Individual AI adoption among legal professionals more than doubled in a year to 69%, and paralegals were the top AI training priority at 39% — ahead of partners (30%) and associates (26%) — in the 8am 2026 Legal Industry Report of 1,300+ legal professionalsSource ↗
79% of legal professionals now use AI in their practice, up from 19% in 2023, according to Clio's Legal Trends Report dataSource ↗
The most common uses of general-purpose AI tools in legal work are drafting correspondence (58%), general research (58%), and summarizing documents (47%)Source ↗
A public database maintained by legal researcher Damien Charlotin tracks over 1,700 court decisions involving AI-hallucinated citations or fake authorities in filings, most of them from 2025 onwardSource ↗
analysisClaudeChatGPT

Deposition digests with page-line citations in an hour

Summarizing a 200-plus-page deposition transcript is classic paralegal work that eats a full day, and it usually lands right when trial prep is busiest. Long-context AI tools can now digest an entire transcript in one pass — but the digest is only useful if every point carries a page-line citation the attorney can check.

Prompt
You are a litigation paralegal's drafting assistant. I will paste a deposition transcript below. Create a two-part digest for attorney review.

Part 1 — Topic summary: organize the testimony under these headings: {{key_topics}}. Use 2-4 bullets per heading, and end every bullet with a page:line citation from the transcript.

Part 2 — Flags, each with a page:line citation:
- Testimony that conflicts with itself or with the case context below
- Admissions or damaging statements (quote the exact words)
- Questions the witness did not answer or was instructed not to answer

Case context (already cleared for this tool): {{case_context}}

Transcript: {{transcript_text}}

Rules:
- Use only what is in the transcript. Do not add background facts, characterize the witness's credibility, or draw legal conclusions — that is the attorney's job.
- If one of my topics is not covered in the testimony, write "Not addressed in this excerpt" instead of inferring.
- Never paraphrase an admission — quote it.
- Keep the full digest under 700 words.

Fill in your details and the prompt updates live — then copy.

writingClaudeChatGPT

First drafts of discovery requests ready for attorney markup

Drafting interrogatories and document requests from scratch is slow, and starting from a stale form means half the requests don't fit the case. Paralegals now prompt AI with the case facts to get a tailored first set, then refine it — vendors and practitioners report drafting time cut by half or more, with the attorney still reviewing and signing everything served.

Prompt
You are assisting a litigation paralegal preparing a first draft of discovery for attorney review. Draft {{discovery_type}} in a {{case_type}} pending in {{jurisdiction}}.

Facts and issues to cover (use only these): {{case_facts}}

Output format:
- Caption placeholder, then a short definitions and instructions section, then numbered requests.
- Group requests by issue, most important first.
- Make each request specific enough to survive an overbreadth objection — tie it to a party, time period, or event from my facts.
- After the requests, add a "For attorney review" list: requests that may exceed local numerical limits, potential privilege issues, and facts I did not provide that would strengthen the set.

Rules:
- Do not invent facts, dates, parties, or events not in my summary. Where a detail is needed, write [CONFIRM: detail].
- Do not cite rules or case law from memory — write [CITE: rule] wherever authority belongs so we insert the verified rule text.
- Put "DRAFT — FOR ATTORNEY REVIEW" at the top.
- Plain, precise language. No "any and all" boilerplate chains.

Fill in your details and the prompt updates live — then copy.

planningClaudeChatGPT

Medical chronologies that cite every source page

Building a medical chronology for a personal injury case traditionally takes 40-60 paralegal hours of reading, extracting, and formatting. AI can produce a first-pass chronology in a fraction of that time — but only a version where every row cites a source page is worth QC-ing, and only a firm-approved, HIPAA-appropriate tool should ever touch the records.

Prompt
You are helping a personal injury paralegal build a medical chronology. I will paste text extracted from medical records that our firm has approved for processing in this tool.

Create a chronology table with columns: Date of service | Provider/facility | Event type (visit, imaging, procedure, prescription) | Findings/diagnosis | Treatment | Source (document name and page).

Records text: {{records_text}}
Incident date: {{incident_date}}
Focus injuries: {{focus_injuries}}

Rules:
- One row per dated encounter, in chronological order. Every row must include a source page — no row without one.
- Use the records' own words for diagnoses and findings. Do not interpret, upgrade, or restate clinical language.
- If a date is illegible or missing, write [DATE UNCLEAR] and still include the row.
- Do not fill gaps. If there is a treatment gap longer than 30 days, list it in a "Gaps and follow-ups" section after the table instead of guessing what happened.
- After the table, also list: pre-existing conditions mentioned that predate the incident date, and records referenced but not provided (for example, a referral with no matching report).

Fill in your details and the prompt updates live — then copy.

communicationChatGPTClaudeCopilot

Client status updates the attorney approves in one pass

Drafting correspondence is the single most common AI use in legal work — 58% of professionals using general-purpose tools do it — and routine client status letters are the paralegal version. The draft has to walk a line: informative and plain-English, but with zero legal advice, predictions, or opinions, because those belong to the attorney.

Prompt
You draft client correspondence for a law firm. I am a paralegal preparing a status update for my supervising attorney to approve. Write a client status email based only on the case activity below.

