Legal AI

Legal AI Software: A Buyer's Map Sorted by the Job You Need Done

Tomas ArkwrightLegal technology writer11 min read

Written by Tomas Arkwright Shopping for legal AI software is confusing because the label covers products that have almost nothing in common. A tool that reviews a supplier agreement, a tool that searches case law, a platform that sorts two million emails for a litigation hold and a chat assistant for a whole law firm can all be sold as AI legal software. They are built on different data, priced in different ways and judged by different tests. The quickest way to a good purchase is to name the job first and only then look at vendors. This guide maps the main kinds of legaltech AI by the work they do, names some of the products in each group as their makers describe them, and then gives you a test plan you can run on your own documents before you sign anything. ## Five jobs hiding under one label Most products fall into one of five buckets. Some vendors sell into two or three, but each product usually has a centre of gravity. | Job to be done | Typical buyer | What the AI works on | What a good result looks like | |---|---|---|---| | Contract review and drafting | In-house legal, commercial teams, transactional lawyers | Your contracts and templates | Correct issue spotting against your positions, usable redlines | | Legal research | Litigators, advisory lawyers, knowledge teams | Case law, statutes, commentary held by the vendor | Answers that cite real, current authority you can open | | E-discovery and document review | Litigation teams, investigators, regulators | Large sets of emails, chats and files from custodians | Relevant documents found early, privilege caught, defensible process | | Firm-wide legal AI assistants | Law firms and large legal departments | Whatever the lawyer uploads, plus connected sources | Fast first drafts, summaries and tables across many matters | | Practice and operations tools | Small firms, legal ops | Time, billing, intake and matter data | Less admin, cleaner records | Before you read any vendor page, write down which row your problem sits in. If you cannot choose one row, you probably have two purchases, not one. ## Contract review and drafting tools This is the most crowded corner of the market. The products read agreements, compare them with a playbook or a standard template, suggest edits and, in some cases, draft new contracts from a brief. A few examples, described in their makers' own words: - Spellbook calls itself a contract AI suite for in-house teams and law firms, with drafting, redlining, playbooks and a repository of signed contracts. Its site lists Microsoft Word among the tools it works with.

