Legal AI for Lawyers: AI for Law Firms, Legal AI Tools and AI Legal Research
Sorted by the task you are actually trying to get done, with what each tool produces, what it costs, and which output a court will accept. Start with the research question on the right.
How courts have ruled
Sample results, illustrative only. Informational research, not legal advice. Verify every citation.
In short
Legal AI is software that applies machine learning and generative models to legal work: case law research, citation checking, document and contract review, drafting, transcript summaries and litigation analytics. Lawyers are permitted to use it. ABA Formal Opinion 512, issued 29 July 2024, confirms that generative AI is compatible with the Model Rules so long as the existing duties of competence, confidentiality, client communication, candor, supervision and reasonable fees are met. The one distinction that decides everything downstream is whether a tool retrieves from an actual body of law and shows you the source, or generates text from a model with nothing underneath it. Research tools of the second kind are what produce fake citations, and US courts sanctioned lawyers for exactly that in at least six federal decisions between February and April 2026 alone. Cases is a research tool of the first kind: ask in plain English, get on-point US precedents with the holding and a citation you can open. Published at 29 dollars per month. This is informational research, not legal advice, and you should always verify the citation.
Last updated August 2026
Buy for a task, not for a category
What AI actually does for lawyers
Almost every roundup of AI tools for lawyers is organized by vendor. That is the wrong axis for a buyer, because the vendors do not compete with each other evenly. A contract review tool and a case law research tool have almost nothing in common except the word AI in the marketing. Sorting by the job gets you to a shortlist much faster.
The column worth reading twice is the third one. It asks whether the output can go into something you file. Most legal AI output cannot, and that is fine, because it is an input to your judgment rather than a substitute for it. The tasks where the answer is yes are the ones that need the most care.
| Task | What the tool produces | Can it go in a filing | Who verifies |
|---|---|---|---|
| Case law research | On-point precedents with a summary, the holding and a citation | Yes, once the citation is checked against the reporter | The lawyer reads the case before citing it |
| Citation checking | Whether a case exists, and whether later courts have treated it badly | Not filed, but it decides whether your brief holds | A citator, not a chatbot. Ask which citator the tool uses |
| Document review and discovery | Ranked or coded documents from a review set | The underlying documents are evidence, the ranking is not | Sampling and a privilege check by a human reviewer |
| Contract review and drafting | Redlines, clause suggestions, deviation from a playbook | Not applicable, this is work product | The lawyer who signs the deal |
| Brief and memo drafting | A first draft with argument structure and cited authority | Only after every citation is independently confirmed | The filing attorney, under Rule 11 |
| Deposition and transcript summaries | Issue summaries with page and line references | The transcript is the record, the summary is a shortcut | Spot check the cited page and line |
| Judge and litigation analytics | How a specific judge or court has ruled on a motion type | No, this is strategy input | Check the coverage window and the sample size |
| Client intake and admin | Qualified intake, scheduling, time capture, billing narratives | No | Ordinary supervision, plus a fee-reasonableness check |
Cases sits in the first row and nowhere else. It answers a research question against US federal and state case law and hands back the cases with the holding and the citation. It does not review your contracts, code your discovery set or draft your brief, and we would rather say that plainly than sell a firm a research tool as an everything tool. If you want the fuller landscape of what each vendor covers, our comparison of legal research platforms goes vendor by vendor.
Most of this market will not tell you a number
What legal AI costs in August 2026
Legal AI pricing is unusually opaque, and the opacity is not accidental. Enterprise vendors price by negotiation because negotiated prices are higher. The table below separates what a vendor publishes on its own site from what it does not, because those are different kinds of fact and most comparisons blur them.
