How to evaluate a legal AI drafting tool: a checklist
How to evaluate a legal AI drafting tool is a question more transactional firms are asking as the market fills with options, each making similar-sounding claims. Most demos look impressive. The test is what happens in ordinary daily use, six months after you sign.
Does it work where your lawyers already work?
The first question is the simplest. Does the tool live inside Microsoft Word, or does it require lawyers to use a separate portal? This distinction matters more to adoption than any feature comparison. Lawyers under deadline pressure will not switch applications to get a clause suggestion. A tool that demands a change in workflow gets abandoned quietly, without anyone formally deciding to stop using it.
Ask the vendor: what does the Word integration actually look like? Can you see a live demonstration in a real document, not a screen recording of a controlled environment?
What does it actually learn from?
There are two possible answers here, and they lead to very different tools. Either the product drafts from generic training data gathered across the internet, or it drafts from your firm's own precedents. The second is considerably more useful for a transactional practice with house styles, preferred clause positions and client-specific requirements.
Ask specifically: when this tool suggests a clause, what corpus is that suggestion drawn from? Is it your documents, or a general dataset? Can you verify the answer with a concrete example using a document type your team handles regularly?
Does it build something your firm owns?
Most AI tools reset at the end of each session. The system learns nothing from the clause choices your lawyers make, nothing from the negotiation positions they take or concede, nothing from patterns across dozens of similar transactions. That is a significant limitation for a practice with repeat counterparties and a consistent approach to deal terms.
Look for tools that capture negotiation history, clause decisions and deal outcomes as a byproduct of normal work, producing a firm-owned record that survives staff turnover and compounds in value over time.
Is pricing published and all-in?
Legal tech pricing is often opaque until you are deep into a sales process. Ask for a per-user all-in figure before you invest time in a full evaluation. Scriptum publishes its pricing, starting at £79 per user per month on the Starter plan, so you can assess the financial case before any sales conversation begins.
A clear evaluation framework protects you from tools that perform well in demos but poorly in practice. If you would like to put Scriptum through this checklist directly, book a demo and we will walk through a live document with your own precedents.