What is the best AI for lawyers?

The useful comparison starts with the legal task. A solo lawyer running intake, billing and files has different needs from a litigation team reviewing discovery or an in house team handling contracts. Clio combines practice operations with its Work and vLex offerings. Everlaw organizes discovery review, while Ironclad manages contract processes. PatentPal produces patent specifications from claims. According to the AI Legal Index, 15 vendors inherit document system permissions and 156 provide their own matter separation model. That difference matters when an assistant searches across client files. Source coverage, citation checks, data terms and the location of the work all affect fit. A tool that drafts a contract is not automatically a primary law research service or a discovery production system.

What AI tools do law firms use?

Law firms use legal AI across several distinct workflows. Research assistants retrieve law and draft answers, discovery platforms sort and analyze evidence, and drafting tools work inside documents. Paxton covers federal and state law, Clearbrief checks citations in Word, and Briefpoint prepares discovery requests and responses. DeepJudge searches a firm's internal knowledge while retaining source permissions. Draftwise retrieves contract language from prior deals, and Gavel separates document automation from its Exec contract product. The AI Legal Index describes these different functions across products serving firms of different sizes. Products can overlap within one firm because a matter system, research service and citation checker perform different jobs. Their integration points determine how much work stays in an existing document or case file.

Can lawyers use ChatGPT for legal work?

ChatGPT can help lawyers summarize material, outline an argument or prepare draft language. Its consumer and business arrangements differ. OpenAI allows consumer content to improve models unless the user opts out, while ChatGPT Business and Enterprise exclude business inputs and outputs from training by default. Neither account type guarantees that an answer states the law correctly. ABA Formal Opinion 512, issued July 29, 2024, addresses competence, confidentiality, review and supervision when lawyers use generative AI. In Mata v. Avianca, the June 22, 2023 sanctions order concerned fabricated authorities and counsel's conduct after those authorities were challenged. The order did not prohibit reliable AI assistance. A general ChatGPT conversation also lacks the built in combination of a legal database, citator and matter permissions found in some legal products.

What is the best AI for legal research?

Legal research products differ in the law they retrieve, the treatment information they provide and how they support an answer. Paxton researches federal and state primary law and supplies citation treatment badges. Harvey's separately licensed Ask LexisNexis adds LexisNexis research with Shepard's information. Descrybe combines opinion search, verification and its Cytator citator in a research platform. Clearbrief serves another role by checking citations and supporting evidence within Word documents. The AI Legal Index identifies 16 vendors with measured citation accuracy evidence and 245 with grounding or citation checking functions. Those features and measurements vary across products, tasks and jurisdictions. A linked opinion supports source inspection, while treatment checking addresses whether later authority changed its standing. Neither alone establishes that a cited passage supports a particular argument.

Have lawyers been sanctioned over AI filings?

Yes. Courts have imposed monetary and professional sanctions for filings with fabricated authorities, often alongside failures to verify them or correct the record candidly. In Mata v. Avianca on June 22, 2023, two lawyers and their firm received a joint $5,000 penalty. In Wadsworth on February 24, 2025, the court imposed three fines totaling $5,000 and revoked one lawyer's admission for that case. Noland v. Land of the Free brought a $10,000 sanction on September 12, 2025. On June 3, 2026, the Ninth Circuit fined two lawyers $2,500 each in Lnu v. Blanche and suspended them for six months. These were sanctions for particular conduct involving false filings and professional duties. They were not findings that every use of a legal AI product is prohibited.

Do lawyers have to tell a court when they use AI?

Disclosure depends on the court, the judge and the applicable order. Northern District of Texas Local Civil Rule 7.2(f) requires a first page disclosure when a brief was prepared using generative AI. Judge Baylson's June 6, 2023 order in the Eastern District of Pennsylvania covers AI used in preparing filings assigned to him and requires certification that citations were verified. By contrast, the Western District of North Carolina's July 29, 2026 order removed its separate AI certificate while preserving human verification requirements. The AI Legal Index records four vendors offering exportable disclosure records and 275 providing partial support. An activity history, source link or citation report may supply information for a filing, but those features do not determine what a particular court requires.

