Your AI-Drafted China Contract Says It Needs A Lawyer. Listen To It.

Your AI-Drafted China Contract Says It Needs a Lawyer. Listen to It.

In the past couple of months, at least three unrelated companies have asked me to review an AI-drafted China contract containing this exact dispute-resolution provision:

Governing Law; Dispute Resolution; Controlling Language

The governing law, court/arbitration forum, service provisions, controlling language, and enforcement mechanism shall be selected by qualified counsel experienced in PRC manufacturing agreements before execution. Proposed terms: _____________________________. If a bilingual version is executed, counsel shall specify which language controls.

In other words, AI is confessing that it cannot finish the dispute-resolution clause in a China contract. That blank is no minor omission. It covers the decisions that determine whether the agreement will do anything for you when something goes wrong, and at least one AI assistant is warning its users not to guess at them. The warning is sound, and it goes unheeded all the time.

I am talking here about manufacturing and sourcing contracts, where the factory and tooling are in mainland China and fast relief usually must come from a Chinese court. Distribution, licensing, employment, and investment agreements raise different questions and often call for different answers. Earlier this year, I wrote about the AI-generated China contracts people now send us, often asking us to quote a fee just to confirm that the dispute-resolution clause works.

The matters below are composites. I have changed identifying and nonessential details.

Companies That Fill In the Blank Themselves

We have had many companies come to us after doing exactly what that provision tells them not to do. The most recent company wanted us to send a demand letter to a Chinese factory. It had shared its product designs with a Hong Kong sourcing company under an NDA governed by Texas law, and the sourcing company had passed those designs to a factory in mainland China. After the relationship ended, the designs appeared online.

The company that contacted us was not seeking damages. It wanted a letter, in English and Chinese, directing the factory to stop using the designs, return the company’s tooling and unsold inventory, account for anything it could not return, and identify everyone else who had received the designs. In other words, it wanted us to stop the bleeding.

We regularly handle matters like this, but I told the company not to pay us to write that letter. I also asked, as I do with most companies that bring me a worthless China contract, whether a lawyer had drafted the agreement. The answer was the one I now hear most often. The company had drafted the contract using AI. Rather than take the unresolved dispute-resolution terms to a lawyer, it filled in the blank with what seemed most favorable to it and then asked AI whether its choice worked. AI gave it a reassuring answer.

Why the Demand Letter Would Have Had No Teeth

A demand letter is worth sending only when a real claim or some other practical leverage stands behind it. This company had something that looked like a claim and no leverage at all.

Start with the sourcing company. The NDA barred disclosure, and handing the designs to a factory looked like a breach. But suppose the company sued in the United States and won. A U.S. money judgment would still have required a separate Hong Kong proceeding against the sourcing company’s assets, and money was not what the company wanted. It wanted the factory stopped, its tooling back, and a list of everyone holding its designs. Coercive orders of that kind do not travel from a U.S. court to Hong Kong the way a judgment for a fixed sum does.

Had the agreement instead allowed suit in Hong Kong, the company would first have needed to establish the content of Texas law before a Hong Kong court. Either route would have cost too much to make the threat behind the letter credible, and neither would have delivered the relief the company wanted.

The NDA’s other gaps made matters worse. It did not stop the sourcing company from using the designs itself or bypassing the company to compete with it. A properly drafted NNN Agreement should have prohibited both. Nothing made the sourcing company responsible for the factory’s use or disclosure of the designs. The agreement also said nothing about ownership, use, or return of tooling and inventory, so the two demands the company cared most about rested on no contractual obligation at all.

The mainland China factory presented a different problem. It had signed nothing, and based on the facts I had, I saw no commercially viable claim against it. A sophisticated factory or its lawyer would see that as readily as we did. A demand letter gets results only when the recipient believes you can and will make it pay. We could not identify a credible threat worth the cost. A letter with nothing behind it can also provoke the very countermoves a company fears, as I explain in Why Demand Letters to China Suppliers Often Backfire.

The Blank Is the Hard Part

Look again at what the provision leaves open: governing law, forum, service, controlling language, and enforcement. Those choices turn on facts no one involved in the drafting had investigated or verified. Which company is actually signing, and where are its assets? Will you need a court to stop the factory quickly, or only an award of money later? Would a particular Chinese court have jurisdiction over that company? Could a judgment or award be enforced where the assets are?

