The audit table, field by field

Below is the working version we use when reviewing a catalogue. The last column is the honest part: what an AI pass can do with the field, and where it stops being trustworthy.

Note the pattern down the last column. Every row where AI helps is a presence and consistency question. Every row where it fails is a judgement about a legal fact. That division holds for essentially all compliance work, and it is why the sensible framing is an audit tool rather than a compliance tool.

How do you actually run this across a catalogue?

Five steps, and the first two are the ones people skip.

1. Export the catalogue with its fields, not its pages. You want a table of products against attributes. If your compliance information exists only inside description text, the export step will show you that immediately, and that discovery alone is worth the afternoon.
2. Classify by regime. Which products ship where, and which fall into a category with its own rules. This is a grouping exercise a model does well, and the same grouping work we described in the piece on categorising a catalogue with AI pays off twice here.
3. Flag empty required fields. A simple presence check across the matrix. No cleverness required, and no generation. The output is a list of products and missing fields, sorted by how many you sell.
4. Check the wording that exists. Language match, units, and whether a warning is present at all. Use AI to surface suspicious cases for a person to read, never to approve them.
5. Fix at the data layer. Put the values in fields, then regenerate the listings from the fields. This is the difference between a fix that survives and one that disappears at the next content refresh.

Step five is where most of the durable value sits, and it connects directly to how you write the rest of the listing. The method we set out for writing product descriptions from an attribute table works here without modification: the compliance fields are simply more attributes, and a description generated from the row cannot lose them.

What goes wrong when a model is left to write compliance text?

It writes something that reads correct. That is the entire problem, and it is worse than an obvious error because nobody catches it in review. Four failure modes recur.

The first is the invented responsible person. Asked to complete a listing template, a model will produce a plausible company name and address for the EU representative field, because the field wanted filling. There is no signal in the output that this was fabricated. The second is the confident conformity claim: a generated line stating that a product meets a named standard, which is a claim about a document you either hold or do not. The third is a translated warning that is linguistically fine and legally wrong, because required safety wording in many categories is prescribed rather than described. The fourth is quieter, which is silently dropping a warning that was present in the source when reformatting a description.

The defence is procedural rather than technical. Compliance fields are copied, never generated. If a value did not come from a document you can produce on request, it does not go on the page. We made a version of that argument about advertising language in the piece on the claims you can and cannot make about AI in your own marketing, and the logic is identical: a claim is only as good as the evidence behind it.

What actually happens if a listing is incomplete?

Rarely a knock at the door. The first consequence for a small seller is usually commercial and quiet. A marketplace suppresses the listing or removes it, since platforms carry their own duties and enforce them by removing risk rather than by investigating it. Guidance for sellers into the EU is explicit that non compliant goods can be taken down. Payment providers and fulfilment partners have their own versions of this.

Then there is customs. The responsible person requirement exists partly so that authorities have someone in the union to talk to, and goods without one can be held. For a seller whose stock is sitting in a container, that is a cash flow event rather than a legal one, and it arrives faster than any enforcement action. If you ship across borders regularly, our walkthrough of what AI can and cannot do with cross border customs paperwork covers the adjacent documentation.

Formal enforcement sits at the end of that chain, not the start. UK guidance on product safety for businesses describes the shape of it: local Trading Standards investigate, the Office for Product Safety and Standards handles national or novel cases, and remedial action can mean new instructions, product modification, or a recall with direct consumer notification. Retailers are separately expected to keep supplier records for traceability, which is a requirement a lot of small shops discover only when asked for one.

Does this apply to a shop selling twenty handmade items?

Yes, with proportionate effort. The regulation attaches to products placed on the market, not to businesses above a size. What changes with scale is how much of it is already handled for you: a shop reselling branded goods inherits most of the required information from its supplier, while a maker producing their own items is the manufacturer and owns the whole file.

For the maker the practical starting point is short. Establish your own details as the manufacturer, keep a record of what each product is made of and any assessment you have done, give each production run an identifier, and put the safety information that belongs to the category on the offer. That is a morning's work for a small range, and it is a great deal less painful than assembling it after a complaint.

The one thing worth doing regardless of size is keeping the compliance fields inside your own system rather than inside a marketplace's listing form. Marketplaces change their forms, and a shop that owns its product data can republish everywhere in an afternoon. That is the same argument we make for owning the storefront code itself, which is why our ecommerce website builder generates a catalogue schema you control instead of renting you fields on someone else's platform.

Where does this information physically go on a page?

Closer to the product than most shops put it. The requirement is that a consumer can find the information in connection with the offer, which in practice means it belongs on the product page rather than behind a link labelled Legal. Sellers who have been through a marketplace review tend to converge on the same layout: a short block below the description, or a clearly labelled tab beside specifications, carrying the manufacturer block, the responsible person block, the identifier and the warnings.

Two placements cause trouble. Putting the block inside an image is the first, because the text is then invisible to translation, to search and to anyone using a screen reader. Putting it only in a downloadable PDF is the second. Neither is an obviously bad faith choice, and both are the kind of thing a shop does once and repeats across four hundred pages.

Language is the other placement question, and it is the one AI genuinely helps with as long as the boundary is respected. The requirement is that warnings reach the consumer in a language they readily understand, which for a shop selling into several member states means several versions. Machine translation of a warning you already hold is a reasonable use with human review. Machine authoring of a warning you do not hold is not a translation problem at all, it is an invented claim.

Used, refurbished and secondhand items

Frequently assumed to be outside all of this, and they are not. The Commission's summary is explicit that the regulation reaches consumer products sold online whether they are new, used, repaired or reconditioned. For anyone running a vintage or refurb business that is a meaningful finding, because the supplier chain that would normally hand you the manufacturer file does not exist.

What is workable in practice is to record what you can establish and to be accurate about what you cannot. The original manufacturer's identity, where it is legible on the item, belongs on the offer. Condition, any repair carried out and any part replaced belong there too, both because buyers need them and because they are the traceability record you would be asked for. A refurbisher who changes the safety characteristics of a product moves closer to the manufacturer's position, which is a conversation worth having with an adviser rather than with a chatbot.

Keeping the audit alive after the first pass

The first audit is a project. Everything after it is maintenance, and maintenance is where compliance work usually dies. Three habits keep it from decaying.

Attach the check to the moment a product is created rather than to a calendar. A new product that cannot be published without its required fields populated is a rule you enforce once, in the system, instead of a review you promise to run quarterly. This is the same reasoning behind writing your operating procedures down before you automate them, which we covered in the piece on standard operating procedures as an AI first hire.

Re run the presence check when a supplier changes, not when the law changes. Supplier substitution is the most common way a compliant catalogue quietly stops being one, because the new manufacturer details never make it into the fields. A monthly diff of manufacturer values against last month catches it in seconds.

Keep the evidence with the value. For every compliance field, store where it came from: the invoice, the supplier declaration, the appointment letter. When somebody asks, and the person asking is usually a marketplace rather than a regulator, you want to answer the same afternoon. A shop that has to reconstruct its evidence from memory is the shop that ends up delisted while it looks.

A short checklist to run this quarter

Export your catalogue to a table. Add a column for each required field above. Sort by units sold and start at the top, because the exposure is concentrated there. Ask an AI tool to tell you which cells are empty and which warnings look absent, then fill them from documents rather than from prompts. Put the result back in your product data. Then set a reminder for six months, because supplier changes are what reopen the gaps.