AI Agents for Lawyers

Comparisons

Legal-Specific AI vs General-Purpose AI: Do Lawyers Need a Vertical Tool?

Legal-specific AI vs general-purpose AI: why legal tool use fell to 40%, what Harvey, CoCounsel and Legora add over Claude or ChatGPT, and a decision by task.

By Dr. Niklas SchmidtUpdated 13 November 202610 min read
On this page
  1. The paradox in the survey numbers
  2. What a vertical tool adds
  3. What the labs now hand you directly
  4. Where general models fail without help
  5. The cost per lawyer
  6. A decision framework by task
  7. A decision by firm size
  8. A month-long test that answers the question for your firm
  9. Where this leaves you

Most lawyers do not need a vertical legal AI tool; they need a business-tier general model with the training switch off and a citation-verification habit. Legal-specific tool use fell from 58% to 40% in a year (Clio, 2025) while overall use held at 79%. A vertical tool earns its price only when it grounds citations in a citator, connects to your document system under ethical walls, or runs a repeated workflow at volume.

The paradox in the survey numbers

Two things are true at once. Lawyers use general chatbots more every year, and law firms keep buying legal-specific platforms at rising prices.

The Clio Legal Trends Report 2025 is the clearest statement of the first: 79% of legal professionals use AI, but use of legal-specific tools fell to 40% from 58% the year before, as lawyers defaulted to general chatbots. The 8am 2026 Legal Industry Report (1,300-plus respondents) found general-purpose tools in 46% of firms and legal-specific ones in 34%, though the latter was up from 21%. The Texas Bar's 2026 survey of 1,553 lawyers had ChatGPT used by 63% of AI users and Westlaw or CoCounsel by 30%. Axiom's July 2026 survey of 528 in-house leaders found two-thirds running general-purpose AI in its default configuration, which is the real risk in that trend.

The second truth is in the money and the enterprise surveys. Harvey raised $550m at $15.5bn on 9 September 2026 with revenue above $400m; Legora is at $5.6bn. The ILTA 2026 Technology Survey of 500-plus firms puts Microsoft 365 Copilot at 76%, CoCounsel and Claude at 44% each, Harvey at 43%, ChatGPT Enterprise at 22% and Legora at 16%. A LexisNexis survey of 543 lawyers found 81% "more comfortable" with AI grounded in legal sources. Firms buy grounding and governance; individual lawyers use whatever is open in the next tab. Both are rational, and the decision framework on this site starts from that fact.

What a vertical tool adds

Strip the marketing and a legal-specific product adds five things a general model does not have out of the box.

  1. A corpus with a citator. CoCounsel answers from Westlaw with KeyCite; Lexis+ with Protégé from Lexis content with Shepard's Trust Markers; Clio Work from vLex's billion-plus documents in 110-plus jurisdictions. Thomson Reuters' chief technology officer put the pitch in one sentence: the control point "is whether the output is accurate, grounded in authoritative sources".
  2. Workflows built by lawyers. Harvey ships 500-plus pre-built agents and its customers have built 25,000-plus more; Lexis ships 300-plus configurable workflows; Legora's Tabular Review reads up to 100,000 documents in parallel; Thomson Reuters' Tabular Analysis runs 10,000 documents against 100 questions.
  3. Permissions, ethical walls and audit trails. Lexis Workrooms have dual approvals and role-based permissions; Harvey added Intapp ethical-wall enforcement on 23 July 2026; the vendors log what the agent did and who approved it.
  4. Matter-scale storage. Harvey's Vault and Lexis' Vault (100,000 documents) hold a data room where the agent can reach it; a chat window holds a conversation.
  5. One price and one throat to choke. As a lawyer put it in an August 2026 thread on rejecting Harvey, "there is also no token usage to manage. It is an all inclusive price", and another wrote that "Harvey is encrypted and trusted".

The Harvey vs Legora and CoCounsel vs Protégé pages compare the vendors on each of these; the point here is that these five things, not the model, are what the premium buys.

