Verify a Supplier in China Before You Pay: What AI Can Check

Check four things before any production payment: the supplier's company record on the official enterprise register, a live video call that shows the premises, a paid sample you inspect, and a bank account whose name matches the registered company. AI can translate, draft questions and compare documents, but it verifies nothing by itself.

The danger with AI here is a false sense of safety. A chatbot will happily say a licence "looks valid" because it cannot tell a real scan from a forged one. Treat it as a fast assistant for the paperwork around verification: translating, spotting mismatched names and addresses, writing the script for the call and keeping the score sheet. The evidence has to come from sources it cannot invent.

Follow me on Instagram@sagnikteaches

What can go wrong, in the order it usually does

First orders from unfamiliar suppliers fail in a handful of repeatable ways. Knowing them tells you which check is aimed at which risk:

Connect on LinkedInSagnik Bhattacharya

• The supplier does not exist, or exists only as a listing and a payment account.
• The supplier exists but is a trading company presenting itself as a factory, so the price carries a margin and quality control is two steps away.
• The sample is excellent and the production run is not.
• The payment goes to someone else: a personal account, a different company, or an account changed by an email "from the supplier" that was not from them.
• The goods differ from the specification, and the supplier's position is that the sample was an example.

A trading company is not a scam. Many are honest, useful intermediaries. It becomes a problem when the claim is false or when you wanted factory-direct pricing and a factory's accountability. The checks below tell you which you are dealing with.

Subscribe on YouTube@sagnikteaches

Step 1: Look the company up on the official register

Each registered company in the country carries a business licence, which typically shows a unified social credit code of 18 characters, a registered Chinese name, a legal representative, a registered address, a registered capital figure and a business scope. The market regulator runs the National Enterprise Credit Information Publicity System at gsxt.gov.cn, where you can search by company name or code and see the company's status. Type the address yourself rather than following a link from the supplier's email, and expect the interface to be in Chinese.

Ask the supplier for a photo or scan of the business licence and the exact Chinese company name. Then check, with the register open in a separate window:

• The code you were given matches the code the register shows for that name.
• The status is active, not cancelled or revoked.
• The legal representative's name matches the licence copy.
• The registered address is plausible for the claimed business (an office-tower unit for a trading company, an industrial district for a factory).
• The business scope includes what they say they sell. A scope of "garment sales" does not make a company a maker of electronics.
• The registration date: a company registered three weeks ago claiming ten years of experience deserves a direct question.

The register shows company details, not whether the company makes good products or ships on time. A genuine company can still be a bad supplier, and a mismatch can be an innocent translation or transliteration difference, which is why you ask before you conclude.

Where AI helps with the register check

Paste the supplier's licence scan or text and the register's record side by side and ask:

Below are two records. A is the business licence the supplier sent me (text copied or typed from the scan). B is what the official register shows for the same code (copied by me).
1. Translate both into English.
2. List every field that differs, even slightly: name, code, legal representative, address, registered capital, scope.
3. Say which differences could be a translation or spelling variation and which could not.
Do not tell me whether either is genuine. You cannot know.

That last instruction matters. A model will otherwise tell you the document "appears authentic". An illustrative reply lists: the code matches; the English name on the supplier's website differs from the registered Chinese name's translation by one word; the address on the licence names a different floor from the website. It is a to-do list for your next email, not a verdict. If you send documents into a public translation tool, read whether that is safe first and remove anything that is not needed for the check, such as bank details.

Step 2: A video call that proves the place exists

A call tests three things a document cannot: the person answers questions about the product without scripts, the place looks like the one claimed, and the supplier will do something live that you pick, which a faked listing struggles with. Book 30 minutes, ask for video on, and ask the contact to show:

1. The entrance with the company name on it, then the production floor, panning slowly, not a cut video.
2. The machine or process that makes your product, running.
3. A finished unit of your product picked at random from stock, with the batch label.
4. A piece of paper they write today's date and your company name on, held next to the product.

Prepare questions with AI. An illustrative prompt for a first order of 2,000 silicone phone cases:

I am buying 2,000 silicone phone cases for a small online shop. The supplier says they are a factory. Write 15 questions for a 30-minute video call that would show whether it is a factory or a trading company, in plain short English a non-native speaker can follow. Include questions about tooling, minimum order, who owns the moulds, sample lead time and quality checks. Add the answer that would worry me for each.

An illustrative extract of the output: "Who made the mould for this case? Worrying answer: 'Our partner factory.' Why: it suggests a trading company. Can you show the injection press today? Worrying answer: 'The machine is in another building and not working.'" Delete questions you would not know how to judge; fifteen is too many for half an hour, and eight good ones beat fifteen rushed ones.

Expect an interpreter or a sales contact with good English, which is fine. Note the faces on the call and ask for their name and job title, then check that the same name appears on their business card or email signature. If a call is cancelled three times, or the camera "does not work" on every attempt, record that in the score sheet below.

