Stack guide · 2026
A sane AI stack for legal / law firms (2026): a ladder, not a 12-logo AI lawyer
Start with the leak you can name. Add Spellbook when the file is open in Word, CoCounsel when Westlaw is already the desk, Harvey when the firm will implement a platform. Skip the rest until a step is leaking. Not legal advice.
Gregory Greenstein, Aiacent
- Published
- Updated
- Pricing verified
- 16 min read
- How we rank
Harvey
8.3/10Best enterprise legal AI
The enterprise legal assistant — firm workflows, not a gavel on ChatGPT.
Harvey is the legal AI we would shortlist when a firm already lost the ChatGPT argument with IT and still needs generation on matter files. Assistants, workflows, a knowledge layer. That is a platform. It is not 'ChatGPT for lawyers' with a better landing page.
- Best for
- Law firms and legal teams that will implement a platform, not a $20 tab.
- Skip if
- You are a solo who needed Word redlines or a careful generalist. That is Spellbook or Claude — and this is not legal advice either way.
Why it ranks
- The legal-AI name large firms actually put through security
- Workflows for research, drafting, and review that are not a naked chat window
- Built for privilege, audit, and a knowledge base you will govern
Watch-outs
- You will implement it like software. There is no honest $20 Harvey
- Overkill for a solo who needed a redline in Word
- A model that sounds like a partner is still not a partner. You sign the brief
Product look
Harvey
Figure slot. No vendor screenshot until we have a rights-cleared file.
CoCounsel
8.1/10Best TR-stack legal AI
Thomson Reuters' legal AI — research and drafting next to Westlaw, not a consumer tab.
CoCounsel (the Casetext product inside Thomson Reuters) is the legal research and drafting assistant we would start a Westlaw shop on. Skills for review, memos, and deposition prep that sit next to sources the firm already pays for. That is a different seat from Harvey's platform sale and from Claude's $20 tab.
- Best for
- Firms and departments already in the Thomson Reuters / Westlaw orbit who needed AI on that stack.
- Skip if
- You are not buying research from TR, or the job is a redline in Word you already have open. Spellbook. And this is not legal advice.
Why it ranks
- The honest AI layer if Westlaw is already how you research
- Drafting and review skills that assume a legal source, not a blog
- Easier security conversation than a consumer chat window
Watch-outs
- A bad buy if you do not already live in Thomson Reuters
- Not Spellbook — you will miss the Word pane on a Monday markup
- Citations still need a human. A footnote from a model is not a table of authorities
Product look
CoCounsel
Figure slot. No vendor screenshot until we have a rights-cleared file.
Spellbook
7.9/10Best Word contract AI
AI in Microsoft Word — redlines and drafts on the contract you already have open.
Spellbook is the legal tool we would give a deal lawyer on a Tuesday. It sits in Word. It drafts a clause, flags a missing provision, suggests a redline. That is a real job. It is not Harvey, and it is not CoCounsel, and it is not a reason to skip either if the buying meeting is research or a firm OS.
- Best for
- Transactional lawyers who mark up contracts in Word and will review every suggestion.
- Skip if
- The job is legal research you will cite, or a firm-wide assistant. CoCounsel or Harvey. Not legal advice.
Why it ranks
- The add-in is the workflow. You do not paste a 40-page SPA into a chat tab
- Draft, review, and redline suggestions a transactional lawyer will actually try
- The honest 'AI for lawyers' buy when the file is already a .docx
Watch-outs
- Not a research engine. Asking it for case law is how you invent a holding
- A confident markup is still a first pass. You own the blackline
- Litigation and knowledge-management shops will feel the missing platform
Product look
Spellbook
Figure slot. No vendor screenshot until we have a rights-cleared file.
Claude
8.2/10Best careful ChatGPT alternative
The careful generalist — long drafts, messy docs, and fewer confident inventions.
Claude is the ChatGPT alternative we would actually switch to for writing, not the one we would buy to save money. Same $20 band. Different temperament: longer context, fewer swaggering citations, better at 'read this pile and do not hallucinate the brief.' Claude vs ChatGPT is the head-to-head by writing, coding, research, and long docs. Perplexity vs ChatGPT is the research-tab fight — Claude is the pile you already have, Perplexity is the public web.