What the labs now hand you directly

Three of the five are no longer vendor-only, which is why the comparison has changed since 2025.

Anthropic's Claude for Legal (12 May 2026; Apache 2.0) ships twelve practice-area plugins, 150-plus skills and 13 scheduled agents at no extra charge to paid Claude customers: tabular diligence with "every cell cited", DSAR responses on the statutory clock, MSA and NDA review to Word tracked changes, a 90-day renewal tracker. Its 20-plus MCP connectors reach iManage, NetDocuments, DocuSign, Ironclad, Relativity, Everlaw, Box, Datasite, CoCounsel, LexisNexis and Harvey, and iManage's MCP server (14 May 2026) is "permission-bound, fully logged" and "respects existing ethical walls" whether the client is Claude, ChatGPT, Copilot, Harvey or Legora. OpenAI's Astra for Law (17 September 2026) adds a Legal Search Index of 230 million-plus URLs of US law, 26 partner plugins and a zero-data-retention API. Microsoft's Word Legal Agent redlines against a playbook for $30 a month inside Copilot. Perplexity's Computer for Counsel routes sub-tasks across 20-plus models, and Gunderson Dettmer runs 35,000 queries a month on it.

So workflows (item 2) and permissions (item 3) are available direct, and a citator (item 1) is reachable through a connector if you hold the underlying subscription. What the labs still do not own is the corpus, and what they do not offer is a single vendor accountable for the whole stack. The Claude for Legal plugins page walks through what is in the box.

Where general models fail without help

The failure mode is the same in both categories, and the evidence says neither is safe unverified.

HAQQ's June 2026 benchmark of 3,000 graded answers across 51 practice areas found that 24% of all answers cited or misapplied law that did not support the claim, that every model, including the leaders, fabricated somewhere, and that Mistral Large produced citation errors 64% of the time; GPT-5.5's 3% was the best citation-hallucination rate, at 134 seconds and $0.082 a task. Vals' October 2025 study had ChatGPT Team at 80% against a 69% lawyer baseline, but every system lost 11–14 points on multi-jurisdiction questions, and the legal-specific tools beat ChatGPT on authoritativeness by six points. A University of Pennsylvania study in April 2026 found that search-augmented models and deep-research agents fabricate 3–13% of the URLs they cite, with agents hallucinating more because they cite more.

The verticals are not exempt. Stanford's 2024 test found 17% hallucination in Lexis+ AI and 33% in Westlaw AI-Assisted Research, and neither company has submitted a current product to an independent test since. In the hallucination cases that named the tool in an early-2026 analysis, ChatGPT accounted for 50%, Westlaw AI 18% and Claude 15%. Lacey v. State Farm (C.D. Cal., 2025) cost $31,100 and involved CoCounsel and Westlaw Precision; Fletcher v. Experian (5th Cir., 2026) involved vLex and CoCounsel; Wadsworth v. Walmart (D. Wyo., 2025) involved Morgan & Morgan's in-house platform. And in a 30-lawyer firm's bake-off of CoCounsel, Harvey and Legora in August 2026, "every system missed" a disqualifying heirship issue in intake notes; only Claude Opus at maximum effort caught it.

The verification protocol is therefore not a general-model precaution. It is the cost of using either category, and it should be in the budget before the tool is.

The cost per lawyer

TierProductPrice per user per month (Sep 2026)Minimum
General, businessClaude Team; ChatGPT Business; Microsoft 365 Copilot; Perplexity Enterprise Pro$25 (Claude standard; $125 premium); $25 ($20 annual); $30; $40Claude 5 seats
Legal, entryAI:ssociate (Austria, ÖRAK partner); CoCounsel Legal; Spellbook; Clio Workfrom €39; $104–$639; about $99–$350 (mid about $180); $199None reported
Legal, in-houseGC AI$500Not reported
Legal, platformLegora; Harveyreported from about $3,000 a year (about $250 a month); estimated $1,000–1,200 plus consumption10 seats; 20 seats (both reported)

The pricing guide sources every figure. The ratio to notice is that a platform seat costs forty times a general seat, and the platform runs the same Claude or GPT model underneath. The premium has to be justified by the five items above, on your matters, in a pilot.