Do legal AI tools train on client data?

Some prohibit it, some permit it and some condition it on an account setting or agreement. The AI Legal Index records 102 contractual prohibitions on training with client data, 123 prohibitions stated in policy and 19 opt in arrangements. These are different commitments. DeepJudge's terms bar model training with customer data unless the customer gives written consent. Ironclad's position depends on an opt in, while Filevine describes restrictions on its model providers' training and retention. A training prohibition addresses one use of information. It does not by itself say how long files remain stored, which subprocessors receive them or which matter participants can retrieve them. Consumer assistants also differ from business workspaces, even when both carry the same product name and a paid subscription.

How much does legal AI cost?

Published entry subscriptions and full legal assistants occupy different price levels. The AI Legal Index records Irys Lite at $19.99 a month with 12 AI messages a day, Descrybe Legal Engine at $25 per user a month and the full Descrybe Platform at $50. Gavel Exec costs $160 per user a month. Paxton offers $499 per user monthly or $2,999 per user annually paid up front, while GC AI Individual is $500 per user a month. Solve Intelligence Starter is $599 per user a month on an annual commitment with 2,662 monthly credits. Harvey, Legora and Filevine require custom quotes. Seat counts, credit allowances, separately licensed research and implementation can change what the final contract covers. These figures describe different products and billing terms, rather than interchangeable bundles.

Will AI replace lawyers?

The products in the index automate parts of legal work rather than assume a lawyer's professional responsibilities. EvenUp prepares demand materials from case records, Briefpoint drafts discovery documents, and Brightflag analyzes legal invoices. Each changes the work involved in producing or reviewing a particular output. Their effects on employment and client demand are still uncertain. California's practical guidance, approved May 14, 2026, expressly addresses agentic systems that can initiate tasks and act with less prompting. It retains attorney responsibility for professional judgment and meaningful supervision. The NYSBA task force report approved April 6, 2024, likewise connects AI use with independent judgment, confidentiality and oversight. Current products therefore support specific legal processes, while the professional duties described in those sources remain with the lawyer.

Index · AI for lawyers

AI for lawyers: which legal AI tools fit which practice

Last reviewed November 11, 2026

There is no single best AI for lawyers, because the tools are built for different jobs. Clio and Smokeball put AI inside a small firm's daily operations. Paxton does primary law research, Everlaw handles discovery, Ironclad manages contracts, and Solve Intelligence drafts patents. The right one depends on the work a practice does most.

What separates them is less the drafting than what sits around it. The tools differ in which law they search, how they check a citation, which client files they can open and whether the contract bars training on client data. According to the AI Legal Index, 15 vendors follow the firm's document permissions and 156 keep a separate permission system. Prices run from $19.99 a month to custom enterprise contracts.

The AI Legal Index holds 466 AI vendors in legal across 9 categories. Each one is researched from public sources, and no vendor pays to be listed.

By kind of practice

Solo and small firms

Small firms use AI for work that repeats across matters, from turning intake answers into documents to finding facts for a motion. A practice management system puts that work beside the calendar, billing and client file. Clio Manage handles those daily operations, while Clio Work adds analysis, research and drafting using matter information and the vLex legal library. Smokeball's Archie works inside the current matter and in Word and Outlook. Its AutoTime feature records time spent working, and its document automation draws on more than 20,000 forms.

Separate tools cover narrower needs without replacing the firm's matter system. Gavel Workflows turns questionnaires and Word or PDF templates into finished documents for work such as estate planning, family law and probate. Gavel Exec reviews contracts against precedent and playbooks. Paxton brings case files, primary law research and litigation drafting into one workspace, with source links and citator treatment badges. Which one fits depends on whether the firm's work usually starts with client intake, a contract or an active dispute.

Midsize and large firms

Larger firms need AI to work across teams, document systems and existing client restrictions. According to the AI Legal Index, DeepJudge searches a firm's documents across its document management system, email and other repositories while carrying over source permissions. That makes its starting point the firm's accumulated work. Draftwise applies deal history to new contracts in Word. Its Playbook Studio builds review rules from example agreements, and its Deal Table compares terms across repositories such as iManage and NetDocuments.

