Updated May 2026. Reading time: 11 minutes.
An AI paralegal is a custom-built AI specialist trained on your firm's matters, templates, and tone. At FellowHire we call ours Quinn. Quinn drafts client correspondence, summarizes intakes, prepares discovery, and tracks deadlines. Here is what an AI paralegal actually does, what it cannot do, and how to pick the right one for your firm.
A good AI paralegal does the recurring paralegal work that buries a small or mid-size firm. Not "answers questions about law". The actual paralegal work. Here is the real list.
Client intake summaries.
The fellow reads the intake form, the engagement notes, and any uploaded documents. It writes a one-page summary for the attorney with the matter type, the parties, the key dates, and the open questions.
Document review.
The fellow reviews discovery, contracts, and pleadings against your firm's review checklists. It flags issues, surfaces relevant clauses, and groups documents by topic for the attorney.
Matter prep.
The fellow prepares the matter file. Cover sheets, engagement letters from your template, conflict checks, calendared deadlines, and a starter task list.
Deadline tracking.
The fellow tracks every deadline across every active matter. It posts daily reminders in your firm's Slack or Teams channel. It flags conflicts. It nags before deadlines slip.
Conflicts checks.
The fellow runs conflict checks against your firm's database before each new matter opens. It produces a clean report with any potential matches and a recommended action.
Discovery review.
The fellow scans through document productions, identifies privileged material, flags hot documents for attorney review, and produces a privilege log.
Drafting client correspondence.
The fellow drafts client status updates, follow-up emails, and document requests in your firm's voice. The attorney reviews and sends.
Each of these is recurring work. Each of these takes hours of paralegal time per week. Each of these is exactly what an AI paralegal is built for.
Generic AI tools are general-purpose. ChatGPT, Claude, and Microsoft Copilot are excellent at general writing and reasoning. They are not built for your firm. They have not seen your matter intake form. They do not know your engagement letter template. They have never read your firm's tone of voice in a status update.
An AI paralegal is the opposite bet. It is custom-trained on your firm's material before it ever takes a task. The fellow gets your matter types, your filing structure, your client correspondence templates, your retainer language, your jurisdiction-specific quirks. The fellow learns your firm.
This is the wedge. Generic tools start every task from zero. An AI paralegal starts every task from your firm's standard. The output is closer to what your senior paralegal would produce on day one. The attorney edits less.
The pattern matters more in legal than in most fields. A status email in a paralegal's voice is fine. A status email in a generic AI tool's voice gets caught by the partner on review. Voice and template fidelity are the work. The fellow is built for that.
A contract paralegal is a real human you hire by the hour or by the project. Many small firms run on contract paralegal labor. The model works. It also has limits.
A contract paralegal is not always available. They have other clients. They have heavy weeks and light weeks. The work that is urgent today may have to wait until they have bandwidth.
A contract paralegal does not stay scoped to your firm. They learn your matter types over time, but they also work for other firms. They do not absorb your tone the way a fellow does, because they are also writing in three other firms' tones.
An AI paralegal is always available. It is scoped to your firm only. It does not burn out. It also does not replace the contract paralegal entirely. The model that works best is paired: the AI paralegal handles the recurring drafting and review work, the contract paralegal handles the judgment-heavy work and the work the attorney wants a human owning. The two do not compete. They cover different ground.
The honest scope. We are clear about this because trust in legal markets is built on accuracy about limits.
Give legal advice.
The fellow is not a lawyer. It does not give legal advice to clients or to your firm. It drafts. The attorney advises.
File documents with courts.
The fellow does not file in PACER, state courts, or any other filing system. The attorney files. The fellow can prepare the document.
Replace attorney judgment.
The fellow does not decide whether to settle, what to argue, or how to advise a client. Those are attorney decisions. The fellow supports the work.
Sign engagement letters or fee agreements.
The fellow drafts, the attorney signs.
Make client commitments.
The fellow does not promise outcomes, deadlines, or fees on behalf of the firm. The attorney commits.
Run with no oversight.
Every fellow output goes through attorney review on the work that matters. The fellow is paired with attorney review by design, not as a workaround.