A decision framework by task

TaskBusiness-tier general modelA vertical tool earns its price when
Summarise or compare documents you supplyYesNever for this alone
First-draft clauses, emails, memos from your own materialsYes, with Projects, Skills or a custom GPTContract volume is high and a playbook must be enforced across a team (Spellbook, Word Legal Agent, Harvey)
Legal research with citationsNot unverified; use it for issue-framing and search stringsAlways, if you can: a citator-grounded tool (CoCounsel, Protégé, Clio Work) or Astra's index for US law, then open every source
Contract review at scalePartly, one document at a timeMore than a few dozen contracts a month, or approvals and audit trails are required
Due diligence in a data roomNoAt any real scale: Harvey Vault, Legora Tabular Review, Thomson Reuters Tabular Analysis, or Claude for Legal's diligence grid connected to Datasite
E-discovery reviewNoAlways: Relativity aiR and its peers, validated under the court's protocol
Regulatory monitoringPartly, with a scheduled agentMulti-jurisdiction coverage and change logs are needed
Intake, scheduling, admin automationYes, with n8n, Zapier or MakeNever

Two rows deserve emphasis. Research is the row where the general model is weakest and the vertical strongest, because grounding in a citator is exactly what a general model lacks; even then, the research use case shows why the protocol remains the same. Admin automation is the row where the vertical adds nothing: a three-lawyer German firm runs intake, onboarding and deadline capture on self-hosted n8n and a local model for €20–30 a month.

A decision by firm size

Solo and small firms (under ten lawyers). A business-tier general model for everyone, with training off and the confidentiality checklist applied, plus one contract add-in or Clio Work if the work justifies it, and a research subscription you already hold. Harvey's reported 20-seat minimum and Legora's 10-seat minimum make the platform question moot; the small-firm guide sets out a stack.

Mid-sized firms (ten to fifty). A general tier for all, a citator-grounded research tool for the litigators, a contract add-in for the transactional group, and a pilot of Legora at ten seats if diligence volume is real. Run the pilot as a six-week test with a go/no-go, not a rolling trial.

Large firms (fifty-plus). A platform, a general tier and the research incumbent, which Law.com has called Big Law's "AI Tower of Babel". The discipline that separates Axiom's 7% of teams that measure results from the rest is UBS's: 80 candidate use cases narrowed to three with a 140% return hurdle.

In-house teams. The buyers' market has shifted toward them: Spellbook draws more than half its revenue from in-house, GC AI prices at $500 a seat, and Legora now runs Salesforce's, Barclays' and Erste Group's legal functions. The in-house guide covers the build-versus-buy question that Deloitte's June 2026 survey says 61% of departments are now piloting as agents.

A month-long test that answers the question for your firm

  1. Week one: put every lawyer on a business-tier general model with training off, and teach the verification protocol. Module 1 of the AI Lab for Lawyers course covers the prompting and the privacy settings that make this tier safe.
  2. Week two: log the tasks people actually run. Most firms find the top five are summarising, drafting from their own materials, email, research framing and translation.
  3. Week three: take the two tasks from the log that a vertical claims to do better, usually research and volume review, and run five real (closed-file) examples through the vertical's trial and through the general model. Open every citation in both.
  4. Week four: price the vertical against the measured difference, per seat, for the lawyers who ran those tasks only. Buy seats for them, not for the firm.

Where this leaves you

The question in the title has a short answer for each half of the profession. For the lawyer, no: a business-tier general model, correctly configured, is the daily tool, and it now carries legal plugins and connectors that used to be the vertical's whole pitch. For the firm, sometimes: research that must be grounded, review at data-room scale and workflows that need approvals and audit are still where CoCounsel, Protégé, Harvey and Legora justify seats, and those seats should go to the lawyers who do that work. What neither half can skip is the verification step, because the database of 2,041 hallucination decisions worldwide includes filings produced with tools from every row of the price table.