Our lawyers often wait to draft the dispute-resolution clause until we understand the full scope of the deal. For that reason, it is often the last thing we draft in the contract. The questions are so fact-specific that we sometimes pull two to four of our international lawyers into a short meeting to decide what belongs in that single clause. I have written before about how often companies get this choice wrong and how much it costs them.

“Most favorable to me” is the wrong test. The right question is where you can obtain real relief and enforce it against the company whose conduct and assets matter. Most companies reach for what they know: their own law, their own courts, and their own language. In a contract with a Chinese company, those are often the worst choices. No treaty requires China and the United States to recognize each other’s civil judgments.

Chinese courts can recognize a U.S. judgment under reciprocity principles, but recognition requires a separate proceeding with its own cost, delay, and uncertainty. An English-controlling contract ordinarily must be translated for a Chinese proceeding, giving the factory one more issue to argue about. When an AI-drafted contract includes a Chinese version, the Chinese-language version is often its weakest part.

I have written about a company that chose U.S. law and then needed a Chinese court to stop its Chinese partner from misusing its trade secrets. Establishing the applicable U.S. law added enough cost and delay to defeat the purpose of seeking emergency relief.

Do not count on AI to stop you when you fill in the blank. A disclaimer is not a safety system. Ask an AI tool to defend a choice you have already made, and it will produce a plausible defense. That is exactly what happened to the company that came to us.

When the Factory Uses AI Too

The buyer is not the only one using AI. More and more often, a Western company sends its AI-drafted contract to its Chinese manufacturer, which uses another AI tool to revise it. The manufacturer’s tool proposes changes favoring the factory, and the factory adds them. By the time the finished agreement reaches me, it is long, bloated, difficult to follow, and internally inconsistent. I recently wrote about a 48-page contract whose provisions I traced in circles without ever resolving what the contract required. Contracts like that are expensive to enforce, and an experienced factory knows it.

Think about who is at the table. You have a Chinese manufacturer with no lawyer using AI and an American, Australian, or European buyer with no lawyer doing the same. AI is on both sides, eager to please both of them. The companies want the deal to go forward, so they compromise by putting both positions into the same document.

A contract that crossed my desk shows where this can lead. Its dispute-resolution clause required disputes to be heard in either Houston or Shanghai. If the parties could not agree on the city, an arbitrator in Singapore would choose between them. The American buyer had no connection to Texas, and the Chinese manufacturer had no connection to Shanghai. The buyer explained that it had selected those cities so neither side would enjoy a hometown advantage. But “fair to both sides” is no better a drafting standard than “most favorable to me.”

That was only the dispute-resolution clause. I found four sets of conflicting provisions scattered throughout the contract. My favorite involved delivery. One provision required delivery within 21 days after issuance of a purchase order and made any delay a breach. Another allowed 30 days from the factory’s receipt and acknowledgment of the purchase order, then imposed liquidated damages of 0.05 percent per day only after an additional ten-day grace period. It was easy to see what had happened: the American company’s AI had supplied the first provision, and the Chinese company’s AI had supplied the second. The parties left both in the final contract.

The conflicting clauses were perfect for the Chinese manufacturer. Before the buyer could reach the merits of any dispute, it would have to sort out a Singapore proceeding that the contract barely described. The clause did not say what law governed that proceeding, which institution would administer it, what rules would apply, who would appoint the arbitrator, or whether the arbitrator would decide only where the case belonged or the merits as well. It did not even establish that the parties had agreed to arbitrate the underlying dispute. The buyer’s first fight would not concern its designs or defective goods. It would be about how to decide where the real fight belonged.

The buyer could spend six months and $50,000 deciding that question alone. It would then have to bring its actual case in whichever city the arbitrator selected, litigate any jurisdictional objections the manufacturer raised there, and, if it ended up in Houston and won, start over in China to seek recognition of the judgment. I told the buyer that unless the amount at stake reached into the millions, I did not think we would take the case. It thanked me.

These supposed compromises usually favor the Chinese factory. The foreign buyer is far more likely to need the contract because it is the party exposed to IP misuse, late delivery, or defective products. A clause that makes litigation equally difficult for both sides therefore favors the side less likely to sue. A court is unlikely to rescue the buyer by rewriting a commercially inconvenient clause merely because enforcing it is expensive.

By the time the buyer needs its contract, one provision says A, another says not A, and a third says it could be A, B, C, or D. Litigation begins with a fight over what the contract means rather than what the factory did. That uncertainty increases costs before anyone reaches the merits and gives the factory leverage.

My other favorite involved a Latvian company that sent me an 18-page contract. I told the company that the draft omitted several essential provisions and was not worth reviewing in its current form. It returned with a 26-page version that still omitted every one of them.

