Iro AIWhat you'll be able to do
- Summarize a long contract or filing into plain-English bullet points with the key risks flagged
- Draft a first-pass clause, letter, or client email you can then edit and own
- Turn dense statute or jargon into an explanation a client actually understands
- Write prompts that constrain scope and force the model to flag its own uncertainty
- Catch invented citations and hallucinated case law before they ever reach a filing
Inside the path
A focused set of five-minute lessons — each one ends with a hands-on exercise, not a quiz you can guess.
Summarize without missing the risk 6 min
Prompt patterns for turning long agreements and filings into plain-English summaries that surface the clauses that matter.
First-draft clauses and correspondence 5 min
Get usable first drafts of clauses, engagement letters, and client emails by giving the model your constraints and voice.
Explain the law in plain English 5 min
Translate statute, jargon, and dense provisions into something a client or junior can follow.
The hallucinated-citation trap 6 min
Why AI invents realistic-looking case names and how to make verification a non-negotiable step.
Research starting points, not answers 5 min
Use AI to map issues and generate leads to check — then confirm every one against a real source.
Try a sample exercise
This is the kind of card you'd practice inside Iro — you do the thinking, then get feedback.
◆ Sample exercise · Prompt practice
A client sends a 40-page vendor agreement and asks what the indemnification clause actually means. You want AI to speed this up without inviting made-up law into your review.
Your task: Pick the prompt that saves time safely — and doesn't tempt the model to hallucinate.
- "Find all the cases that support challenging this indemnification clause and cite them."
- "Summarize this indemnification clause in plain English in 5 bullet points. List any terms that are ambiguous, one-sided, or unusual, and flag anything you're not certain about. Do not cite cases or statutes — I'll verify the law myself."
- "Is this indemnification clause enforceable? Give me a yes or no."
- "Rewrite this entire agreement to favor my client."
See why the second prompt wins
The winning prompt uses four techniques that keep AI in its lane: it constrains the scope to one clause and a plain-English summary, sets a clear output format (5 bullets plus a list of ambiguous or one-sided terms), demands uncertainty flags so you know exactly what to double-check, and explicitly forbids invented citations — the single biggest AI risk in legal work. The "find cases and cite them" option is how lawyers end up filing fake case law; models generate realistic-looking citations that don't exist. Asking "is it enforceable" pushes the model to give a legal conclusion it isn't qualified to make, and "rewrite the whole agreement" is too broad to verify line by line. In Iro you'd write your own version and get feedback on scope, format, and where verification has to happen.
Where AI helps a lawyer — and where it's dangerous
The safe, high-value uses of AI in legal work all share one trait: you can verify the output against the source in front of you. Summarizing a document you have, drafting a clause you'll edit, or explaining a concept in plainer words are all things you can check line by line. That's where AI earns its keep — reading fast, drafting fast, and freeing you to do the judgment work.
The dangerous uses are the ones where the model fills a gap you can't see. Ask it for supporting case law and it will produce citations that look perfect and don't exist. Ask it whether something is enforceable and it will give a confident legal conclusion with no accountability behind it. The rule that keeps you safe is simple: AI drafts and summarizes; you verify and advise. Nothing with an invented fact in it should ever reach a client, a court, or opposing counsel.
Verification is the skill, not the prompt
A good legal AI habit isn't a magic prompt — it's a workflow with a verification step baked in:
- Constrain the task: ask for a summary, a first draft, or an explanation of a document you already have, not for new facts or law.
- Force uncertainty flags: tell the model to mark anything ambiguous or anything it isn't sure of, so your review time goes where the risk is.
- Ban invented citations: tell it not to cite cases or statutes, or treat any citation it gives as an unverified lead to confirm in a real database.
- Keep PII and privilege in mind: be careful what confidential client material you paste into a general tool, and follow your firm's policy.
Do this and AI becomes a genuine force multiplier without ever putting your name behind something that isn't true. This page is about productivity, not legal advice — the responsibility for the law always stays with you.