AI Prompts for Paralegals (2026)

This article is informational only and is **not legal advice**. AI prompts do not create an attorney-client relationship and do not substitute for the judgment of a licensed attorney. Every AI output must be reviewed and approved by a supervising attorney before it is relied on or sent.

**Never paste client confidential data, privileged material, personally identifiable information (PII), or case-identifying details into a public chatbot.** Treat consumer AI tools as you would a public forum unless your firm has a vetted, contractually protected enterprise deployment with a signed data-processing agreement and assurances that inputs are not used for training. Redact names, account numbers, dates of birth, and matter identifiers, or work only with hypothetical or already-public facts. When in doubt, ask your firm's general counsel or IT security team, and verify every citation, statute, and factual claim against the primary source — large language models can fabricate case names and holdings.

Each prompt is written as a reusable template. Copy one, paste it into your AI tool, and fill in the [bracketed] placeholders with **redacted or public** information. Set a role ("you are a litigation paralegal"), give it context, and always specify the output format — a table, a numbered list, a memo. The more constraints you add, the less the model improvises.

Work in small, verifiable chunks. Ask for a summary of one document at a time so you can check it against the source, rather than dumping a 200-page record and trusting a single pass. For background on why structured prompts beat one-liners, see what is prompt engineering and our prompt engineering cheat sheet.

These prompts compress long, redacted, or public documents into something you can scan and verify. Always cross-check page and paragraph references against the original.

**1. Neutral document summary** — "You are a litigation paralegal. Summarize the document below in plain English for an attorney who has not read it. Produce: (a) a 3-sentence overview, (b) a bulleted list of key facts with the page or paragraph number for each, and (c) a list of any defined terms. Do not infer facts that are not stated. Flag anything ambiguous. Document: [paste redacted/public text]."

**2. Issue spotter** — "Acting as a paralegal, read the [contract/pleading/policy] below and list every potential legal issue, obligation, deadline, or risk you can identify. For each, quote the exact clause it comes from and note the page/section. Do not provide legal conclusions — only flag items for attorney review. Text: [paste redacted/public text]."

**3. Clause comparison** — "Compare these two versions of the same [clause/section] and produce a table with columns: Topic, Version A language, Version B language, Substantive change (yes/no), Plain-English note. Highlight any change that shifts a deadline, liability, or obligation. Version A: [paste]. Version B: [paste]."

**4. Deposition/transcript digest** — "You are a paralegal preparing a deposition digest from the public/redacted transcript below. Output a table: Page:Line, Topic, Testimony summary (1 sentence), Follow-up flag. Keep summaries factual and neutral. Transcript excerpt: [paste]."

Use these for first drafts only. Every draft goes to your supervising attorney before it leaves the building.

**5. Routine client-status letter (draft)** — "Draft a professional, plain-English status update letter to a client about a [matter type] matter. Use a calm, reassuring but non-committal tone. Do not state legal conclusions or predict outcomes. Leave [DATE], [NEXT STEP], and [ATTORNEY NAME] as placeholders for me to fill. Keep it under 250 words."

**6. Internal case memo skeleton** — "You are a paralegal. Create the skeleton of an internal case memo with these sections: Question Presented, Short Answer (leave blank for attorney), Facts, Procedural Posture, Issues to Research, Open Tasks. Populate only the Facts and Procedural Posture sections from this redacted summary: [paste]. Mark every other section 'For attorney input.'"

**7. Meet-and-confer / scheduling email** — "Draft a brief, courteous email to opposing counsel proposing dates for a [meet-and-confer / deposition / call]. Offer three time windows: [WINDOW 1], [WINDOW 2], [WINDOW 3]. Professional and neutral tone, no substantive case discussion. Sign-off placeholder: [NAME, TITLE]."

Chronologies and fact tables are where AI saves the most time — but they are also where fabricated dates do the most damage. Verify every entry.

**8. Chronology builder** — "From the redacted facts below, build a chronology table with columns: Date, Event, Source (document name + page), Significance (1 line). Sort earliest to latest. If a date is approximate or missing, write 'UNDATED — verify' instead of guessing. Facts: [paste redacted text]."

**9. Document index / privilege log skeleton** — "Create a document-index table with columns: Doc No., Date, Type, Author/Recipient (redacted), Description, Reviewed (Y/N). Leave privilege determinations blank for attorney review. Populate from this list of public/redacted document descriptions: [paste]."

**10. Plain-English explainer for a client** — "Rewrite the following procedural step into plain English a non-lawyer client can understand, at roughly an 8th-grade reading level. Do not give advice or predict outcomes — only explain what the step is and what generally happens next. Add a closing line reminding the reader to direct questions to their attorney. Step: [paste]."

**Do not** paste privileged, confidential, or PII-laden material into a consumer chatbot. **Do not** ask the model to perform legal research and trust the citations — LLMs invent realistic-looking case names and holdings. If you use a prompt to find authority, treat every citation as a lead to verify in Westlaw, Lexis, or the primary source, never as a finished cite.

**Do not** let an AI draft go out without attorney review, and **do not** rely on AI for filing deadlines or limitations periods — compute and confirm those yourself against the rules. Avoid vague prompts ("summarize this case"); instead specify the format, the source references, and the no-inference rule. For more on keeping models from leaking or being manipulated, see our prompt injection defense checklist.

All three major assistants handle summarization and drafting well; the differences are durable ones around long-document handling, free access, and a built-in reasoning mode for careful comparison work. Pricing and exact limits change often — check the official pages linked in the table footnote before committing your firm to a plan.