The page count was not the real problem. The original draft combined what should have been two separate agreements, although it was not clear the company needed either one. The revised draft combined three, even though the third was neither necessary nor relevant. The AI left the real gaps untouched and filled eight additional pages with protections the company did not need.

Why a Bad China Contract Is Often Worse Than No Contract

The examples above are extreme, but the underlying problem is common. When we review AI-drafted China contracts, we almost always find provisions that do not belong and important protections that are missing. The documents are long and polished enough to look thorough, which makes their defects harder for a nonlawyer to see.

In my experience, about half of the AI-drafted contracts companies bring us provide no meaningful protection. The first harm they cause is false confidence. A company that knows it has no contract tends to act cautiously. It limits what it shares, presses for answers about who will actually make its products, and files for trademark and design protection early. A polished but useless contract encourages the company to relax those precautions, especially after AI has assured it that the contract is sound.

A bad contract also advertises weakness to the factory. If the agreement is governed by U.S. law, makes English controlling, and confines all disputes to a U.S. court, it signals that the buyer is unlikely to pursue the factory in China. Experienced factory managers and their lawyers will recognize and exploit that weakness.

Bad drafting can also leave the buyer worse off than it would have been without a contract. An exclusive U.S. forum clause can make it difficult or impossible to sue the factory in China, where its assets are. We also see supposed compromises that declare both the English and Chinese versions official but make the Chinese version controlling if the two conflict. The result is three contracts to analyze: the English version, the Chinese version, and whatever a court decides the two mean together.

The identity of the drafter matters for another reason. If a lawyer drafts a defective China contract, the client may have a malpractice claim. That claim might be its only realistic source of recovery. When a company drafts its agreement with AI, there is no drafting lawyer and no malpractice claim against the drafter. I am currently serving as an expert in a legal-malpractice case arising from a lawyer’s failure to identify the correct contracting party.

The AI-drafted contracts we see are almost always worse than agreements drafted by independent lawyers, including bad ones. If I had to rank the options from best to worst, the list would look like this:

1. A contract drafted by a good lawyer.
2. A contract drafted by a bad but honest lawyer.
3. No contract at all.
4. A contract drafted by a fake lawyer who is not secretly being paid by the other side.
5. A contract drafted by AI.
6. A contract drafted by a fake or dishonest lawyer who is being paid by the other side to tilt the agreement in its favor.

A well-drafted China contract is designed to influence conduct before a dispute begins. Its remedies matter because they are credible. A Chinese manufacturer that knows from the outset that the buyer can pursue practical relief in China has a much stronger reason to keep its promises.

We have drafted thousands of China manufacturing contracts, so we know how long proper drafting takes and can quote a flat fee for it. Disputes are different. Once a contract has failed, no one can predict how much work the resulting fight will require. To my knowledge, not one of our clients has had to sue a manufacturer for breaching one of our contracts.

What to Do If You Have Already Signed

If something has already gone wrong, start by not making it worse. Do not send the factory a threatening letter, whether from a lawyer or from you, until you know what leverage you have. If you have not protected your brand, designs, and tooling in China, an angry letter can prompt the factory to register your brand or designs in its own name, record those registrations with China Customs to block your exports, hold on to your molds and inventory, or make your product for someone else. Protect what you still can first.

For continuing or future production, try to replace the defective agreement before you share more information, pay for more tooling, or place another order. For many mainland manufacturing relationships, particularly those in which the factory holds your tooling or quick court relief may become necessary, that means an agreement with Chinese as the controlling language, Chinese law as the governing law, and disputes heard in a Chinese court with jurisdiction over the proper Chinese defendants. Those defendants are not always limited to the factory.

Other deals call for different terms. That is why the decision belongs with a lawyer who knows your situation rather than in a blank you fill in yourself. Determine whether you need an NNN Agreement, a manufacturing agreement, or both. If your brand is not registered in China, register it. A China trademark and a China contract usually solve different problems, and many companies need both.

I use AI every day to organize information, identify questions I have not asked, proofread what I have written (including this blog post), and test my thinking. It is useful when I can evaluate its output. Ask it to draft a specialized contract without the knowledge to check its work, and it will often build you a monster that reads beautifully, covers everything you thought to ask about, and confidently tells you that you got it right.

The most honest provision I have seen in any of these contracts is the one admitting that qualified counsel needs to make the decisions that matter most when something goes wrong. On that point, AI is right. Take it at its word.