| Tool | What it is | Published rate | Trial |
|---|---|---|---|
| Cases | AI case law search | 29 dollars per user per month, published | Self-serve |
| midpage | AI legal research | 30 dollars per month, 80 dollars on Pro billed annually, published | Two weeks |
| Westlaw Advantage | Research platform with agentic AI | 256.75 single circuit, 399.75 all states plus federal, per user per month, firms up to 10 attorneys | 7 days, no credit card |
| Paxton AI | AI legal assistant | 499 dollars per user per month, or 2,999 per user per year, published | 7 days |
| Lexis+ with Protege | Agentic legal AI | No rate published | 2 days |
| CoCounsel Legal | Agentic legal AI | No rate published. Contains Westlaw Advantage, whose published rate is the floor | On request |
| vLex Vincent AI | AI legal research | No rate published | Offered, length not stated |
| Bloomberg Law | Research platform | No rate published | 14 days, no credit card |
| Harvey | Enterprise legal AI | No rate published, sold with seat minimums | Not published |
Rates and trial terms read from each vendor's own site, August 2026. Where the cell says no rate published, that is the finding: we are not substituting a third-party estimate for a price. Verify before purchase, this market moves.
Two things fall out of that table. The spread between the cheapest and the most expensive published per-seat rate is about seventeen times, and the products at either end do genuinely different jobs, so the comparison is only useful once you have fixed the task. And the trial column is the closest thing this market has to an apples-to-apples number, since most vendors publish a trial length even when they publish no price. The most opaque row is the last one: our Harvey AI guide works through what an enterprise legal AI platform costs when the vendor publishes nothing at all. Our legal research pricing breakdown holds the full set, and the Westlaw free trial guide covers what to test while a trial clock is running.
The 2026 record, in citations you can pull
What happens when nobody opens the case
This is the part of the legal AI conversation with actual evidence behind it, so it deserves numbers rather than warnings. Damien Charlotin maintains a public database of court decisions in which a tribunal found or clearly implied that a party relied on hallucinated material. As of its 8 August 2026 update it lists 1,868 cases. Its inclusion bar is deliberately strict: allegations alone do not qualify, the court has to have made the finding.
Norton Rose Fulbright collected six federal decisions from a single ten week window, February to April 2026, and they are worth reading as a set because the outcomes vary so much. Two produced no monetary sanction at all. One produced a five figure penalty per attorney plus the other side's fees.
| Decision | Court | Date | Citation | Outcome |
|---|---|---|---|---|
| Fletcher v. Experian Information Solutions, Inc. | Fifth Circuit | 18 February 2026 | 168 F.4th 231 | 2,500 dollar monetary sanction |
| In re: Eric Chibueze Nwaubani | Fourth Circuit | 11 March 2026 | 2026 WL 687194 | Public admonishment |
| Whiting v. City of Athens, Tennessee | Sixth Circuit | 13 March 2026 | 170 F.4th 455 | Appellees' reasonable fees on appeal, double costs, and 15,000 dollars each in punitive sanctions |
| United States v. Farris | Sixth Circuit | April 2026 | 171 F.4th 920 | No monetary sanction, but Criminal Justice Act compensation denied |
| Gamez v. County of Fresno | E.D. California | 9 April 2026 | 2026 WL 925944 | No sanctions imposed |
| Fivehouse v. U.S. Department of Defense | E.D. North Carolina | 27 April 2026 | 2026 WL 1278575 | Public reprimand |
Case names, courts, dates, reporter citations and outcomes as collected and cited in Norton Rose Fulbright, "AI in litigation: Update on Gen AI sanctions in 2026". Database count from the AI Hallucination Cases database maintained by Damien Charlotin, last updated 8 August 2026. Pull the opinions before relying on any of this.
Whiting v. City of Athens is the one to read if you only read one. The Sixth Circuit found that counsel had filed briefs containing more than two dozen fake citations, along with quotations that did not appear in the cases cited and citations that did not support the propositions they were attached to. The panel ordered the attorneys to pay the appellees' reasonable fees on appeal, pay double appellate costs, and each pay 15,000 dollars in punitive sanctions, treating the fake citations as misconduct in arguing the appeal.
Notice what is not in that list: nobody was sanctioned for using AI. They were sanctioned for filing authority they had not read. That distinction is the whole practical lesson, and it points at a workflow rather than a policy. We go through the verification steps in detail in can lawyers use AI for legal research, and the mechanics of confirming a citation exists in our case citation checker guide.