Step 3: A paid sample, inspected against a written specification

Order the sample as if it were the real thing: same material, same finish, same packaging, paid by the same route you will use for the order. A sample hand-picked from stock and sent free tells you what the best unit looks like, not what a run of 2,000 will look like. Ask for one that comes off the production line, and for two from different batches if the supplier will.

Write the specification before the sample arrives. A page with measurable points is enough: dimensions with tolerance (65.0 mm wide, plus or minus 0.5), material and hardness, colour reference, print method, packaging, labelling, and the tests you require if the product is regulated in your market. AI can turn a rough brief into a tidy specification, and you can ask it to list what is missing:

Here is my rough product brief for silicone phone cases. Turn it into a one-page specification with measurable acceptance criteria for each point. List the points where I have not given a number, and ask me for them. Do not invent tolerances or standards.

Measure the sample against that page with a ruler, a scale and a good light. Keep photos. Use the same sheet for the first production run, so "it matches the sample" has a number behind it. If the product has safety or labelling rules where you sell it, ask a qualified adviser or the relevant authority what is required; do not rely on a chatbot for that.

Step 4: Make the bank account match the company

The supplier's invoice or pro forma lists a beneficiary name, a bank and an account number. The beneficiary should be the exact registered company name you found in step 1, not a person and not a similar-sounding company. Ask for the details in writing on the supplier's letterhead and confirm them by a call to a number you found yourself, not one in the email.

Be wary of three patterns: a request to pay a personal account "for tax reasons", a late change of bank details by email, and pressure to pay today for a discount that expires today. If details change, treat it as a new payment verification, even if the email looks like earlier ones. Business email compromise, where a criminal impersonates a supplier or takes over their mailbox, is common enough that any change of account deserves a phone call. This tutorial on catching duplicate invoices and payment fraud covers the checks on your side of the payment.

A practical staging for a first order is a modest split: pay a share before production, inspect, then pay the balance before shipping or against documents, whatever you agree in writing. The shares depend on the order, the supplier and the payment protection you have, so settle them in the contract and not over chat. Some importers use an escrow-style service or a platform's payment protection for first orders; read the current terms of whichever you use, including what it does not cover and how long you have to raise a claim.

A worked example with numbers

An illustrative case: a small online shop plans a first order of 2,000 silicone phone cases at $1.10 a unit, $2,200 for goods, plus about $400 shipping. Sales price $9.99, so the order could return around $19,980 gross if all sell, which is why a failed batch hurts. Two suppliers answered:

The saving at supplier B is $0.22 a unit, or $440 across the order. Against that, the whole $1,760 you would send supplier B is at risk, four times the saving, before shipping is even counted. The cheap quote is information about the risk, not a bargain. Even with supplier A, the shop might begin with 500 units at $550 to test production quality before committing to the 2,000.

A score sheet you can reuse for each supplier

Give each check a score of 0, 1 or 2 and total them. Zero is a fail or refusal, 1 is partial, 2 is clear. A page like this also lets a colleague see why you chose someone:

• Register record matches licence and name (0-2)
• Business scope fits the product (0-2)
• Video call: live, unscripted, premises shown (0-2)
• Answers on tooling, minimum order and quality (0-2)
• Sample matches the written specification (0-2)
• Bank beneficiary equals registered company (0-2 - and a 0 here ends the process)
• Consistency of names, addresses, phone and email across documents (0-2)
• Willingness to put terms in a written contract (0-2)

A supplier scoring 14 or more of 16 with no zero is a sound candidate for a small first order. Anything with a zero in the bank line or the register line stops until resolved. Ask AI to fill the sheet from your notes, but you supply the scores. Keep the sheet and the supplier's documents in one folder per supplier so they sit with your wider supplier records.

Filled in for the two suppliers in the worked example, the sheet reads like this (illustrative scores):

Supplier A clears the bar with no zero, so a small paid order is reasonable, and the two scores of 1 become the questions for the contract stage. Supplier B stops on the bank line alone; the other zeros only confirm it. Notice that the sheet would have stopped B even if its price had been half, which is the point of writing the rule down before the quotes arrive.

Using AI to write messages across a language gap

Messages between you and a supplier often pass through a translation layer on both sides, so short sentences with one question each work best. Ask the AI to rewrite your drafts in simple English and to list every question in the message so none is lost. An illustrative before and after:

Before: "Could you please let us know whether it would be feasible to supply a pre-production sample prior to the deposit, and what your MOQ is?"

After: "Can you send one sample from your production line before I pay a deposit? Please tell me: 1) sample price, 2) shipping time, 3) your minimum order quantity."

