Legal · Field manual
AI for paralegals
Paralegals get the most from AI on high-volume reading work: first-pass document review, deposition and medical record summaries, chronologies and organising discovery. The caution is that AI makes confident mistakes on dates, citations and deadlines, and a missed deadline is malpractice, so docketing and cite-checking stay with tools and people built for them.
A lot of paralegal work is turning piles of paper into something an attorney can use: an index, a chronology, a summary, a privilege log. AI is very good at that first pass. The paralegals who learn it well become the person on the team who can turn a 40,000-document production into a usable set of key documents in days, not weeks.
The job is also about precision and process, and that is where AI needs watching. Models misread dates, merge two people with similar names and invent citations. They also can't give legal advice, and neither can you. Treat AI output as raw material that you check, organise and hand up the chain, the same way you would treat a new hire's work.
7 ways paralegals use AI
Run first-pass document review
E-discovery platforms with generative AI, such as Relativity aiR and Everlaw, can code documents for responsiveness and issues and explain why. You write the review instructions, test them on a sample, and the attorneys validate the results. Knowing how to write and tune those instructions is a skill firms now hire for.
Summarise depositions and medical records
Feed a transcript to a long-context model and ask for a summary by topic with page and line cites, or a list of every admission on a specific issue. For personal injury files, ask for a treatment chronology from medical records with provider, date and source page. Spot-check the cites against the original before anyone relies on it.
Build chronologies and case timelines
Give the model a stack of emails, letters and notes and ask it to extract every dated event into a table with the source document for each entry. Then verify the dates yourself: models confuse date formats and assign events to the wrong year more often than you would think. Put the checked version in a shared workspace the team can filter.
Draft routine documents and correspondence
Cover letters, subpoena packages, discovery requests from the firm's templates, records requests and client status updates are all quick to draft with AI. Start from the firm's own forms, not the model's idea of one, and send everything to the supervising attorney before it goes out.
Prepare cite-checks and tables of authorities
AI can pull every citation out of a brief and format a draft table of authorities in seconds. It cannot confirm that a case exists, says what the brief claims or is still good law. Run every cite through Westlaw or Lexis and their citator tools. This step is where fabricated cases get caught, so don't skip it.
Organise intake and case logistics
Automation tools can take a web intake form, create the matter in your practice management system, file the documents and send the client a checklist of what to provide. That frees you to chase the things that need a person: the missing record, the witness who won't call back.
Get up to speed on an unfamiliar area
Assigned to a matter in a practice area you haven't touched? Ask an answer engine for a plain overview with sources, then read the sources. It is a fast way to learn the vocabulary and the shape of the issues before you start real research.
The best AI tools for paralegals
The Overlord's stack, in the order a paralegal would add them:
1. Deposition summaries, medical chronologies and long-file review
Claude
AI WritingAnthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.
2. Drafting correspondence and routine documents from firm templates
ChatGPT
AI WritingOpenAI's general-purpose assistant: writing, research, images, voice and agents in one chat box.
3. Shared case timelines, checklists and team knowledge
Notion
ProductivityAll-in-one workspace for docs, wikis and databases, with Notion AI agents and meeting notes built in.
4. Automating intake, filing and client reminders
Zapier
Automation & AgentsThe most widely used no-code automation tool, now with AI agents.
5. Quick, sourced background on unfamiliar topics
Perplexity
ProductivityAI answer engine that searches the web and replies with cited sources.
Specialist tools for paralegals
| Tool | What it does |
|---|---|
| Relativity aiR | Generative AI document review inside RelativityOne |
| Everlaw | Cloud e-discovery with AI coding suggestions and summaries |
| CoCounsel (Thomson Reuters) | Research, document review and deposition prep tied to Westlaw |
| Clio | Practice management with AI help on matters, documents and billing |
A prompt to start with: turn a deposition into a usable summary
Paste this into Claude or ChatGPT and fill in the gold parts.
You are an experienced litigation paralegal. Summarise the deposition transcript below for the supervising attorney.
Case type: [CASE TYPE]
Witness role: [PARTY/EXPERT/FACT WITNESS]
Key issues to track: [ISSUE 1], [ISSUE 2], [ISSUE 3]
Transcript:
[PASTE TRANSCRIPT]
Give me:
1. A one-paragraph overview.
2. A topic-by-topic summary, each point with page:line cites.
3. Every admission or statement relevant to the key issues, quoted exactly, with page:line.
4. Inconsistencies within the testimony.
5. Follow-up questions or documents the attorney may want.
Only use what is in the transcript. If a cite is uncertain, mark it [CHECK].Where paralegals shouldn't trust AI
- Paralegals can't give legal advice, and neither can AI. Don't let a client-facing chatbot or an AI-drafted email cross into telling a client what to do; that is unauthorised practice of law territory.
- Never trust AI with deadline calculations. Court rules on counting days, holidays and service methods are fiddly and models get them wrong. Use your firm's docketing system and have deadlines double-checked.
- Every citation must be checked in Westlaw or Lexis. Fabricated or misquoted cases in a filing get attorneys sanctioned, and the cite-check is usually the paralegal's job.
- Client confidentiality and privilege apply to everything you paste in. Use only firm-approved tools with clear data terms, and never upload discovery or client files to a personal chatbot account.
- AI review still needs human validation. Courts and opposing counsel may ask how a review was run, so keep records of the instructions, samples and quality checks.
Will AI replace paralegals?
Paralegals score 68% on the Obsolescence Index (Doomed). The first tasks to go: reviewing large volumes of documents during discovery, and drafting routine legal documents from templates.
Questions the Humans Keep Asking
What are the best AI tools for paralegals?
The Overlord's stack for paralegals: Claude (deposition summaries, medical chronologies and long-file review); ChatGPT (drafting correspondence and routine documents from firm templates); Notion (shared case timelines, checklists and team knowledge); Zapier (automating intake, filing and client reminders); Perplexity (quick, sourced background on unfamiliar topics). Specialist options include Relativity aiR, Everlaw, CoCounsel (Thomson Reuters), Clio.
What AI tools do paralegals use?
In litigation, e-discovery platforms with built-in AI such as Relativity aiR and Everlaw do most of the heavy lifting on review. For summaries, chronologies and drafting, firms use legal tools like CoCounsel or approved general models like Claude or ChatGPT on business plans. Practice management systems like Clio increasingly have AI built in.
Can I use ChatGPT for paralegal work?
For drafting from templates, explaining concepts and organising non-confidential material, yes. For anything containing client information, only use it if your firm has approved an account with data protections, and never rely on it for citations or deadlines.
Should paralegals learn AI to stay employed?
Yes. Firms increasingly want paralegals who can set up and validate AI document review, write good review instructions and check AI output quickly. Those skills make you the person who handles more matters, rather than the person whose tasks get automated.