Your attorneys are likely already using one of these. The question is which tier, with what controls in place.
Last verified September 14, 2026 against vendor pages
Ratings are L2's judgment as of the date shown.
Swipe sideways to compare all three →
|
Anthropic
Claude
|
OpenAI
ChatGPT
|
Microsoft
Copilot
|
|
|---|---|---|---|
| Best for | Document & template creation across file formats; complex multi-step workflows | Broad general assistant; custom GPTs; complex workflows; computer control | Security and org-level controls for firms already using M365 |
| Capabilities relevant to law firms L2 rating | ★★★★★ | ★★★★★ | ★★★★★ |
| Connector compatibility L2 rating | ★★★★★ | ★★★★★ | ★★★★★ |
| Entry price | $$$ | $$$ | $$$+ requires M365 |
| Full governance tier | $$$seats + usage billed separately | $$$quote-only; sales-assisted | $$$+ requires M365 |
| HIPAA BAA | ✓Enterprise only | ✓Sales-managed Enterprise only | ✓Included with M365 |
| No training on your data business tiers | ✓Yes | ✓Yes | ✓Yes |
Price legend (relative only): $ = lowest entry cost · $$ = mid-range or seat fee + metered usage · $$$ = premium / quote-only. Dollar figures change too often to publish; see the "last verified" date above.
Copilot ranks last on functionality and first on HIPAA convenience. That is the real tradeoff for a small firm handling PHI.
Ratings are L2's judgment as of the date shown. We re-evaluate as the tools change — stay up to date with what's current.
The one rule that matters most: consumer tiers are for personal use; business tiers are for client work. Consumer plans train on your data unless you opt out, have no admin controls, and no HIPAA path.
Set expectations up front: these three products change monthly. Features, security controls, and pricing models shift, and the most capable tool for legal work today may not be the most capable one next quarter. This page reflects what we verified on the date shown; treat it as a starting point, not a final answer.
For the ethics side — ABA Formal Opinion 512 and the state bar opinions, court rules and guidance that apply to any of these tools — see our Formal Legal Guidance on the Use of AI Tools.
L2 tracks these changes for the firms we work with. If you want a current read on your specific situation, start with an AI Governance Review.
On this page
Claude's edge is how well it works with the file formats lawyers actually use — Word, PDF, Excel, PowerPoint, scanned records — and how reliably it carries a complex, multi-step workflow from start to finish. Hand it a form agreement, a term sheet, and the firm's drafting conventions and it produces a usable first draft; ask it to reconcile a redline, build a closing checklist from a purchase agreement, or turn a deposition into a cited digest and it follows through. Works in chat and in Cowork. Very long context handles full contracts, records, and transcripts in one pass. Skills let a firm package repeatable workflows and deploy them firm-wide. Works inside Word and Outlook via Claude for Microsoft 365.
Caveat: Cowork and the Office add-ins are not covered by Anthropic's BAA — HIPAA firms keep PHI in chat and Projects.
Widest custom-GPT ecosystem; shared Projects, workspace agents, deep research, computer control, Excel/PowerPoint extensions. Excellent for research memos, summaries, and correspondence, and capable on multi-step tasks. Less purpose-built for template-driven document production than Claude.
Caveat: use reasoning mode for long documents; the default mode has a shorter working memory on the Business tier.
Already in Word, Outlook, and Teams with your permissions applied — the lowest change-management cost of the three. Good for email triage, meeting recaps, and first-pass edits. Weaker on complex, multi-step legal work and template creation.
Caveat: Copilot surfaces whatever a user can already access — clean up SharePoint permissions before turning it on.
None of these replace legal research databases. All three can invent citations.
This ranking will move. New models ship every few months and the gap between these three is smaller than it was a year ago. L2's AI Pilot Program puts the current versions in front of your attorneys on your real work so the decision is based on results, not marketing.
Connectors are how the AI reaches the firm's documents, email, calendar, DMS, and practice-management systems. Without them the tool only knows what you paste in.