Harvey and Legora cover a broader mix of research, review and drafting. Harvey's Vault supports tabular review of up to 100,000 documents, while Ask LexisNexis adds case law and Shepard's Citations to the workspace. Legora's Tabular Review attaches a source to each cell and its Portal delivers work to clients under the firm's brand. Legora also connects to Intapp Walls and iManage Security Policy Manager. Many large firms run more than one, since precedent search, deal review and client delivery are separate steps in one matter.

In house legal teams

Corporate legal departments use AI to answer business requests, negotiate agreements and control outside counsel spending. Those jobs sit in different systems. Checkbox captures requests from email, Slack, Teams and web forms, then routes them by matter type, expertise and capacity. Its dashboards measure demand and cycle times. Ironclad manages the contract itself, from intake and approvals to signature and the repository. Contract Review Agent redlines against company playbooks and precedent, while other agents handle search, renewals and portfolio questions.

GC AI gives the individual lawyer a research and drafting workspace with shared projects and a Word add in. Its Exact Quote feature links answers to highlighted source text, and Contract Intelligence extracts terms across a connected portfolio. Brightflag starts with the legal bill. Its model classifies invoice lines by activity, and billing rules flag entries before approval and payment. According to the AI Legal Index, Brightflag's subscription is based on outside counsel spend. That purchasing model differs from a research seat or a contract workflow subscription.

Litigation and e-discovery

Litigation AI handles several distinct bodies of material. Discovery platforms process collections, help reviewers classify documents and prepare productions. Everlaw combines review and production with Storybuilder, where its Writing Assistant drafts from the evidence. Coding Suggestions supports first pass review. Relativity covers legal hold, collection, processing and production, with separate aiR products for review, privilege and case strategy. These products organize large evidence collections and the decisions made about them.

Drafting and checking tools operate closer to the document that leaves the firm. Briefpoint generates written discovery from a complaint or answer, drafts responses to opposing requests and brings client answers into those responses through a secure link. Its production responses can cite Bates or exhibit numbers. Clearbrief works in Word, comparing factual and legal citations with their sources and producing a downloadable Cite Check Report. Sorting which documents are responsive and checking whether a source supports a sentence are separate jobs, and a team in active litigation usually needs both. Neither one writes the argument.

Plaintiff and personal injury firms

Personal injury tools turn medical records, bills, police reports and intake notes into chronologies, demands and case analysis. EvenUp's MedChrons extracts treatment timelines and diagnosis codes, while its demand service ranges from immediate AI generation to an expert reviewed tier. Its Settlement Repository supplies past settlements for valuation work. Supio also builds medical chronologies and demands, with page citations for claims. For high value cases, human experts verify bill totals and damage calculations before delivery.

Eve covers a wider sequence from intake and evaluation through complaints, discovery and resolution. Its drafts link back to the firm's documents and adapt to the firm's style. Filevine places AI inside case management, with MedChron, DemandsAI and Depo CoPilot beside the matter's operational record. The practical difference is how much of the case the tool takes on. Some produce one demand package, some analyze the medical record, and some run the whole matter. According to the AI Legal Index, EvenUp prices by case, while Filevine combines case management with separate AI capabilities.

EvenUp

Supio

Eve

Filevine

IP and patent practices

Patent AI spans drafting, prosecution, prior art research and disputes over existing rights. PatentPal starts with claims that the attorney supplies. It generates supporting specification text, flowcharts, block diagrams and figure descriptions, then exports the result to Word and drawing formats. DeepIP works inside Word and starts earlier, drafting applications from invention disclosures. It also reviews claims for antecedent basis and section 101 risk, drafts office action responses and searches patent and nonpatent literature.

Solve Intelligence uses a browser editor for applications, figures, chemical structures and sequence listings. It also builds infringement, invalidity and freedom to operate charts. Patlytics connects those analysis tasks with portfolio work, claim construction and prosecution, with a source behind each claim chart cell. These are different jobs. Turning finished claims into a specification is not the same work as finding prior art or mapping a claim onto a product. The products overlap in drafting, and where each one fits depends on what it starts from and what it produces.

