Skip to content
All insights

AI That Assembles Vetted Clauses vs AI That Invents Them (2026)

On this page
  1. How 99 Data Rooms AI Legal Drafting works
  2. Why "inventing" clauses is risky
  3. What to look for in a drafting tool
  4. From safe draft to controlled sharing in 99 Data Rooms
  5. Draft from vetted clauses for free
  6. Sources

When it comes to AI legal drafting safety, the single most important distinction is this: some tools assemble documents from vetted, human-reviewed clauses, while others generate legal wording from scratch with a language model. The first approach gives you text a lawyer has already checked, selected to fit your answers. The second gives you plausible-sounding wording that no one has verified and that can be subtly, or badly, wrong. The AI Legal Drafting feature in 99 Data Rooms is firmly in the first camp: it is live in the product, and it assembles vetted England and Wales clauses by ID, and it never invents wording. This guide explains why that difference matters and what to look for. It is general information, not legal advice.

The rise of general-purpose AI has made it trivial to ask a chatbot to "write me an NDA". The output looks like a contract. The problem is that looking like a contract and being an enforceable, current, jurisdiction-correct contract are not the same thing, and the gap between them is exactly where risk lives.

Start with the feature itself, because it is the clearest example of the safer approach. The AI Legal Drafting feature in 99 Data Rooms, live and available in the product, does not write law. It works from a library of vetted England and Wales templates, seventeen of them, each built and reviewed by people before it ever reaches you. When you draft, the assistant asks a few plain questions: who the parties are, what the document is for, how long it should run, and so on. Based on your answers, it selects the appropriate vetted clauses by their ID and assembles them into a finished draft.

The thing it deliberately does not do is generate clause wording itself. A language model is not asked to compose a limitation of liability paragraph or invent an indemnity. Those already exist, checked, in the library, and the AI's job is selection and assembly, not authorship. That is what makes the output predictable: every clause in your document is one a human has already vetted, and the same inputs produce the same trusted text.

The honest framing, which we keep on every drafted document, is that this is a starting point, not legal advice. It gives you a solid, current England and Wales baseline fast. Anything with real money or real risk attached is still worth a human review before you rely on it. You can browse the full AI Legal Drafting template library to see the documents available, including the mutual NDA, the founder agreement and others.

Why "inventing" clauses is risky

The alternative approach, letting a language model generate legal wording on demand, carries risks that are easy to underestimate because the output reads so fluently.

The first is hallucination. Language models produce text that is statistically likely, not text that is verified. They can invent clauses that sound authoritative but misstate the law, reference statutes that do not apply, or cite cases that do not exist. There have been well-publicised instances of AI-generated legal documents and submissions containing fabricated citations. In a contract, an invented or garbled clause might do nothing, or it might create an obligation you never intended.

The second is jurisdiction drift. A great deal of the legal text a general model has learned from is American. Ask it for an NDA and you may get wording built around US concepts, references to state law, or terminology that has no equivalent in England and Wales. It looks right to a non-lawyer and is quietly wrong for your jurisdiction. This is the same provenance problem that plagues free download sites, which we cover in our comparison of vetted versus cheap NDA template sites.

The third is staleness and inconsistency. A generated clause is not tied to a reviewed, maintained source, so you have no way of knowing whether it reflects current law, and two runs of the same prompt can produce materially different documents. With an assembled-from-vetted-clauses approach, by contrast, the clauses are maintained centrally, so when the underlying template is updated everyone drafting from it gets the current version.

None of this means AI has no place in legal drafting. It means the safe role for AI is to route you to the right vetted wording quickly, not to write the wording. Understanding the difference is the core of AI legal drafting safety, and it is why our explainer on what an NDA is and when you need one stresses that the concept is easy but the drafting is where things fail.

What to look for in a drafting tool

If you are evaluating any AI-assisted legal drafting tool, a few questions cut straight to whether it is safe to rely on.

Ask where the clauses come from. If the answer is "the model writes them", treat the output as an unreviewed draft at best. If the answer is "a library of clauses reviewed by people", you are on firmer ground. Ask which jurisdiction the templates are written for. For a UK business you want England and Wales wording, not repurposed US text. Ask whether the templates are maintained and updated, and whether the same inputs reliably produce the same document. And check how the tool describes its own output: a responsible one tells you plainly that it is a starting point and not legal advice, rather than implying it replaces a solicitor.

Finally, look at what happens after drafting. A document you cannot then control is only half the job. The ability to gate, track, sign and revoke the document you have just drafted is what turns a drafting tool into something you can actually run a deal on. For genuinely low-stakes needs, a good free vetted template such as our free NDA template for England and Wales may be all you need; for anything heavier, the assemble-from-vetted-clauses route plus a human review is the sensible baseline.

From safe draft to controlled sharing in 99 Data Rooms

The reason 99 Data Rooms takes the assembly approach is that drafting is only the first step of a journey the whole product is built around. A safely drafted document is worth far more when it never leaves your control afterward.

Once the AI has assembled your draft from vetted clauses, it stays in the platform. You can gate it so only a verified email holder with a one-time code can open it, share it as a tracked and revocable link rather than an attachment, and see page-by-page analytics on who read it. When it is ready to execute, you send it for signature in the browser, and the signed PDF comes back with an audit certificate recording who signed, when, their IP and intent, plus a SHA-256 fingerprint. Electronic signatures are admissible for most commercial documents in England and Wales, with exceptions such as deeds and land transfers (see Sources), which we cover in our guide on electronic signatures in England and Wales. If anything changes, one click revokes access. Drafted from vetted clauses, then gated, tracked, signed and revocable: safe wording and safe handling, in one place.

Draft from vetted clauses for free

You can draft from vetted England and Wales clauses, keep the draft, and gate, track, sign or revoke it inside 99 Data Rooms. The free tier is a real tier, not a trial: three rooms, twenty-five active links, forever, no card required. Start for free, assemble the document you need from vetted clauses, and move up only when you want unlimited links, NDA gating or watermarking. The wider platform is in beta and improving fast, but the AI Legal Drafting feature is live and the loop from safe draft to signed already works.

Sources

Questions, answered
Does 99 Data Rooms use AI to write legal clauses?

No. The AI Legal Drafting feature assembles documents from a library of vetted England and Wales clauses, selecting them by ID based on your answers. It does not generate or invent clause wording. The feature is live in the product.

Is an AI-assembled document legal advice?

No. It is a starting point that gives you a solid, current England and Wales baseline. It is not legal advice, and anything material is worth reviewing with a qualified adviser before you rely on it. This article is general information, not legal advice.

Why is AI that invents clauses risky?

Language models produce statistically likely text, not verified law. They can hallucinate clauses, cite the wrong or non-existent authorities, and drift into US wording that does not fit England and Wales. Assembling from vetted clauses avoids all three problems.

Can I trust a document a general chatbot writes for me?

Treat it as an unreviewed draft at best. You cannot easily tell whether the wording is current, correct for your jurisdiction, or internally consistent. A vetted-clause approach plus a human review is far safer for anything that matters.

What can I draft in 99 Data Rooms?

There are seventeen vetted England and Wales templates, spanning NDAs, founder and employment agreements, board minutes and resolutions, due diligence request lists and more. You can browse them in the template library and assemble a draft in minutes.

Keep reading