The point of being clear about limits is that trust is built on accuracy. A fellow that knows what it cannot do is more valuable than a tool that pretends it can do everything.
Legal practice is regulated. The ABA Model Rules and your state's parallel rules govern competence, confidentiality, supervision, and client communication. An AI paralegal has to respect all of them. We build for that.
Competence and supervision. Under ABA guidance, attorneys are responsible for the work product their team produces, including AI-assisted work. The fellow operates under attorney supervision. Every meaningful output is reviewed before it goes to a client or a court.
Confidentiality. Client material is privileged. The fellow handles it as privileged. Per-firm isolation, no cross-firm data leakage, no training on your matters. Engagement letters can include language about AI-assisted work where your jurisdiction recommends or requires disclosure. We help your firm draft that language.
Malpractice posture. The fellow does not give legal advice and does not file documents. The attorney remains the responsible party of record. The fellow is a tool that produces drafts, summaries, and reviews. Treating the output as draft work product matches existing malpractice patterns.
FellowHire is SOC 2 and ISO 27001 compliant. Encrypted in transit (TLS 1.3) and at rest (AES-256). No-training enterprise agreements with OpenAI and Anthropic.
The fellow is only as good as the tools it can read and write. Quinn connects to the systems your firm already runs.
If your firm uses a tool not on this list, ask. We add integrations on customer demand. See the Clio integration deep dive →
AI paralegal (Quinn)
Paxton.ai
ChatGPT or Claude
All three have a place. The question is what your firm needs. For one-off legal research, ChatGPT or Claude are fine. For broad legal-domain drafting tools that work across firms, Paxton is the leader. For a paralegal that knows your firm and lives in your firm's workflow, an AI fellow is the model.
Questions to ask any AI paralegal vendor before you sign.
Bring this list to every demo. The vendor's answers will tell you whether they built a real tool for legal teams or a generic AI with a paralegal label.
Quinn is on the FellowHire Standard tier. The Standard tier is $18,000 per year, per fellow, billed annually. That includes the custom build, the firm-specific training, the integrations, and the ongoing recalibration.
The whole firm uses Quinn for that one price. Five attorneys or fifteen, the price is the same. There are no per-seat fees. There are no per-message credits. There are no surprise bills. One annual price, predictable in your budget.
If your firm needs a Senior fellow for more complex matter types or a higher autonomy level, that is the Senior tier at $48,000 per year per fellow. Most paralegal work fits in the Standard tier. Your firm decides which fits based on the role scope at the demo.
No. The model is paired. The AI paralegal handles the recurring drafting and review work. The human paralegal handles the judgment-heavy work, the client-facing work the attorney wants a human owning, and the supervision of the AI. Most firms keep their paralegal and add a fellow.
No. The fellow drafts and reviews. The attorney advises. Giving legal advice is unauthorized practice for non-attorneys, including AI. We build the fellow to respect that line.
Yes. FellowHire is SOC 2 and ISO 27001 compliant. Encrypted in transit (TLS 1.3) and at rest (AES-256). No-training enterprise agreements with OpenAI and Anthropic.
Harvey is a legal AI tool focused on BigLaw and corporate legal. It is excellent at legal research and document drafting at the firm-wide enterprise level. FellowHire's Quinn is built for small and mid-size firms that want a paralegal-shaped fellow inside their Slack or Teams. Different scope, different price point, different deployment model.
Paxton is a vertical legal AI with its own web app. It is trained on legal data broadly and works across firms with the same underlying model. Quinn is custom-built per firm and lives in your Slack or Teams. The pricing models are different. The deployment model is different. Both are legitimate options for different firm sizes.
ChatGPT is a general-purpose AI assistant. It is not trained on your firm. It does not connect to Clio. It does not live in your firm's Slack or Teams. For ad-hoc research or general writing it works fine. For repeat paralegal work it is not the right tool.
One call to scope the role. About a week to build. Quinn lives in your Slack or Microsoft Teams, connects to Clio or your case management tool, and starts on day eight. Predictable annual pricing.