Where the tools differ

What happens to client data

Client data can stay out of shared model training under very different arrangements. According to the AI Legal Index, 102 vendors prohibit training in their agreements, 123 state a prohibition in a policy, and 19 use an opt in arrangement. A contractual restriction, a policy statement and a customer permission switch are different commitments. Training also differs from retaining a prompt, letting support staff inspect an error or sending text to a model provider to produce an answer.

DeepJudge's terms prohibit training on customer data without prior written consent, while its deployment options include the firm's own cloud or premises. Ironclad trains its own models only when customers opt in. Filevine bars third party model providers from training on or keeping customer data. Those examples address different parts of the processing chain. The application can retain a matter even when its model provider keeps no prompt history. So a promise about the model says nothing by itself about how the application stores a matter, deletes it or controls who can open it.

DeepJudge

Ironclad

Filevine

How legal AI tools handle client data in training →

Whether the citations can be trusted

A citation can point to a real opinion and still misstate what it holds. Source links, checks for later treatment and measured task accuracy answer separate questions. The AI Legal Index identifies 16 vendors with checkable measured accuracy evidence and 245 with grounding and citation checking but without equivalent measured evidence. Those figures cover different kinds of legal software. A result on document classification cannot establish that a research assistant cites the right authority.

Harvey's task accuracy benchmarks measure performance on specific legal work, rather than whether a citation exists. Ask LexisNexis brings Shepard's Citations into its workspace. Paxton links citations to the cited page and adds treatment badges from its own citator. Paxton's 2024 benchmark results do not measure the accuracy of the current product. Clearbrief checks citations and supporting passages inside Word and produces a Cite Check Report. That report makes the review traceable. Cite Check has no stated false positive or false negative rate. These features show the source and the checking. They do not make the generated analysis right.

Paxton AI

Harvey

Clearbrief

How legal AI tools support and check citations →

Privilege and ethical walls

Confidentiality controls govern who can reach a client's material and what happens when someone requests it. Keeping one firm's files separate from another firm's files is one boundary. Keeping opposing matter teams apart inside the same firm is another. The AI Legal Index identifies 15 vendors that enforce source document permissions during retrieval and 156 with their own permission systems. The difference is whether a change in the document system carries into the AI's answers or has to be reflected in a separate access model.

DeepJudge continuously syncs source permissions across the repositories it searches. Draftwise retrieves under the individual user's identity and follows live document system permissions. Legora connects those controls to Intapp Walls and iManage Security Policy Manager. These mechanisms affect which passages can enter an answer, even when a search covers a large collection. Access restrictions also differ from promises about compelled disclosure. The index's separate notice signal identifies 204 vendors with commitments to notify customers when legally permitted. Such a commitment addresses the opportunity to respond to a request, rather than deciding whether particular material is privileged.

DeepJudge

Draftwise

Legora

How legal AI handles ethical walls and confidential matters →

Security certifications

Security assurance concerns the controls around a service. It does not measure whether a generated citation is correct. According to the AI Legal Index, 32 vendors have current independent attestation with a named scope and an accessible route to the report. Another 267 state a certification or attestation without the same visibility into its scope, date or report. A SOC 2 Type II report concerns controls over a period. An ISO 27001 certificate concerns an information security management system within a defined scope.

DeepIP's trust center carries SOC 2 Type 2, ISO 27001 and ISO 42001 material, while its product offers cloud and on premises deployment. Solve Intelligence holds SOC 2 Type II and ISO 27001 and offers separate US and EU deployments. Everlaw holds FedRAMP Moderate and GovRAMP authorizations and hosts in several regions. These are distinct forms of assurance and deployment choice. None automatically answers which model provider processes a prompt, how long an application retains it or whether a source permission follows it into a generated response.

