Introduction
Can AI draft a contract well enough to actually sign? After 17 years of drafting NDAs, founder agreements, and privacy policies for businesses across India, my honest answer is: yes, and also, absolutely not — depending entirely on which part of the job you’re asking it to do. AI is now genuinely good at producing a clean first draft in seconds. It is also, provably, capable of inventing case law, missing jurisdiction-specific requirements, and getting a founder sued or sanctioned with total confidence. Both of those things are true at the same time, and any lawyer telling you otherwise — in either direction — is selling you something.
This isn’t a hot take. It’s what the actual research, including some fairly alarming court sanctions data from 2025 and 2026, says about where AI belongs in contract drafting and where it categorically doesn’t.
Where AI Genuinely Excels: Speed and First Drafts
Let’s give credit where it’s due, because the data here is stronger than most lawyers want to admit. A well-known LawGeex study comparing AI against experienced attorneys on NDA review found AI achieved 94% accuracy compared to an 85% average for the human lawyers, and cut review time from 92 minutes down to 26 seconds. A more recent 2025 LegalBenchmarks.ai study found that human lawyers produced reliable first drafts about 56.7% of the time, while the top AI system reached 73.3% — and delivered it almost instantly.
Even more interesting: in scenarios carrying real legal risk, specialized legal AI tools raised explicit risk warnings in 83% of outputs, compared to 55% for general-purpose tools — and in that same study, the human lawyers raised no risk warning at all in several of those scenarios. That’s not a small finding. It means AI, used correctly, can catch things an experienced professional misses on a busy day.
For routine, repeatable documents — a standard NDA, a basic freelance agreement, a first-pass privacy policy — AI is a genuinely useful starting point. I’ve said as much when writing about what actually needs to go into a freelance agreement: a template, AI-generated or otherwise, is a starting point for the conversation, never the end of it.
Where AI Fails — Sometimes Catastrophically
Here’s where the honest part of “an honest lawyer’s take” comes in.
The Hallucination Problem Is Not Theoretical Anymore
A 2024 Stanford study on legal AI reliability found hallucination rates of 43% for GPT-4 on legal queries, 33% for Westlaw’s AI research tool, and 17% for Lexis+ — and a separate, broader Stanford study testing general-purpose models on over 800,000 verifiable legal questions found hallucination rates between 58% and 88%. Even the paid, purpose-built legal AI tools that are supposedly “safe” hallucinate on somewhere between one in six and one in three queries.
This isn’t just a research-paper statistic. It’s already cost real lawyers their careers. In the now-infamous Mata v. Avianca case, attorneys submitted six completely fabricated court decisions generated by ChatGPT into a federal brief — and when questioned, the same tool confirmed its own fabrications were real. The presiding judge sanctioned the attorneys $5,000 and published an opinion that bar associations across the US still cite in ethics guidance. Since then, a database tracking these incidents has logged over 1,500 court decisions worldwide involving AI-hallucinated content, with penalties escalating fast: a record $110,000 fine in Oregon in 2025 for 23 fabricated citations, and in 2026, the first indefinite bar suspension in US history tied directly to AI-generated fabrications.
The uncomfortable truth the data reveals is that “just use a specialized legal AI tool” isn’t a safe fix on its own — Lexis+ and Westlaw’s own AI products have both appeared in sanctions data alongside general chatbots. Verification by a human being who actually understands the law is still the only real safeguard.
AI Doesn’t Know Which Jurisdiction It’s Writing For
This is, in my experience, the single biggest blind spot for founders drafting contracts in India using US-trained AI tools. Most large language models are trained overwhelmingly on US and UK legal content. Ask one to draft a founder agreement, and it will often produce clauses that assume a Delaware LLC structure, US-style at-will employment, or dispute resolution norms that simply don’t map onto how founder agreements and shareholders’ agreements actually work under Indian company law. It won’t flag India-specific requirements like proper stamp duty on the agreement, registration requirements, or jurisdiction and governing-law clauses that actually hold up in an Indian court.
The same blind spot shows up in privacy documentation. An AI tool asked to draft a “privacy policy” will very often produce something modeled on GDPR or CCPA — not the specific obligations under India’s DPDP Act, 2023, which has its own consent, data-retention, and breach-notification requirements. A founder who ships that document believing it’s “done” is carrying real compliance risk they don’t know they have.
AI Struggles With Context, Intent, and Multi-Source Judgment
The 2025 LegalBenchmarks study is genuinely useful here because it isolated exactly where humans still win. In a task requiring a lawyer to integrate a template, a term sheet, and an email thread — including pulling accurate company details from a screenshot — only the human lawyer got it right. Every AI output in that test contained incomplete or inaccurate party information. Human lawyers also demonstrated clear advantages in interpreting what a client actually wants, and in avoiding unnecessary concessions to the other side during negotiation — both of which require reading intent and relationship dynamics that AI, so far, cannot reliably do.
This matches exactly what I see when advising founders on term sheets or remote hiring agreements across Indian states: the hard part was never producing text. It was understanding what the founder actually needed protected, and negotiating that intent into language both sides would honor.
