Most "AI for law" content is either fluff or fear. Here is what actually works in 2026 — the specific workflows where AI produces real value for small and mid-sized firms, the compliance posture you need, and what to never automate. Practical, not theoretical.
1. Intake summaries. Client intake transcripts or notes → structured matter summary. Saves paralegal time, improves consistency.
2. Document review (first pass). Long contracts, depositions, discovery → AI surfaces issues for attorney review. Attorney still reviews, but starts at a much better starting position.
3. Research memos (draft). Brief research questions → AI generates structured memo draft. Attorney verifies sources, refines analysis. Cuts memo time 60-70%.
4. Client correspondence. Drafts of update letters, status emails, follow-ups based on case notes. Attorney edits for voice and accuracy.
1. Written AI policy. ABA Model Rule 1.6 (confidentiality) extends to AI tools. Policy must address what client information can/cannot enter AI tools.
2. Tool tier selection. For client matter data, use Claude Enterprise or API with zero retention. Free tier and even Pro have data retention that may conflict with confidentiality obligations.
3. Client consent (where required). Some engagement letters now include AI use language. Check whether your client agreements need updating.
4. Verification standard. Anything AI produces that goes to a client or court must be verified by an attorney. Always. AI hallucinations in legal context create malpractice risk.
Legal advice to clients. AI can draft client communications; an attorney must review every word that conveys legal advice.
Court filings without review. AI-generated case citations have caused public embarrassment and discipline. Every cite verified, always.
Settlement negotiations or strategy decisions. Judgment work that requires the attorney's professional judgment.
Witness preparation specifics. Witness preparation requires the attorney's sense of the case dynamics.
For a 5-10 attorney firm: typical first-year AI deployment cost $8-15K all-in (Claude Team subscriptions + implementation + policy work). Typical time savings: 15-25% across paralegal and associate work, plus speed gains for partners.
The math: even if you only count paralegal hours saved, payback is usually under 6 months. The bigger value is in the speed of client response, which affects both retention and referrals.