Everlaw

DeepIP

Solve Intelligence

Which legal AI tools have checkable security assurance →

Disclosure to a court

A record of AI use and a check of legal authority serve different purposes. The first explains how a document was produced. The second tests the sources and propositions in it. The AI Legal Index identifies 4 products with exportable document records and 275 with parts of a record. Another 4 provide guidance or templates without corresponding product support. Those counts describe available features, rather than a finding that a particular export satisfies a court's order.

Clearbrief's downloadable Cite Check Report lists factual and legal citations, flags missing sources and links each item to its context. Its checking process uses classic machine learning, so the report does not identify a generative model for that step. Everlaw's AI results cite source documents, and its platform preserves review and production histories. Those features do not combine model, retrieved sources and human verification in one document export. So a history of review decisions, a citation report and a declaration of AI use can each say something different, even about the same filing.

Everlaw

Clearbrief

Which legal AI tools keep records for court disclosure →

General AI or legal AI

ChatGPT, Claude, Gemini and Microsoft Copilot can summarize supplied material, reorganize notes and draft explanations in plain language. Copilot also works with permitted emails and documents in Microsoft 365. These functions are useful for a first draft or an outline, but a general assistant does not automatically provide a legal database, a citator and matter restrictions together. The account and the connected tools determine much of that difference.

OpenAI's consumer terms permit model improvement, with an opt out for new conversations. ChatGPT Business and Enterprise exclude inputs and outputs from training by default. Claude Free, Pro and Max use conversations for improvement when users allow it, with separate feedback and safety review provisions. Claude's commercial services exclude inputs and outputs from training by default, although voluntarily submitted feedback can be used. A paid individual subscription is therefore not the same arrangement as a business workspace.

Consumer Gemini can use activity to train models when Keep Activity is on. With it off, future chats are excluded unless feedback is submitted. Qualifying Workspace services require customer permission or instructions before training on customer data. Consumer Copilot permits training for eligible signed in users who have not opted out. Organizational accounts are excluded, as are Copilot conversations inside Microsoft 365 Personal and Family apps. Microsoft's commercial Copilot also excludes prompts, responses and Microsoft Graph data from foundation model training.

Tools built for law add specific workflows around those models. Paxton supplies primary law research and treatment badges, Clearbrief checks citations in Word, and Draftwise follows document system permissions during retrieval. Those additions connect drafting to authority and matter controls. Each product combines a different set of sources, checks and permissions around the model.

Paxton AI

Draftwise

Clearbrief

What it costs

Legal AI prices describe different units of work. A seat subscription buys access for a person, while credits, document allowances and case charges limit or price the work processed. Per seat legal assistants also differ from practice management subscriptions, such as Clio's starting plan. A platform subscription does not necessarily include its AI products.

Billing periods and usage limits vary across the plans below. Irys Lite has a message allowance, Solve Intelligence uses credits, and enterprise contracts can add separate research licenses or implementation charges. A monthly subscription price and a custom quote can cover different combinations of users, functions and usage.

Kind of plan	Typical price	Examples
Entry subscriptions Irys Lite includes 12 AI messages a day. Descrybe separates its research connector from the full research platform.	Irys Lite is $19.99 a month. Descrybe Legal Engine is $25 per user a month and the full Descrybe Platform is $50.	Irys Descrybe
Practice and drafting software Clio's figure is the practice platform starting price. Gavel Exec is the contract drafting product and includes 1,000 monthly completions.	Clio starts at $49 per user a month. Gavel Exec is $160 per user a month or $1,740 per user billed annually.	Clio Gavel
Individual legal assistants Paxton includes primary law research and case work. GC AI Individual includes unlimited chats and skills. Their team or enterprise offerings use separate quotes.	Paxton is $499 per user a month or $2,999 per user a year paid up front. GC AI Individual is $500 per user a month.	Paxton AI GC AI
Patent drafting subscriptions Solve Starter includes 2,662 monthly credits and stops AI work at the limit. Its annual commitment is invoiced monthly. DeepIP sells monthly or annual seat subscriptions.	Solve Intelligence Starter is $599 per user a month on an annual commitment or $799 with a monthly commitment. DeepIP requires a quote.	Solve Intelligence DeepIP
Enterprise contracts Harvey licenses seats and charges separately for Ask LexisNexis. Legora combines a platform fee with credits. Brightflag prices by outside counsel spend and charges implementation. Filevine packages its platform and LOIS products by sales order.	Custom quotes without a published rate card	Harvey Legora Brightflag Filevine

What courts and bars have said

Courts and bar associations address different parts of legal AI use. Ethics opinions explain how professional duties apply to confidentiality, competence, fees and supervision. Sanctions orders decide what happened in a particular case. Local rules and standing orders govern filings in a particular court or before a particular judge.

