AI customs classification: what a proposal must show

AI customs classification gives you a proposal, not a decision. A useful proposal names a code, shows the path through heading, notes and subheading that leads to it, and asks a question when the invoice does not hold the facts the tariff classifies on. The decision stays with a person: whoever lodges a customs declaration is responsible for the accuracy and completeness of its information, and so is a customs representative who lodges it [1]. A Binding Tariff Information (BTI) decision binds customs and its holder on a classification [1]; a proposal, whether it comes from AI, a database or a colleague, does not.

What HS classification AI has to follow

The General Rules for the Interpretation of the Combined Nomenclature govern every classification [2], and German customs states that they must be applied [3]. Rule 1 sets the order of reading: the titles of sections, chapters and sub-chapters are for ease of reference only, and for legal purposes classification is determined by the terms of the headings and any relative section or chapter notes [2]. German customs adds that other classification decisions have to be taken into account, such as classification regulations, Harmonized System opinions and court judgments, and that the explanatory notes are an aid to interpretation [3].

For AI HS code classification, that sets the bar. An HS code AI gives without its reasons cannot be checked. It becomes checkable when the proposal shows which heading text and which notes it applied, which is what you would ask of a colleague who calls out a code across the office.

What a good proposal shows

1. The full code, with its date. An import declaration in Germany always takes the 11 digit code [3]; the guide on how a commodity code is built covers the Dutch ten. The Commission publishes the complete Combined Nomenclature every year, applying from 1 January of the following year [3], and TARIC data is sent to the national administrations every day [4].
2. The path. Chapter, heading and subheading, with the rules used to get there: rule 1 with a chapter note, for example, or rule 6, under which subheadings are compared by their own terms and subheading notes, and only subheadings at the same level are comparable [2].
3. The facts it relies on, each tied to the invoice line it came from. German customs notes that many goods can be classified only when both their material and their intended use are taken into account [3].
4. What is open, named as a gap rather than filled with a silent assumption.
5. Comparable decisions, such as BTI decisions on similar goods. The Commission's database publishes them with their non-confidential details, among them the description of the goods, the validity, the classification and the reasoning behind it [5].

A proposal with these five parts can be checked in a few steps. A proposal without them leaves you to do the classification again yourself.

When the invoice is too thin

Take the invoice line "ceramic dinner plate, 27 cm". In the 2026 Combined Nomenclature, tableware of porcelain or china is 6911 10 00, with a conventional duty of 12%, and ceramic tableware other than of porcelain or china falls under 6912 00 21 to 6912 00 29, at 5% to 9% depending on whether it is common pottery, stoneware, earthenware or fine pottery, or another ceramic [2]. The invoice does not say which.

A proposal that still names a code here is guessing. The useful output is one question for the supplier, answerable from the product specification: "Is the plate porcelain or china? If not, is it common pottery, stoneware, or earthenware?" The declaration needs this anyway. For release for free circulation, the description of goods is the normal trade description, in terms sufficiently precise to enable immediate and unambiguous identification and classification of the goods [6]. The guide when the invoice is not enough has the other questions worth asking.

Where the declarant decides

Lodging a customs declaration makes the person concerned responsible for the accuracy and completeness of its information and for the authenticity, accuracy and validity of the documents supporting it; where a customs representative lodges it, the representative is bound by the same obligations [1]. For release for free circulation, the declarant is the debtor of the customs debt, and under indirect representation so is the person on whose behalf the declaration is made [1].

A proposal from tariff classification AI changes none of this. It is preparation, which the declarant checks, adopts or rejects.

Checking an HS code from ChatGPT

Whether the code comes from ChatGPT, a classification tool or memory, the check is the same. In the TARIC consultation, search by goods code and country of origin, with the date of the declaration as the reference date [7]. Three things show quickly:

• Does the code exist? If a goods code is not or no longer valid in the EU, the consultation says so and suggests entering the first six digits and browsing the nomenclature [7].
• Does the text fit? Browse the nomenclature down to the code and read each description above it. If the path does not describe your goods, the proposal is wrong.
• Which measures apply? TARIC holds the measures attached to the code, from tariff preferences and quotas to anti-dumping duties and prohibitions, but not national VAT or excise rates [4].

When a BTI is worth it

A BTI decision is worth applying for when goods recur and their classification is disputed. It binds the customs authorities and its holder on the classification, and it is valid for three years [1]. The duty rate and the other measures that follow from the code are not part of it [5]. Before applying, anyone can ask the customs authority for general, non-binding information on the classification of goods [8]. How to search existing decisions, and how to apply, is in the guide on binding tariff information.

Athren prepares the case from the shipment email and proposes a code for clear lines; your declarant checks what is open and files it. Athren files nothing.