My Honest Framework for Using AI in Contract Drafting
Here’s how I’d actually advise a founder to use AI in this process, having tested it extensively against my own drafting:
- Use AI for the first draft of routine, low-stakes documents — a basic NDA skeleton, a starting freelance agreement, a rough outline of clauses to consider.
- Never use AI as the final word on anything jurisdiction-specific. Indian contract law, the Indian Contract Act 1872, stamp duty rules, and sector-specific compliance like the DPDP Act need a human who actually practices in that jurisdiction.
- Never trust an AI-cited case or provision without independently verifying it. Given hallucination rates as high as 88% on some benchmarks, a citation “sounding right” is not verification.
- Have a human review anything that will actually be signed, particularly founder agreements, equity documents, and anything involving breach-of-contract remedies — these are exactly the high-stakes, hard-to-reverse documents where a mistake is expensive to unwind.
- Treat the contract as a trust-building document, not just a legal formality — something I’ve written about in more depth in Contracts Are Marketing Tools, Not Just Legal Paperwork — and AI-generated boilerplate rarely reads like it was written with the other party in mind.
Conclusion: AI Is a Fast Assistant, Not a Replacement Lawyer
Can AI draft a contract? Yes, faster than any human, and for routine documents, sometimes more consistently. Can it draft one you should sign without a qualified human reviewing it for your actual jurisdiction, your actual intent, and your actual risk? The research — and a growing pile of five- and six-figure court sanctions — says no, not yet, and not without real consequences for the founder who assumed otherwise. The honest answer is the boring one: use AI to move fast, and use a lawyer to make sure fast didn’t cost you something expensive later. If you’d like a founder agreement, NDA, or privacy policy actually reviewed for Indian law rather than assumed to be compliant, you can reach me directly here, or read more of my writing on contract drafting and startup legal risk.
Frequently Asked Questions (FAQ)
1. Can AI draft a legally valid contract in India? AI can generate contract-like text, but validity under Indian law depends on proper stamping, registration where required, and compliance with the Indian Contract Act, 1872 — none of which AI reliably verifies. A human lawyer should review any AI-drafted contract before signing.
2. Is it safe to use ChatGPT to write an NDA or founder agreement? It can be a reasonable starting point for a first draft, but ChatGPT and similar general-purpose tools are not trained specifically on Indian law and have documented hallucination rates as high as 43% on legal queries, so the output should always be reviewed by a qualified lawyer.
3. What is an AI hallucination in a legal context? It’s when an AI tool generates a confident-sounding but false statement — a fabricated case citation, a misstated law, or a nonexistent clause requirement. Courts have sanctioned lawyers heavily for filing AI-hallucinated content without verifying it first.
4. Are specialized legal AI tools safer than general chatbots like ChatGPT? They’re meaningfully safer but not risk-free. Stanford research found purpose-built legal AI tools like Lexis+ and Westlaw’s AI research hallucinate at rates between 17% and 33%, compared to around 43% for general-purpose models — safer, but still far from zero.
5. What are the biggest risks of using AI to draft contracts? The main risks are jurisdiction blindness (AI trained mostly on US/UK law missing Indian-specific requirements), fabricated legal citations or clauses, missing context about the actual business relationship, and a false sense of security that the document is “done” without human review.
6. Has a lawyer ever gotten in trouble for using AI to draft legal documents? Yes, repeatedly. The best-known case is Mata v. Avianca, where attorneys filed six fabricated AI-generated court cases and were sanctioned. Since then, over 1,500 similar incidents have been logged globally, with penalties reaching over $100,000 in some cases.
7. Can AI understand the specific business context behind a contract? Not reliably. Research shows AI struggles to integrate multiple sources of context — like a term sheet, an email thread, and a template — as accurately as a human lawyer who understands the client’s actual intent and negotiating position.
8. Should startups use AI-generated privacy policies? Only as a starting draft. AI-generated privacy policies often default to GDPR or CCPA assumptions and can miss specific requirements under India’s DPDP Act, 2023, leaving a business non-compliant without realizing it.
9. What should founders do instead of fully relying on AI for contracts? Use AI to speed up first drafts of routine documents, then have a lawyer familiar with the relevant jurisdiction review anything that will actually be signed — especially founder agreements, equity documents, and privacy policies.
10. Will AI eventually replace lawyers for contract drafting entirely? Unlikely in the near term. Current research shows AI is closing the gap on drafting speed and even some risk-flagging, but still lags significantly behind humans on contextual judgment, negotiation strategy, and jurisdiction-specific accuracy — the parts of the job that actually protect a client.
Internal linking note: All internal links above point to real, live pages on parvezali.me. This piece pairs naturally with “Contracts Are Marketing Tools, Not Just Legal Paperwork” and the DPDP Act compliance checklist — consider linking back to this post from both.
Keyword density note: This draft targets roughly 1% density for the focus keyword “can AI draft a contract” and its close variants (AI contract drafting, AI legal drafting). Run it through Rank Math’s live content analysis after pasting into WordPress, since exact density shifts slightly with final formatting and any edits you make.