The record below separates those actions by date and issuing body. A disclosure requirement in one court is not a nationwide rule. Nor is a citation checker a substitute for the verification described in these orders. The July 2026 North Carolina order, for example, removed a separate certificate while retaining human verification.

July 29, 2026 · United States District Court for the Western District of North Carolina

In Re: Use of Artificial Intelligence

The court required human verification of factual contentions, legal authorities and quotations in filings. It superseded its June 18, 2024 standing order and ended the separate AI certificate requirement. Filing a document itself carries the stated certifications.

Western District of North Carolina AI order
June 3, 2026 · United States Court of Appeals for the Ninth Circuit

Lnu v. Blanche

The court fined Mike Singh Sethi and William Rounds $2,500 each and suspended both from practice before it for six months. The order addressed fabricated authorities and repeated failures of candor. It also imposed AI disclosure and personal verification statements on the firm's filings for two years.

Ninth Circuit order in Lnu v. Blanche
May 14, 2026 · The State Bar of California

Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law

The Board of Trustees approved guidance replacing the November 2023 version. It expressly addresses agentic AI, retaining attorney responsibility for professional judgment and requiring meaningful supervision and review when systems perform legal tasks, communicate advice or prepare filings.

California's revised AI guidance
September 12, 2025 · Court of Appeal of the State of California, Second Appellate District, Division Three

Noland v. Land of the Free, L.P.

The court imposed $10,000 in sanctions on attorney Amir Mostafavi after appellate briefs included fabricated quotations, nonexistent cases and misrepresented authorities generated with AI. It directed service of the opinion on his client and transmission to the State Bar.

Noland sanctions opinion
February 24, 2025 · United States District Court for the District of Wyoming

Order on Sanctions and Other Disciplinary Action

In Wadsworth v. Walmart, eight of nine cases cited in motions did not exist. The court fined Rudwin Ayala $3,000 and revoked his admission for that case. It also fined T. Michael Morgan and Taly Goody $1,000 each for their failures to verify the filing.

Wadsworth sanctions order
July 29, 2024 · American Bar Association Standing Committee on Ethics and Professional Responsibility

Formal Opinion 512

The opinion applies existing professional duties to generative AI, including competence, confidentiality, client communication, candor, supervision and reasonable fees. It addresses reviewing outputs, protecting client information and billing hourly clients for actual time rather than hours that automation saved.

ABA Formal Opinion 512 on generative AI
June 10, 2024 · United States District Court for the Northern District of Texas

Special Order adopting Local Civil Rule 7.2(f) and Local Criminal Rule 47.2(e)

After a public comment period, the court adopted rules that took effect September 3, 2024. A brief prepared using generative AI discloses that on its first page under the heading Use of Generative Artificial Intelligence. A brief without the disclosure certifies that no part of it was prepared that way.

Northern District of Texas order adopting AI disclosure rules
April 6, 2024 · New York State Bar Association

Report and Recommendations of the New York State Bar Association Task Force on Artificial Intelligence

The House of Delegates approved the task force report and its recommended AI guidelines. The report covers client confidentiality, supervision, independent judgment and verification. It also recommends education for lawyers and judges and continuing review of the guidelines.

NYSBA AI report
January 30, 2024 · United States Court of Appeals for the Second Circuit

Park v. Kim

After counsel cited a nonexistent decision supplied by ChatGPT without reading or confirming it, the court referred attorney Jae S. Lee to its Grievance Panel. It also ordered her to provide the decision to her client.