ABA Formal Opinion 512, issued 29 July 2024
What the ethics rules require of a firm using AI
Formal Opinion 512 is the first ABA ethics guidance written specifically about generative AI, and its structure is reassuringly boring: it does not create new duties, it maps the existing Model Rules onto the technology. That is good news for a firm building a policy, because it means you already know the standard.
| Duty | What it means in practice |
|---|---|
| Competence, Model Rule 1.1 | Understand what the tool can and cannot do, and keep that understanding current. The opinion is explicit that a generative AI tool cannot by itself substitute for a lawyer's own competent work. |
| Confidentiality, Model Rule 1.6 | Know how the tool handles your data and put safeguards in place. The opinion says lawyers should generally obtain informed client consent before putting client confidences into a generative AI tool, and that boilerplate consent buried in an engagement letter is not enough. |
| Communication, Model Rule 1.4 | Tell the client when the use of the tool is material to the representation, for example when client information goes into it or when it affects how the matter is handled. |
| Candor to the tribunal, Model Rule 3.3 | What you file is your representation to the court. Output you did not read is still your signature. |
| Supervision, Model Rules 5.1 and 5.3 | Partners and managers have to put policies and training in place, and the duty extends to non-lawyer assistance, which is how the opinion treats the tools themselves. |
| Fees, Model Rule 1.5 | Fees must stay reasonable. You cannot bill an hour that the tool did in a minute, and you cannot pass through the cost of learning the software as client work. |
Summarized from ABA Formal Opinion 512, "Generative Artificial Intelligence Tools", 29 July 2024. State bars have issued their own guidance and it is not uniform, so read your jurisdiction's opinion alongside this one. Informational research, not legal advice.
The confidentiality point is the one firms underestimate. The opinion puts the burden on the lawyer to know how a tool handles input, which means self-serve consumer AI is a harder sell inside a firm than a vendor with a written retention policy and a security posture you can inspect. Two practical questions cover most of it: does the vendor train on your input, and how long is it retained. Ask both in writing before a matter goes anywhere near the tool.
The fee point is the sleeper. If a research task that used to take three billable hours now takes twenty minutes, the reasonable fee is twenty minutes. Firms that treat AI as a margin lever rather than a capacity lever are the ones most likely to end up in a fee dispute, and Formal Opinion 512 says so more or less directly.
A shortlist in four questions
How to choose legal AI for a law firm
1. What corpus does it search
This is the question that separates the products. A tool that retrieves from a defined body of case law can be checked. A tool answering from a general model cannot, because there is no source to open. Ask for the coverage: which courts, which years, and how current. A vendor that cannot answer precisely is telling you something.
2. Does it show its sources
Every claim should carry a citation you can open in one click, and the case you land on should say what the summary said it says. Test this deliberately during a trial with a question you already know the answer to. It is the fastest way to tell a retrieval tool from a text generator.
3. Is there a citator, and whose
Finding a case and knowing it is still good law are different capabilities. KeyCite and Shepard's are the two established citators, and neither is perfect: Paul Hellyer's study in Law Library Journal volume 110 reviewed 357 citing relationships and found each missed or mislabeled roughly a third of negative treatments. We do not have a citator and say so on every page. Know which side of that line your tool sits on.
4. What does it cost when the trial ends
Ask for the renewal rate and the term length before you sign, not the first year discount. Legal research contracts commonly run two to three years, and the published multi-year discounts in this market are the vendor pricing the lock-in. A published, cancellable monthly rate is worth real money against a cheaper three year term.
For solo attorneys and small firms the practical order is usually research first, then drafting, then everything else, because research is where the hours go and where the cheapest published options exist. Larger litigation teams tend to have the reverse problem: discovery volume dominates and the research spend is already committed under a multi-year contract. If you are working out where a switch would land you, our Westlaw alternatives and LexisNexis alternatives pages lay out what you keep and what you give up.
Questions lawyers ask before they buy
Legal AI, answered
What is legal AI?