  • Luminance describes a contract intelligence platform with three product areas: a repository for finding risk and obligations across a portfolio, a negotiation product that applies precedent and playbooks, and workflows from intake to signature.
  • Contract lifecycle platforms add intake, approvals, signature and storage around the review step. They suit teams that need process control as much as review help. What separates good contract AI from average contract AI is not the model behind it. It is whether the tool knows your positions. A review that flags every limitation of liability clause as "risky" is noise. A review that says "this cap is twelve months of fees, your standard is twenty-four, here is your fallback wording" saves real time. ## Legal research tools, as a category Research products answer questions about the law: what a statute says, how courts have read it, which cases support a point. Their value depends on the content library behind them. A research tool with a thin or stale library will produce confident answers that rest on nothing. Three checks matter more than any feature list. First, every answer should cite sources you can click and read in full. Second, the tool should show you when the cited material was last current. Third, it should say plainly when it found nothing, rather than filling the gap with a plausible paragraph. Ask the vendor to run five questions you already know the answers to, including one where the honest answer is "there is no authority on this". ## E-discovery and large-scale document review E-discovery platforms handle collection, processing, review and production of large data sets for litigation and investigations. AI here means ranking documents by likely relevance, clustering similar material, suggesting coding decisions and summarising what a custodian knew and when. Two established names show how the category describes itself. Relativity presents Relativity aiR as its AI platform, with offerings for e-discovery, investigations, case strategy and contract review, and a conversational tool called Relativity claiR that answers questions about a case record with citations to source documents. Everlaw describes itself as AI built for litigation and investigations, and lists Deep Dive, Review Assistant, Predictive Coding and Clustering among its AI features. The test for this category is defensibility. You must be able to explain to a court or a regulator how documents were chosen and checked. Ask how the platform records reviewer decisions, how it samples for quality control and how it handles privilege. ## Firm-wide legal AI assistants This group grew fastest over the last two years. These are general platforms for lawyers: upload documents, ask questions, draft memos, build comparison tables across a data room, run repeatable workflows. Harvey calls itself one platform for legal AI work, serving law firms and in-house teams, and lists components such as Vault, Knowledge, Review Tables and Agents, along with apps for contract intelligence and intake. Legora describes collaborative AI for lawyers built around what it calls an agentic operating system for legal work, with Tabular Review, Workflows, Legal Research and Word and Outlook add-ins, serving firms and in-house teams. Legora's site also states that customer data is not used for AI training. These platforms are wide by design. That is their strength for a firm doing many kinds of work, and it is also why you should test them on the narrow task you care about most. A broad assistant may be very good at summarising a deposition and only average at applying your negotiation playbook. ## Practice and operations tools Practice management, time capture, billing and intake tools now ship AI features too: drafting time entries from calendar data, summarising matter activity, answering client intake forms. These help small firms most. They rarely touch the legal judgment at the centre of a matter, so the buying test is simpler: does it save admin time without creating billing errors? ## A test plan you can run on your own documents Demos use the vendor's best examples. Your decision should rest on yours. The plan below takes about two weeks and works for any of the five groups. ### Step 1: build a small, honest test set Pick 15 to 25 real documents that represent your normal work, not your hardest edge cases and not your cleanest templates. For contract tools, include at least a few agreements on the other side's paper. For research tools, write ten questions with known answers. Remove or mask anything you are not allowed to share, and check that your engagement letters or internal policies permit the test at all. ### Step 2: write the answer key before the trial starts For each document, note what a competent reviewer on your team would flag. This is the step most buyers skip, and without it every output looks impressive. ### Step 3: spot check accuracy, not just speed Score each output in three columns: issues the tool found that your key also has, issues your key has that the tool missed, and issues the tool raised that are wrong or irrelevant. Misses are the expensive column. A tool that is fast but misses an uncapped indemnity is a liability, not a saving. ### Step 4: check citations and grounding For every answer that points to a clause, a page or a case, open the source and confirm it says what the tool claims. Count how many pointers lead to the wrong place or to nothing. Grounding failures are the clearest signal of how much checking your team will still need to do. ### Step 5: read the data use terms Ask in writing whether your documents, prompts or outputs are used to train any model, whose model sits underneath, how long data is kept and where it is stored. Our page on whether AI legal software trains on your contracts walks through the wording to look for in vendor terms. ### Step 6: run the security questions Access controls, single sign-on, audit logs, retention and deletion, sub-processors and incident notice all belong in the review. The CLM vendor security questionnaire lists the questions and what a good answer sounds like; most of it applies to any legal AI vendor, not just contract platforms. ## Scoring sheet for comparing two or three vendors Keep the comparison simple enough that a partner or general counsel can read it in five minutes. | Criterion | Weight you choose | Vendor A | Vendor B | Notes | |---|---|---|---|---| | Issues caught against your answer key | High | | | Count misses separately | | Wrong or irrelevant flags | Medium | | | Noise costs reviewer time | | Citation and source accuracy | High | | | Open every pointer | | Fit with where your team works (Word, browser, email) | Medium | | | | | Data use and retention terms | Pass or fail | | | Get it in the contract | | Security answers | Pass or fail | | | | | Pricing model and minimums | Medium | | | Seats, usage, matters | Treat data use and security as gates. A product that fails either one should not reach the scoring stage. ## Questions that expose weak AI legaltech quickly - "Show me an answer where your tool said it did not know." A product that never declines is guessing.
  • "What happens when two of our playbook positions conflict?" Good contract tools let you rank them.
  • "Which parts of the output come from our documents and which from your own content?" The vendor should be able to separate the two.
  • "What does the tool do with a scanned PDF with poor image quality?" Real files are messy.
  • "Who at your company can see our documents, and under what process?" Expect names of roles, not reassurance. ## Matching budget and team size to a category Pricing models differ by group. Contract tools often price per seat. E-discovery platforms commonly price by data volume or by matter. Broad assistants for firms are usually quote-based enterprise deals. Many vendors in every group do not publish prices, so plan for a sales process and ask early about minimum seats and term length. A useful rule of thumb: buy the narrowest tool that solves the job in front of you, then widen only when you can name the second job. Teams that buy a broad platform hoping to find uses for it often end up paying for seats nobody opens. ## Where Adira fits on this map Adira sits in the first row of the table: contract work. It is a contract workspace, not a legal research product, an e-discovery platform or a practice management system, so if your job is in one of those rows you should look at the tools built for it. Inside contract work, Adira's AI Drafting Studio generates full contracts from a brief and stores each one as a structured clause tree rather than flat text. Its Document Intelligence Dashboard marks each position in an uploaded contract as favourable, neutral or unfavourable to your company and lists red flags, missing clauses, key dates and obligations. Clause-Level Editing takes a plain-English instruction, edits the right clauses and returns a diff, so a reviewer sees exactly what changed. Company Persona builds a profile of your company from its own documents and applies it to every draft. On data use, Adira does not train on customers' contract corpus, and it runs on Anthropic's Claude. Adira's pricing page lists storage, CRM and e-sign integrations by plan; confirm the specific system you need with Adira before you buy. Pricing is public on adiralaw.com: the Practice plan is USD 89 per seat per month billed annually with a three-seat minimum, the Firm plan is USD 179 per seat per month billed annually with a five-seat minimum, and Enterprise is custom. A 7-day free trial is long enough to run steps 1 to 4 of the test plan above on a small set of your own contracts.

Frequently asked questions

Is legal AI software one product category?
No. The label covers at least five different jobs: contract review and drafting, legal research, e-discovery, firm-wide assistants and practice tools. Each is judged by a different test, so decide which job you are buying for before comparing vendors.
How many documents do I need to test a legal AI tool properly?
Fifteen to twenty-five representative documents are usually enough to see patterns in misses and false flags, as long as you write the answer key before the trial starts.
What is the most important thing to check in a trial?
Misses. Count the issues your own reviewer would flag that the tool did not. Speed and polish matter far less than an uncapped indemnity or a missed renewal deadline.
Should I ask vendors whether they train on my documents?
Yes, in writing, and get the answer into the contract. Ask about documents, prompts and outputs separately, and ask which underlying model provider is used and what that provider's terms say.
Can one platform cover contract review and e-discovery?
Some vendors offer both, but the tests differ. E-discovery needs a defensible review process for large data sets; contract review needs accurate application of your negotiation positions. Test each job on its own.
Was this useful?

See how Adira drafts in your voice and reads contracts from your side.

Explore the showroom

Working through a contract like this? Weave is Adira’s free tool to read, mark up, and connect any contract in your browser — no account needed.

Try Weave — free

Book a demo

See Adira on your own contracts

A 30-minute walkthrough with our team. Tell us a little about you, then pick a time that suits you.

Prefer email? [email protected]