Second Circuit opinion in Park v. Kim
January 19, 2024 · The Florida Bar

Opinion 24-1

This advisory opinion permits generative AI use subject to confidentiality, accurate and competent service, proper billing and advertising rules. It addresses retention, sharing and training policies, and requires client facing chatbots to identify themselves as AI rather than a lawyer or firm employee.

Florida Bar advisory ethics opinion on AI
June 22, 2023 · United States District Court for the Southern District of New York

Mata v. Avianca, Inc.

The court imposed a joint $5,000 penalty on two lawyers and their firm after submissions cited invented cases and counsel continued defending them. It also required notice to the client and the judges falsely identified as authors of the fabricated opinions.

Mata sanctions opinion and order
June 6, 2023 · United States District Court for the Eastern District of Pennsylvania

Standing Order Re: Artificial Intelligence (“AI”) in Cases Assigned to Judge Baylson

For filings assigned to Judge Michael M. Baylson, attorneys and parties representing themselves must disclose any AI use in preparing the filing. They must also certify that every citation to law or the record has been verified as accurate.

Judge Baylson's AI standing order

Browse every vendor →

Common questions

What is the best AI for lawyers?: The useful comparison starts with the legal task. A solo lawyer running intake, billing and files has different needs from a litigation team reviewing discovery or an in house team handling contracts. Clio combines practice operations with its Work and vLex offerings. Everlaw organizes discovery review, while Ironclad manages contract processes. PatentPal produces patent specifications from claims. According to the AI Legal Index, 15 vendors inherit document system permissions and 156 provide their own matter separation model. That difference matters when an assistant searches across client files. Source coverage, citation checks, data terms and the location of the work all affect fit. A tool that drafts a contract is not automatically a primary law research service or a discovery production system.
Clio Ironclad Everlaw PatentPal
What AI tools do law firms use?: Law firms use legal AI across several distinct workflows. Research assistants retrieve law and draft answers, discovery platforms sort and analyze evidence, and drafting tools work inside documents. Paxton covers federal and state law, Clearbrief checks citations in Word, and Briefpoint prepares discovery requests and responses. DeepJudge searches a firm's internal knowledge while retaining source permissions. Draftwise retrieves contract language from prior deals, and Gavel separates document automation from its Exec contract product. The AI Legal Index describes these different functions across products serving firms of different sizes. Products can overlap within one firm because a matter system, research service and citation checker perform different jobs. Their integration points determine how much work stays in an existing document or case file.
Gavel Paxton AI DeepJudge Draftwise Briefpoint Clearbrief
Can lawyers use ChatGPT for legal work?: ChatGPT can help lawyers summarize material, outline an argument or prepare draft language. Its consumer and business arrangements differ. OpenAI allows consumer content to improve models unless the user opts out, while ChatGPT Business and Enterprise exclude business inputs and outputs from training by default. Neither account type guarantees that an answer states the law correctly. ABA Formal Opinion 512, issued July 29, 2024, addresses competence, confidentiality, review and supervision when lawyers use generative AI. In Mata v. Avianca, the June 22, 2023 sanctions order concerned fabricated authorities and counsel's conduct after those authorities were challenged. The order did not prohibit reliable AI assistance. A general ChatGPT conversation also lacks the built in combination of a legal database, citator and matter permissions found in some legal products.
What is the best AI for legal research?: Legal research products differ in the law they retrieve, the treatment information they provide and how they support an answer. Paxton researches federal and state primary law and supplies citation treatment badges. Harvey's separately licensed Ask LexisNexis adds LexisNexis research with Shepard's information. Descrybe combines opinion search, verification and its Cytator citator in a research platform. Clearbrief serves another role by checking citations and supporting evidence within Word documents. The AI Legal Index identifies 16 vendors with measured citation accuracy evidence and 245 with grounding or citation checking functions. Those features and measurements vary across products, tasks and jurisdictions. A linked opinion supports source inspection, while treatment checking addresses whether later authority changed its standing. Neither alone establishes that a cited passage supports a particular argument.