Matter type: {{matter_type}}
Recent activity from the file: {{case_activity}}
Next expected steps and dates from our calendar: {{next_steps}}

Format: greeting, a one-paragraph "where things stand," a bulleted "what happened recently," a bulleted "what happens next" with dates, and a closing that invites questions to our office.

Rules:
- Plain English at an 8th-grade level. Define any legal term you must use, e.g., "deposition (sworn testimony given before trial)."
- Report only events and dates I provided. Do not predict outcomes, estimate value, or characterize the strength of the case — no "we expect to win" or "we are in a strong position."
- Do not give advice or recommendations. Where the client faces a decision, say the attorney will discuss options with them.
- End with my signature block: name, the title "Paralegal," and the line "Reviewed by [ATTORNEY NAME]."
- Under 300 words.

Fill in your details and the prompt updates live — then copy.

analysisClaudeChatGPT

First-pass research memos where every citation gets verified

General research is tied for the most common AI use in legal work, and it is also where AI has burned the profession worst: a public database tracks more than 1,700 court decisions involving hallucinated citations, with real sanctions. The workable pattern is using AI to frame the issue and generate leads, then pulling and reading every authority in Westlaw or Lexis before it goes anywhere near work product.

Prompt
You are a legal research assistant helping a paralegal prepare a first-pass research memo for a supervising attorney.

Research question: {{research_question}}
Jurisdiction: {{jurisdiction}}
What we already know from the file: {{known_facts}}

Produce:
1. The issue restated in one sentence.
2. Framework: the elements, tests, or factors courts generally apply to this issue, in plain language.
3. Research leads: statutes, rules, and cases likely to matter. Tag every single citation with [UNVERIFIED — pull and read before use]. Never present a citation as confirmed.
4. Search strategy: 5-8 targeted Westlaw/Lexis queries with terms and connectors.
5. Open questions the attorney will likely ask next.

Rules:
- You may describe general legal principles, but tag any point where the law varies by state or may have changed recently with [CHECK CURRENT LAW].
- Never fabricate a quotation from a case or statute. If you cannot recall exact language, say so plainly.
- This memo is an internal starting point, not authority. End with the line: "All citations unverified — verify in Westlaw/Lexis before any use in work product."

Fill in your details and the prompt updates live — then copy.

automationChatGPTClaudeCopilot

Intake notes turned into structured matter summaries

Raw intake-call notes pile up fast, and turning them into something an attorney can act on — parties, dates, documents, conflicts list — is repetitive formatting work. AI structures the notes in seconds, and a well-built prompt forces out the two things that matter most: the missing questions and the dates that could start a limitations clock.

Prompt
You are helping a paralegal turn raw intake notes into a structured new-matter summary for attorney review. Practice area: {{practice_area}}.

Raw notes from the intake call (names replaced with placeholders): {{intake_notes}}

Output, in this order:
1. Parties: each placeholder name and the relationship between them.
2. Key dates: every date mentioned, labeled with what happened. Flag the earliest date that could start a limitations period as "POSSIBLE SOL TRIGGER — attorney to confirm."
3. Claim summary: 3-5 sentences of what the potential client says happened, in neutral language.
4. Documents mentioned: everything the caller referenced, as a collection checklist.
5. Missing information: every question a {{practice_area}} attorney would want answered that the notes do not cover, phrased as follow-up call questions.
6. Conflicts-check list: every person and entity named.

Rules:
- Use only what is in the notes. Do not infer facts, fault, or damages amounts.
- Do not assess whether the claim has merit and do not calculate any deadline — label date math for the attorney.
- Under 400 words.

Fill in your details and the prompt updates live — then copy.

Common questions from paralegals

Is it against the rules for a paralegal to use ChatGPT or Claude on client work?

Not inherently — ABA Formal Opinion 512 permits generative AI with safeguards, and paralegals are the profession's top AI training priority. The conditions are firm approval of the specific tool, no confidential client information in consumer accounts that train on inputs, and attorney supervision of anything that becomes work product. Check your firm's AI policy before your first prompt, not after.

Can using AI get me into unauthorized-practice-of-law trouble?

The tool changes nothing about UPL. A paralegal still cannot give legal advice, predict outcomes to a client, decide strategy, or sign and serve documents — whether the words came from your head or a chatbot. AI drafts are safe when everything flows through the supervising attorney for review and approval, exactly as NALA Canons 1-3 require.

How do I make sure an AI citation never ends up in a filing?

Treat every AI-supplied citation as fiction until proven otherwise — pull each authority in Westlaw or Lexis, read it, and confirm it says what the AI claims. Over 1,700 court decisions now involve hallucinated citations, and courts have sanctioned filers tens of thousands of dollars. A paralegal who runs a rigorous cite-verification pass is protecting the attorney's license.

Will AI replace paralegals?

The evidence points to changed work, not eliminated jobs. In the 8am 2026 survey, 39% expected fewer support roles but 22% expected more, and paralegals were the top AI training priority — firms are investing in them, not phasing them out. The paralegals gaining ground are the ones who manage AI workflows, QC the output, and handle the judgment calls AI cannot.

Related professions