- Best for
- Operators who write from long sources and want a model that will push back.
- Skip if
- You needed the cheapest Plus-shaped seat with plugins, or a shared brand desk.
Why it ranks
- Still the generalist we trust most on long, messy source material
- Pushes back instead of cheerfully inventing a citation
- Artifacts and projects are close enough to a writing room for one person
Watch-outs
- No shared brand-voice system — that is still Jasper
- Plugin and browsing culture is thinner than ChatGPT's
- Easy to treat a careful draft as a finished one
Product look
Claude
Figure slot. No vendor screenshot until we have a rights-cleared file.
People search 'AI stack for law firms' or 'AI legal stack' after a vendor demoed an associate that researches, redlines, and files. The honest 2026 stack is smaller: a leak you can name in a partners' meeting, then one seat per bottleneck. If you cannot name the bottleneck, you are not shopping for a stack. You are shopping for a slide that will not survive the next ethics memo.
This is a ladder, not a dump — and it is not legal advice. The legal ranking is the scored practice desk: Harvey 8.3 for a firm platform, CoCounsel 8.1 when Thomson Reuters / Westlaw is already the operating system, Spellbook 7.9 for Word redlines, Claude 8.2 as a careful solo tab (not a research product). The writing ranking is the sentence desk. The research ranking is public-web and paper tools — Perplexity, Elicit — not a reporter you will cite. The free-tools ranking is where a solo actually starts without a card. Read this one if you are trying not to buy four legal logos in one quarter — or to add the second one on purpose.
Start here
The matter system you already have, and a number. Word markups that slip? Research that already lives in Westlaw? A ChatGPT ban from IT with no replacement? Long memos a solo will still read? Most 'we need an AI legal stack' tickets are one of those tickets with extra procurement. Claude Pro is $20 — Anthropic's public Pro seat, the same band we cite on is Claude Pro worth it, checked 19 September 2026. It will draft a structure you already understand. It will not become Westlaw, and it will not sit in the Word pane.
Stay on the DMS plus a careful tab until a named leak shows up: the contract is open in Word and paste is the compliance incident, research has to start from sources the firm already pays for, the firm will implement a platform security will sign, or a solo needs a long-doc temperament and will not treat the draft as advice. If none of those are true, another logo will not make Tuesday's markup move. Privileged files in a consumer window are a malpractice story, not a productivity win.
- Start: the DMS and Word you already run, a leak with a number, then one legal job
- Do not start: a 12-logo 'AI lawyer,' Harvey because a keynote said associate, or ChatGPT on a privileged memo 'just this once'
- Honest watch-out: a prettier ranking will not invent a partner, and a stack will not invent a policy
Add the next seat when the bottleneck is named
Add Spellbook when the file is open in Word
The leak is a contract already in .docx: a missing provision, a first redline, a playbook you actually wrote. Spellbook is the pane. Public pricing is a per-lawyer subscription on spellbook.legal — we will not invent a sticker the public page did not print as a dollar this pass. Review every suggested clause. Skip Spellbook if the job is research you will cite (CoCounsel). Skip it if the buying meeting is a firm OS (Harvey). Skip it if you will accept a clause without reading it. The legal ranking is why this seat is third on purpose — most small-firm tickets that are not Claude are this Word-shaped leak.
Add CoCounsel when Westlaw is already the desk
The leak is research and drafting next to sources the firm already pays Thomson Reuters for. CoCounsel (the Casetext product inside TR) is the skill layer, not a consumer tab. Quote-only on the TR bill — we will not invent a typical contract. Read every cite. Skip CoCounsel if you are not in that stack. Skip it if the document is open in Word and the leak is the markup. Lexis+ AI if you already live in Lexis — honorable mention on the legal ranking, same job family, other publisher. ChatGPT and Claude are not research products you will file. Perplexity is a public-web brief. The research ranking is that public desk, not a table of authorities.
