Digital contract drafting with artificial intelligence: opportunities, duties, and liability

Contract creation is changing as profoundly as it did when computers and word processing arrived. No-code platforms, smart online forms, and above all generative artificial intelligence promise contract drafts in minutes that once took hours. For businesses that is attractive because contracts are produced faster and more cheaply. At the same time new legal questions arise: Who is liable for a defective contract produced by AI? When does using such tools become a service subject to legal-services licensing rules? And what does the EU AI Act require?

This article places digital contract drafting in a legal framework and explains what matters in practice so efficiency gains do not become liability risks.

From form books to AI contracts

Contract templates have a long tradition. Even before printing, models were copied by hand, later pre-printed, amended, and tailored. With computers and printers, contracts were copied from older files and adapted. What all these steps share: a complete model is taken first and then trimmed to the specific case.

Using proven models makes sense. If the case is not an atypical special situation, there is no need to reinvent the wheel. Templates save resources and safeguard quality. With generative AI, however, the logic shifts: the system no longer relies only on a stored template but independently produces wording from training data, internal templates, statutes, and case law. That opens possibilities but also moves legal control from choosing the right template to critically reviewing freely generated text.

Three technologies that are often confused

When people talk about digital contract drafting they often mean very different tools. For legal assessment a clear distinction helps.

Rule-based software and no-code platforms work deterministically. The user stores text blocks and defines via query logic how they are assembled with case data into a contract. A case not foreseen in the logic cannot be solved automatically. Programming effort therefore pays off mainly for recurring standard cases.

Generative AI works probabilistically. It produces probability-based results without every constellation being individually programmed and can thus handle previously unknown cases. The outcome is not necessarily correct and not fully predictable.

Smart contracts are different again: they automatically execute predefined if-then conditions, often via blockchain. They do not create a contract but perform individual duties under an existing agreement.

The distinction is not academic. It determines which AI Act obligations apply, who bears responsibility for errors, and whether a legal case-by-case assessment takes place at all.

Why a good prompt alone does not make a good contract

Contract drafting is part of preventive legal care. Unlike assessing a fixed fact pattern after the event, the view is directed at an uncertain future. Clear yes-or-no answers give way to many design options with pros and cons to be weighed. That is where a short prompt reaches its limits.

Before a usable clause exists, numerous framework conditions must be clear: Which law applies, and should the CISG be excluded under German law? Is it a B2B or consumer contract? Are general terms subject to strict content control? Should a clause be balanced or as one-sided as possible for one party, and what risk of invalidity is acceptable?

Correct legal classification of the business model also matters. What counts is not the contract heading but the content of the performance owed. Whether a SaaS agreement is characterised as a lease, whether care services are service or works contracts, or whether a project follows classic or agile methods has immediate effects on warranty, liability, and termination. A wrong classification can invalidate entire sets of clauses.

Professional tools address this via a playbook and preconfigured settings where company policies and standard clauses are stored. That preparatory work—not the individual prompt—holds the real legal know-how. Freely available, vetted, always up-to-date contract models do not yet exist, so the decisive settings must come from the user.

Hallucinations and automation bias

Even carefully guided AI still produces hallucinations: convincing but invented statements on the law or fabricated citations. Studies of leading legal research tools show that even specialised systems return incorrect results in a significant share of queries. Every formulation and every cited source should therefore be checked.

Automation bias adds difficulty—the tendency to adopt a machine suggestion uncritically. The AI Act explicitly addresses this and requires effective human oversight for high-risk systems. Applied to contract drafting: final human review must not be a formality but genuine professional review with real decision-making leeway. Relying on AI especially where current legislation, case law, or supervisory practice matter—which the system may lack—is negligent.

Who is liable if AI produces a defective contract?

If AI produces a contract with invalid or disadvantageous clauses and damage results, the AI itself is not a liable party. Attribution as a vicarious agent under German Civil Code § 278 fails because AI is not a person. Liability may attach to the person who deployed AI and did not review its output. Generated results should undergo final review; the more opaque the system and the less certain its data currency, the more extensive review must be.

A separate issue is conclusion of contract by AI. If AI communicates with a counterparty and concludes an unintended contract, the operator must generally attribute the declaration, comparable to an automated electronic declaration of intent. In Canada a company was already held responsible for incorrect information from its chatbot. Where errors stem from human failure, rescission of the declaration may be considered.

What companies should do now

Using AI in contract drafting makes sense when it is structured and safeguarded. These steps have proven effective:

• Define guardrails: Build a playbook with company policies, standard clauses, and custom instructions instead of relying on single prompts.
• Ensure genuine final review: Assign clear responsibility for legal sign-off so human control does not become a formality.
• Check data protection: Enter into a data processing agreement, clarify data flows, and exclude use of inputs for third-party purposes.
• AI literacy and policy: Train staff and adopt an internal AI policy to meet the competence duty in force since 2025.
• Documentation and labelling: Keep decision paths traceable and observe transparency duties for AI-generated content.

Conclusion

Artificial intelligence is changing contract drafting but does not replace legal responsibility. The main task shifts: away from drafting every clause, toward precisely defining objectives, carefully parameterising settings, and critically reviewing the result. Used this way, AI adds speed and efficiency without sacrificing legal certainty.