Frequently asked questions

Do lawyers need Harvey or is ChatGPT enough?

For most lawyers and most tasks, a business-tier general model is enough, provided it is a tier that does not train on your data and every citation is verified. Harvey, Legora, CoCounsel and Lexis+ Protégé pay for themselves when the work is grounded research, diligence at data-room scale, or workflows that need approvals and audit trails across a team. Harvey's reported 20-seat minimum rules out most small firms in any case.

What does legal-specific AI do that ChatGPT or Claude cannot?

Four things, as of September 2026: it grounds answers in a proprietary corpus with a citator such as KeyCite or Shepard's; it runs pre-built legal workflows at volume, for example Tabular Analysis across 10,000 documents; it enforces permissions, ethical walls and audit trails; and it bills one all-inclusive price with a vendor to hold accountable. General models now reach some of this through plugins and MCP connectors, but they own no legal corpus.

Why did use of legal-specific AI tools fall while ChatGPT use rose?

Clio's 2025 Legal Trends Report found legal-specific AI use fell to 40% from 58% while overall use stayed at 79%, and the 8am 2026 report found general-purpose tools in 46% of firms against 34% for legal-specific ones. Lawyers default to the tool that is open all day, and frontier models improved faster than most vertical products. Firms still buy verticals for research grounding and governance, so both trends run at once.

Is a general AI model safe for legal research?

Not without verification, and neither is a legal-specific one. HAQQ's June 2026 benchmark found 24% of 3,000 answers cited or misapplied law that did not support the claim, with every model fabricating somewhere. Stanford measured 17% and 33% hallucination in earlier Lexis and Westlaw AI products. Use a citator-grounded tool for research where you can, and open every authority before it goes in a filing whichever tool produced it.

How much more does legal-specific AI cost than a general model?

Roughly four to forty times per seat. Claude Team and ChatGPT Business are about $25 a seat per month and Microsoft 365 Copilot $30. CoCounsel Legal publishes $104 to $639, Spellbook about $180 at its mid tier, Clio Work $199, GC AI $500. Harvey is estimated at $1,000 to $1,200 a seat with a 20-seat minimum and Legora from about $3,000 a year with a 10-seat minimum, both from secondary sources.

Put this into practice

AI Lab for Lawyers: a live, hands-on course for practising lawyers

4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.

Reserve a seat · €200 Maven money-back guarantee · certificate of completion

Guides

How to Choose an AI Tool for Your Law Firm: A Decision Framework

How to choose legal AI for your law firm: four routes, six vendor questions, a weighted scoring matrix and a six-week pilot, with September 2026 prices.

11 min read

Guides

What Legal AI Really Costs in 2026: Harvey, Legora, CoCounsel, Claude and ChatGPT

Legal AI cost as of September 2026, from €39 to $1,200 per seat per month, with Harvey, Legora, CoCounsel, Claude and ChatGPT prices and worked budgets for 5 and 50 lawyers.

10 min read

Tools

Claude for Legal Plugins Explained: The 12 Practice-Area Plugins and 20+ Connectors

Claude for Legal gives paid Claude users 12 open-source practice-area plugins, 150+ skills and 20+ connectors to iManage, NetDocuments and CoCounsel.

8 min read

Comparisons

Harvey vs Legora: The Two Legal AI Platforms Compared

Harvey vs Legora in September 2026: $15.5bn vs $5.6bn valuations, ARR, clients, Workflow Agents vs Agentic OS, models, pricing minimums, geography and who should pick which.

10 min read

Playbooks

How to Verify AI Legal Research: A Citation-Checking Protocol

Verify AI legal research with a five-check protocol: open every authority, match the quote, confirm holding and status, log it. Template, time budgets, tools.

11 min read

Guides

AI for Small Law Firms and Solo Practitioners: What to Use and What to Skip

AI for small law firms in 2026 means a business-tier model, one legal tool and simple automation for under €200 a month, plus the policy and privacy settings most solos skip.

9 min read