AI can draft a first-pass contract, summarize a 200-page deposition, or surface the right precedent in minutes. It can also invent a case that does not exist and hand it to you with a straight face — and lawyers who filed those fake citations have been sanctioned and named in the press. So this guide treats AI the way a careful practitioner should: a powerful assistant for the grunt work, never an unsupervised authority. Below are the tools actually worth your money, split honestly between cheap everyday assistants and expensive enterprise platforms, with real pricing, what each is bad at, and the one rule you cannot skip. New to this? Start with our guide on how to use ChatGPT effectively.
The math: Time to set up ~1 day · Tasks automated: summaries, first drafts, research triage · Weekly time reclaimed: several hours. Pricing changes — confirm on each vendor's site, and verify every citation regardless of tool.
Here's the thing: a general chatbot predicts plausible text, and "a plausible-sounding case citation" is exactly the kind of thing it will fabricate. That makes legal AI a special category, because the cost of a confident error isn't an awkward email — it's a sanction, a malpractice exposure, or a breach of confidentiality. The tools that belong in a practice fall into two camps: cheap, general assistants that are safe for language (emails, summaries, first drafts) and purpose-built platforms grounded in real legal databases that are safe for law (research with checkable citations). Mixing those up is how lawyers get into trouble.
Before choosing tools, know where the billable and non-billable hours leak. Four tasks dominate, and they map cleanly onto what AI does well:
| Tool | Best for | Starting price | Standout | The honest catch |
|---|---|---|---|---|
| ChatGPT / Claude | Drafts, summaries, plain English | $0–$20/mo | Fast, flexible, cheap | Fabricates citations — never cite from it |
| Spellbook | Contract review in Word | ~$20–$40+/user/mo | Lives inside Microsoft Word | Custom quotes; real work needs higher tiers |
| Lexis+ AI | Legal research | Subscription | Grounded in real case law | Still has a measurable error rate |
| CoCounsel (TR) | Research + workflows | Bundled w/ Westlaw | Tight Westlaw integration | Hard to price standalone |
| Harvey / Legora | Large-firm platforms | Custom, multi-seat | Built for AmLaw workflows | Seat minimums price out small firms |
Organized by the task eating your time, not alphabetically.
For client emails, first-draft clauses, summarizing long documents, and translating legalese into plain English for a client, a general assistant is fast and cheap. What it must never do is supply case citations.
Who it fits: any lawyer who writes and reads a lot — which is all of them. What it does well: turns a long document into a neat summary of facts, parties, dates, and obligations; drafts routine correspondence; explains a concept simply for a client. Where it falls short: it fabricates convincing, fake citations and will occasionally state law that doesn't exist. It also shouldn't see privileged material in a public tool. Pricing: free, or $20/month.
Spellbook runs where lawyers actually draft — inside Word — suggesting clauses, flagging missing terms, and redlining inline. That makes it far more practical for transactional work than a general chatbot, because it's built around contract structure.
Who it fits: solo and small-firm transactional lawyers doing regular contract drafting and review. What it does well: in-document redlining and clause suggestions without leaving Word. Where it falls short: it doesn't publish fixed pricing — entry tiers are reported around $20–$40/user/month, but serious work generally needs mid-tier plans or higher, so budget above the headline. Pricing: custom quotes. We compare it against LegalOn, Robin AI, and Luminance in our guide to the best AI contract review software.
This is the category where a purpose-built tool is essential. Lexis+ AI and CoCounsel are grounded in real legal databases, so their citations are real and checkable — a general chatbot's are not.
Who it fits: any lawyer doing enough research that verified AI assistance saves real hours. What it does well: fast, grounded research and document workflows; CoCounsel integrates tightly if you're already on Westlaw/Practical Law. Where it falls short: "grounded" doesn't mean perfect — independent Stanford testing put Lexis+ AI's error rate around 17% and Westlaw AI around 34%. That tells you two things: these are far better than a general LLM for law, and you still verify every cite. Pricing: Lexis+ AI is a subscription; CoCounsel is typically bundled into existing contracts. Before relying on any of them, read our evidence-based explainer on whether AI legal research is safe — it covers the measured error rates and the sanctions to avoid.
These platforms are built for AmLaw-scale firms, with workflows designed around the billable hour.
Who it fits: large firms with the matter volume and budget to justify them. What it does well: deep, firm-wide legal workflows. Where it falls short: pricing is custom and assumes multi-seat budgets — Harvey, for instance, has been reported to require large seat minimums, which prices out small practices entirely. Know they exist; they're rarely the right call for a solo or small firm. Pricing: custom, multi-seat. If you run a lean practice, our guide to AI for small law firms covers the tools actually priced for you.
The honest picture depends entirely on firm size. A solo or small firm can run a $0–$20 general assistant plus Spellbook for contracts plus its existing research subscription — modest money for a real productivity lift. A large firm evaluating Harvey or Legora is in a different universe, where multi-seat minimums can run into six figures a year before anyone drafts a document. The lesson: match the tool to your size. Most small practices never need the enterprise tier, and buying it would be the most expensive mistake on this page.
Be honest about your needs. If you rarely draft contracts, you don't need Spellbook — a general assistant plus careful human review covers occasional work. If your research volume is low, your existing Westlaw or Lexis subscription is enough without adding an AI tier. And no small firm should be shopping enterprise platforms; that's solving a problem you don't have. Start with the cheapest tool that fits the task and add up only when the volume justifies it.
Reading about AI changes nothing; piloting it on low-risk work this week builds real confidence. Do it in this order:
Yes for drafting, summarizing, and explaining — with human review. It is not safe to rely on a general chatbot for citations, which it fabricates. Use purpose-built research tools for case law and verify everything.
ChatGPT or Claude for everyday drafting and summaries, plus Spellbook if you handle contracts. That covers most day-to-day work affordably; add a dedicated research tool as your volume grows.
Purpose-built tools like Lexis+ AI are grounded in real databases and far more reliable than general AI, but independent testing still shows meaningful error rates. Treat it as a fast first pass, then confirm every citation yourself.
It has led to court sanctions, public embarrassment, and malpractice exposure for real lawyers. This is the single biggest risk of AI in law, and it's entirely avoidable by verifying every cite in a genuine legal database before filing.
No. It compresses the grunt work — review, summarizing, first drafts — but judgment, advocacy, negotiation, and accountability to the client and court remain human. It changes how you bill, not whether you're needed.