Paxton AI Harvey Clearbrief Descrybe
Have lawyers been sanctioned over AI filings?: Yes. Courts have imposed monetary and professional sanctions for filings with fabricated authorities, often alongside failures to verify them or correct the record candidly. In Mata v. Avianca on June 22, 2023, two lawyers and their firm received a joint $5,000 penalty. In Wadsworth on February 24, 2025, the court imposed three fines totaling $5,000 and revoked one lawyer's admission for that case. Noland v. Land of the Free brought a $10,000 sanction on September 12, 2025. On June 3, 2026, the Ninth Circuit fined two lawyers $2,500 each in Lnu v. Blanche and suspended them for six months. These were sanctions for particular conduct involving false filings and professional duties. They were not findings that every use of a legal AI product is prohibited.
Do lawyers have to tell a court when they use AI?: Disclosure depends on the court, the judge and the applicable order. Northern District of Texas Local Civil Rule 7.2(f) requires a first page disclosure when a brief was prepared using generative AI. Judge Baylson's June 6, 2023 order in the Eastern District of Pennsylvania covers AI used in preparing filings assigned to him and requires certification that citations were verified. By contrast, the Western District of North Carolina's July 29, 2026 order removed its separate AI certificate while preserving human verification requirements. The AI Legal Index records four vendors offering exportable disclosure records and 275 providing partial support. An activity history, source link or citation report may supply information for a filing, but those features do not determine what a particular court requires.
Do legal AI tools train on client data?: Some prohibit it, some permit it and some condition it on an account setting or agreement. The AI Legal Index records 102 contractual prohibitions on training with client data, 123 prohibitions stated in policy and 19 opt in arrangements. These are different commitments. DeepJudge's terms bar model training with customer data unless the customer gives written consent. Ironclad's position depends on an opt in, while Filevine describes restrictions on its model providers' training and retention. A training prohibition addresses one use of information. It does not by itself say how long files remain stored, which subprocessors receive them or which matter participants can retrieve them. Consumer assistants also differ from business workspaces, even when both carry the same product name and a paid subscription.
DeepJudge Ironclad Filevine
How much does legal AI cost?: Published entry subscriptions and full legal assistants occupy different price levels. The AI Legal Index records Irys Lite at $19.99 a month with 12 AI messages a day, Descrybe Legal Engine at $25 per user a month and the full Descrybe Platform at $50. Gavel Exec costs $160 per user a month. Paxton offers $499 per user monthly or $2,999 per user annually paid up front, while GC AI Individual is $500 per user a month. Solve Intelligence Starter is $599 per user a month on an annual commitment with 2,662 monthly credits. Harvey, Legora and Filevine require custom quotes. Seat counts, credit allowances, separately licensed research and implementation can change what the final contract covers. These figures describe different products and billing terms, rather than interchangeable bundles.
Gavel Paxton AI Harvey Legora GC AI Filevine Solve Intelligence Irys Descrybe
Will AI replace lawyers?: The products in the index automate parts of legal work rather than assume a lawyer's professional responsibilities. EvenUp prepares demand materials from case records, Briefpoint drafts discovery documents, and Brightflag analyzes legal invoices. Each changes the work involved in producing or reviewing a particular output. Their effects on employment and client demand are still uncertain. California's practical guidance, approved May 14, 2026, expressly addresses agentic systems that can initiate tasks and act with less prompting. It retains attorney responsibility for professional judgment and meaningful supervision. The NYSBA task force report approved April 6, 2024, likewise connects AI use with independent judgment, confidentiality and oversight. Current products therefore support specific legal processes, while the professional duties described in those sources remain with the lawyer.
Brightflag Briefpoint EvenUp

AI for lawyers: which legal AI tools fit which practice

By kind of practice

Solo and small firms

Midsize and large firms

In house legal teams

Litigation and e-discovery

Plaintiff and personal injury firms

IP and patent practices

Where the tools differ

What happens to client data

Whether the citations can be trusted

Privilege and ethical walls

Security certifications

Disclosure to a court

General AI or legal AI

What it costs

What courts and bars have said

Every category in the index

Common questions

Sources

Correct a record, or ask how something was graded