Legal AI is software that applies machine learning, and now generative models, to legal work: finding case law, checking citations, reviewing documents and contracts, drafting, summarizing transcripts and analyzing how courts rule. The useful distinction is not AI versus not AI. It is whether the tool answers from a controlled body of law with citations you can open and read, or from a general model that produces fluent text with no source underneath it. The first can be verified in minutes. The second cannot be verified at all.
Can lawyers use AI?
Yes. The American Bar Association addressed this directly in Formal Opinion 512, issued 29 July 2024, and the answer is that lawyers may use generative AI provided they satisfy their existing duties: competence, confidentiality, communication with the client, candor to the tribunal, supervision, and reasonable fees. No rule prohibits the technology. What gets lawyers sanctioned is filing output nobody read, which is a Rule 11 and Rule 3.3 problem rather than an AI problem.
What is the best AI for lawyers?
There is no single answer, because the tools split by task. For finding and citing authority you want a tool built on a controlled body of case law that returns real citations. For contract review you want a drafting tool that works inside your document. For discovery you want a review platform. Buying one general assistant for all of it is the most common expensive mistake. Pick the task that is costing your firm the most hours and buy for that first.
Will AI replace lawyers?
No, and the way it is failing tells you why. Generative models are good at producing text that reads like legal writing and bad at knowing whether the authority in it exists. Every 2026 sanctions decision turns on the same gap: a filing that looked professional and cited cases that were not real. Judgment, accountability and the signature on the filing stay with the lawyer. What is genuinely changing is how long the first draft and the first research pass take.
Is AI legal research reliable?
It depends entirely on what the tool searches. A model answering from its own training data will invent citations that look correct, complete with plausible reporter numbers. A tool that retrieves from an actual body of case law and shows you the case can be checked in the time it takes to open it. Treat any citation you have not opened as unverified regardless of how confident the tool sounds. That single habit prevents almost every reported failure.
How much does legal AI cost?
The published range in August 2026 runs from about 29 to 30 dollars per user per month at the self-serve end, through 256.75 to 399.75 for Westlaw Advantage for firms up to ten attorneys, to 499 for Paxton AI. LexisNexis, CoCounsel, Bloomberg Law, vLex and Harvey publish no rate at all, so the figures circulating for those are third-party estimates rather than prices. Most of this market is quote-only, which is why published pricing is worth weighting when you compare.
Do lawyers have to disclose AI use in court filings?
It depends on the judge, and this is genuinely fragmented. There is no blanket federal rule. A number of individual judges and some jurisdictions have adopted standing orders or local rules on generative AI in filings, ranging from a certification that citations were verified to full disclosure of AI use. Check the standing order of your specific judge before you file, every time, because the requirement lives at the chambers level rather than the circuit level.
Is ChatGPT safe for legal work?
For legal research specifically, a general chatbot is the wrong instrument. It does not search a case law database, so it produces citations by pattern rather than by retrieval. On confidentiality, ABA Formal Opinion 512 puts the burden on you to know how the tool handles your input and to get informed client consent before putting client confidences into it. General assistants are useful for non-confidential drafting and explanation. They are not a research tool and should not be treated as one.
What is an AI legal assistant?
It is the marketing term for a chat interface over one or more legal workflows: ask a question, get an answer, ask it to draft something. The label tells you nothing about quality, so ignore it and ask two questions instead. What corpus does it search, and does it show its sources. If the answer to the first is "the model" and to the second is no, you have a text generator rather than a legal assistant.
What tasks can AI do for a law firm?
In practice: first-pass case law research, citation and good-law checking, document review and discovery ranking, contract review against a playbook, first-draft briefs and memos, deposition and transcript summaries, litigation and judge analytics, and intake and billing administration. The pattern worth noticing is that AI is strongest where the output is an input to your judgment and weakest where the output goes straight to a court without a human in between.
Keep researching
Legal AI you can check in one click
Ask your research question in plain English and get on-point US federal and state precedents with the holding and a real citation. Published at 29 dollars per month, no quote required. Informational research, not legal advice, always verify the citation.