The after version is easier for a translation tool and for a person whose English is a second language, and it leaves a clear trail of numbered answers. Go through replies and ask the AI to produce a table of answers to your numbered questions, with "no answer" in blank cells. That makes evasions visible. For tone and wording once the supplier checks out, the negotiation emails tutorial carries on from here. A poor translation can still change a number or a date, so confirm quantities, prices and deadlines in figures in a line of their own, and for contract wording have a bilingual professional check the version that will be signed. For a deeper look at when machine translation is good enough, see this tutorial on client-facing documents.

Red flags, and what each one usually means

• Prices far below every other quote. Either a different (cheaper) specification, a bait quote to be raised later, or a payment scam.
• Refusing a live video. There is no premises, or they do not control one.
• A different company name on every document. A trading company with several entities, or a borrowed licence.
• Brand names you did not ask for. Offering well-known brands at low prices points to counterfeits, which can be seized at the border and expose you to legal claims.
• No contract, only a chat order. You lose your recourse on specification and delivery.
• Very high urgency. The "last units", the "price rises tomorrow" and the "pay now to hold the slot".

When a check goes wrong: three small stories

These are illustrative composites of how problems tend to surface, not accounts of particular firms.

The licence that belonged to a neighbour

A supplier's licence scan carried a valid code and active status, but the company name was one character different from the supplier's website. The importer asked why. The answer was that the supplier used a related company's licence for exports. Not necessarily a scam, but the beneficiary on the invoice would then have been a third company. They asked for the order to be placed in the licence holder's name, or walked away.

The sample that was too good

The sample was flawless and had a different surface finish from the photos on the listing. A pre-production sample from the line, ordered afterwards, was noticeably rougher. The importer rewrote the specification to include a surface-finish reference, and the supplier agreed to it in the contract.

The changed account number

Two days before the balance payment, an email arrived "from the supplier" with a new bank account because of an "audit". The sender address differed by one letter. A phone call to the number from the first contact confirmed the supplier had sent nothing. The balance went to the original account, and the shop added a rule: no account change is accepted by email alone.

The pre-payment routine on a calendar, with your hours

1. Day 1: collect licence, company name, website and listing; do the register lookup and mark mismatches (45 minutes).
2. Day 1: send the questions, the specification draft and a request for a paid sample (30 minutes with AI drafting).
3. Day 2 to 3: hold the video call and fill in the score sheet (45 minutes).
4. Day 10 to 20: sample arrives; measure, photograph and compare (1 hour).
5. Before payment: confirm the bank beneficiary by phone and record who confirmed it and when (15 minutes).

About three and a quarter hours of your time, spread over three weeks, plus the sample cost, to decide whether to commit a few thousand dollars. If you prefer to talk the routine through against the way you actually source, an AI implementation consultation can cover which parts to automate. The wider picture of tools an importer needs is in this look at supply chain software for small importers.

Questions importers ask before a first payment

Can AI tell me whether a supplier is real?

No. It can translate a licence, compare two documents for mismatched names and draft your questions, but it cannot confirm that a factory exists or that a document is genuine. Only an official register lookup, a live video call, a sample you inspect and a payment to a matching company account give real evidence.

Is a supplier badge on a sourcing platform enough?

It is one signal, not proof. Platform badges and payment protection programmes have their own conditions and exclusions, so read the current terms on the platform itself and treat the badge as a reason to keep checking, not to stop. Many scams use real platforms.

Should I pay a deposit before the sample arrives?

Pay for the sample only, and pay the quoted amount to the same company account you will use later. Do not send a production deposit until you have inspected the sample and had the video call. If the supplier refuses a paid sample, that tells you something.

When is a third-party inspection worth paying for?

When the order value is large enough that a failed batch would hurt, or when you cannot attend the factory. An inspection company checks goods before they ship against your written specification. Ask several for quotes, and keep the booking independent from the supplier's own suggestion.

Further reads

• How to Compare Supplier Quotes Side by Side With AI — Line up quotes from checked suppliers side by side.
• How to Automate Purchase Orders and Supplier Emails With AI — Turn an approved supplier into a repeatable purchase order.
• Why Is AI Bad at Maths? What to Check in Quotes and Invoices — What to recheck in quotes and invoices the AI summarised.
• How to Match Supplier Invoices to Purchase Orders Automatically — Check the supplier's invoice against what you ordered.
• AI Freight Quoting: Turn Customer Emails Into Draft Quotes — How forwarders turn emails into quotes, useful when comparing shipping.
• AI Invoice Processing: Stop Typing Supplier Bills by Hand — Stop retyping supplier bills by hand.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: platform buyer-protection pages (check current terms on the platform), published guidance on the 18-character unified social credit code and the National Enterprise Credit Information Publicity System (gsxt.gov.cn; the site itself sits behind a browser check, so its details come from secondary guides read 5 November 2026), and the importer's own practice. Order figures are illustrative.