Claude ★★★★★ — native connectors plus open MCP support: NetDocuments, Microsoft 365, Google Workspace, Slack, and any system with an MCP server. Admins control which connectors are allowed.
ChatGPT ★★★★★ — broad first-party connectors (Microsoft 365, Google Workspace, Slack, GitHub) plus apps and developer mode for custom integrations.
Copilot ★★★★★ — deep inside the Microsoft tenant; reaching outside it depends on Copilot connectors and Copilot Studio, which are heavier to set up. Fine if everything lives in M365; limiting if the DMS or PM system does not.
L2 is a NetDocuments partner and has connected all three tools to law-firm DMS and practice-management systems. We can tell you in an hour which one fits your stack.
Business/Enterprise tiers of all three: no. Consumer tiers: assume yes.
Copilot: yes, already in your M365 agreement; web-search queries excluded.
Claude: Enterprise plans only; admin click-accepts in settings; excludes Cowork and Office add-ins.
ChatGPT: sales-managed Enterprise only; not the Business tier.
A BAA is necessary, not sufficient.
SSO on all business tiers. SCIM, audit logs, and custom retention require the Enterprise tier on Claude and ChatGPT and come with M365 on Copilot.
The 2025 New York Times v. OpenAI litigation is the illustration worth remembering: a court preservation order required OpenAI to retain consumer and API chat logs that users had deleted or that would otherwise have expired. "Deleted" is a policy, not a guarantee — and enterprise tiers with contractual retention controls are worth the step up.
Security terms change as often as features do — BAA scope, retention options, and which tier gets which control have all shifted in the past year. L2's AI Governance Review includes a current-state check of the vendor terms against your firm's obligations.
Claude: flat per-seat on Team; per-seat plus metered usage on Enterprise (admins can cap spend). Predictable seats, variable usage.
ChatGPT: flat per-seat on Business; Enterprise is quote-only and sold sales-assisted. OpenAI publishes no seat minimum, but in practice it is sized for larger organizations than most small firms.
Copilot: an add-on to an existing M365 license — never standalone. Bundles are cheaper than stacking a la carte.
Rule of thumb: at entry level the three are close enough on cost that price should not drive the decision. At the governance tier, structure matters more than list price — metered usage, seat minimums, and base-license requirements each change the real number.
Prices moved on all three vendors in the last six months. L2 models the actual cost for your headcount and usage as part of an AI Governance Review.
Most firms: Claude Team (Enterprise when governance or HIPAA requires it) for drafting and workflows.
Firms that want zero change management: Copilot — accept the functionality ceiling.
Firms that want the broadest general assistant and don't handle PHI: ChatGPT Business.
Common pattern: Copilot firm-wide + Claude for a power-user group. Note the second-vendor SSO/offboarding burden.
Nobody: consumer accounts for client work.
Whichever you pick, plan to revisit the decision in twelve months. The tools will have changed.
Two ways to turn this comparison into a decision for your firm.
Short engagement
We look at your practice areas, data obligations (HIPAA, client confidentiality requirements), existing stack, and headcount, and recommend a provider, tier, and rollout plan. Includes a current-state check of each vendor's security terms and pricing model.
Start a Governance Review30 days · 5–10 attorneys
On your real matters. We configure the tool, connect it to your DMS, build two or three firm-specific workflows, and measure what changed. You decide with evidence.
Ask about a PilotThe path we recommend
Ethics guidance for every step: Formal Legal Guidance on the Use of AI Tools.
And when the next model ships, we'll tell you whether it changes the answer.
Tell us a little about your firm and we'll follow up with a current read on which tool and tier fit.
We typically respond within one business day.
Sources · fetched September 14, 2026
For general guidance only, not legal advice. Ratings reflect L2's judgment as of the verification date and will change as the products do. All product names, logos, and brands are property of their respective owners. Anthropic, OpenAI, Microsoft, and NetDocuments are trademarks of their respective companies. Use of these names does not imply endorsement or affiliation unless otherwise stated.