Add Harvey when the firm will implement software
The leak is matter workflows security will sign: research, draft, review on files you still own, a knowledge layer you will govern. Harvey is a platform sale. Enterprise quote — there is no honest $20 Harvey. Staff the implementation. Skip Harvey if you needed a Word add-in this month. Skip it if research on Westlaw is the actual leak. Skip it if you are three lawyers and a bookkeeper — Claude Pro is the honest tab until you have an IT project. The founder take on /tools/harvey is still the split: most 'we need Harvey' tickets from a five-lawyer shop are a Spellbook or Claude ticket with extra keynote.
Stay on Claude (or ChatGPT) when you are a solo who will not get Harvey — and keep client facts out of the prompt when you can
The leak is a long messy PDF, a memo structure you already understand, a draft that should push back. Claude Pro is $20. ChatGPT Plus is $20 — more eager, same band, honorable mention on the legal ranking because it invents case names with confidence. Neither is counsel. Turn off training on sensitive work if the vendor offers it, prefer a Team or Enterprise seat, and describe structure instead of pasting the file when you can. The writing ranking is the temperament table. The free-tools ranking is the $0 start if you are still proving the habit. Skip both consumer tabs if the file is privileged and you do not have a contract your GC will accept.
Leave public-web research on the research ranking
Perplexity, Elicit, Consensus, ChatGPT vs Perplexity for research — those scores live on the research ranking. They are a brief on the open web or a paper pile, not a reporter. Buying Perplexity Pro to 'do legal research' is how this ladder wastes a morning and a citation. If the buyer needed Westlaw-class answers, that is CoCounsel or Lexis+ AI. If the buyer needed a thinking pass on notes they already have, that is Claude or ChatGPT — then verify every claim.
What is not a rung
- A second 'AI lawyer' because the first one felt narrow. Name the job. The legal ranking already splits platform, research, Word, and a careful tab.
- Consumer ChatGPT on a privileged memo. Honorable mention as a draft tab. Not a research product. Not scored as counsel.
- Perplexity, Elicit, or a 'best AI for research' tab as a table of authorities. The research ranking exists so this stack does not have to pretend a public brief is Westlaw.
- EvenUp as a general desk. Personal-injury damages. A specialist. Wrong default for a general practice.
- Microsoft 365 Copilot as a matter system. Word and Outlook. Productivity ranking.
- Vendor theater that sources, screens, drafts, and files so you do not have to look.
Most 'we need an AI legal stack' tickets are a Spellbook ticket, a CoCounsel ticket, or a Harvey ticket. The ones that are not are a Claude ticket with a malpractice footnote.— Aiacent stack guide, 2026
A solo vs a small firm vs a platform buyer, in one pass
Solo or in-house of one: Claude Pro at $20 if you will read the draft (is Claude Pro worth it), ChatGPT Plus if the job is mixed and everyday and you will not cite it, Spellbook if the contract is already in Word. Stop. Do not buy Harvey. Do not buy CoCounsel 'to try AI' if you are not on Thomson Reuters. The free-tools ranking is the $0 version of this paragraph. The writing ranking is the temperament version.
Small firm, a handful of transactional lawyers: keep the careful tab for structure you already understand. Add Spellbook when markups are weekly. Add CoCounsel only if Westlaw is already on the bill and research is the leak. Add Harvey only if you have an IT project and a security review — not because a demo wrote a nicer memo. That is a stack. A slide with twelve 'AI lawyer' logos is a directory.
AmLaw-shaped or in-house with a GC: Harvey if the firm will implement a platform. CoCounsel if TR is already the research OS. Spellbook if the pane is still the Tuesday leak. Claude Team or Enterprise if the tab has to be a contracted workspace, not a consumer window. Microsoft 365 Copilot if the leak is Word and Outlook, not the matter — productivity ranking. The legal ranking is the RFP if you want the four scores in one table. The research ranking is the RFP if the next buy is a public brief. The writing ranking is the RFP if the next buy is the sentence.
Verdict
Do not start from a list of logos. Start from a leak with a number. Add Spellbook when the file is open in Word. Add CoCounsel when Westlaw is already the desk. Add Harvey when the firm will implement a platform. Stay on Claude when you are a careful solo who will not treat the draft as advice. Then read the legal ranking if you want the scores, the writing ranking if the leak is the sentence, the research ranking if you meant a public brief, and the free-tools ranking if you are still proving the habit without a card. This is software. It is not counsel.