INDUSTRY GUIDES

AI for law firms: intake, discovery responses, and drafting with attorney review

AI for law firms works best on the high-volume work around legal judgment: structured intake and conflict-check prep, first drafts of discovery responses and routine documents, billing narratives, and search across your own matter files. An attorney reviews everything that leaves the firm, and ABA Formal Opinion 512 sets the ethical frame for confidentiality, supervision, and fees.

By Buzzed Technologies ยท Published ยท Updated

What matters

The short version.

  1. 01

    Start with intake and conflict-check prep: a narrow, auditable workflow that collects only what the conflict check needs before anyone hears the full story.

  2. 02

    Discovery responses, billing narratives, and routine drafts are strong targets because an attorney already reviews and signs them. The tool drafts; the lawyer certifies.

  3. 03

    ABA Formal Opinion 512 says informed consent is needed before client information goes into a self-learning AI tool, and boilerplate engagement-letter language is not enough.

Law firm workflows: what AI drafts and who signs off

A planning aid, not a ranking. The right level of review depends on the tool, the task, and your jurisdiction's rules.

Law firm workflows: what AI drafts and who signs off
WorkflowWhat AI doesWho reviewsMain risk to control
Intake and conflictsCollects parties, matter type, jurisdiction, and urgency; runs preliminary name checks; books consultationsIntake staff, then the attorney who clears conflictsCollecting detailed facts before conflicts clear
Document reviewSummarizes depositions and contracts, extracts clauses, flags unusual termsAssociate or paralegal, after sample testingMissed or mischaracterized terms
Discovery responsesDrafts responses and objections from your library and the recordSigning attorney; client verifies interrogatory answersBoilerplate objections and unsupported facts
Client status updatesDrafts plain-language notes when the matter changesResponsible attorney or paralegalSending advice or wrong dates without approval
Billing narrativesDrafts entries from calendar, documents, and email with task codesEach timekeeperBilling for time the tool saved
Knowledge searchFinds and summarizes prior work product with source linksThe lawyer using itCrossing ethical walls; treating old memos as current law

01

Where AI fits in a law firm, and where it doesn't

AI for law firms is most useful on the volume work that surrounds legal judgment: collecting intake facts, preparing conflict checks, drafting discovery responses and routine documents, summarizing records, writing billing narratives, and finding what the firm already knows. It should not give legal advice, sign anything, or decide strategy.

The test for a good first project is simple. The work repeats every week, the inputs are documents or structured facts, and an attorney or paralegal already reviews the output before it goes anywhere. When all three are true, you can measure the time saved and catch errors without changing who is accountable.

Poor first projects are the ones where a wrong answer reaches a client, opposing counsel, or a court with no person in between: an after-hours chatbot that answers 'do I have a case?', or a research tool whose citations nobody checks.

02

Intake and conflict-check prep without Rule 1.18 trouble

Prospective clients inquire at all hours by web form, chat, phone, and email. An intake workflow can collect case type, jurisdiction, key dates, opposing parties, and urgency, run those names as preliminary conflict-check inputs against your matter database, and schedule a consultation with the right attorney, so no inquiry waits overnight.

The trap is collecting too much. Model Rule 1.18 protects information from prospective clients even when you never take the matter, and detailed facts from someone adverse to an existing client can create a disqualification problem. Design intake to gather what the conflict check needs first (every party, related entities, the matter type) and hold the full narrative until conflicts clear.

Keep people on two decisions: whether a potential conflict is real, and whether to accept the matter. The system can flag near-matches on names, maiden names, and corporate affiliates; an attorney decides. Send declination and non-engagement letters from approved templates so the record is clean.

Intake automation in AI for law firms: an inquiry at any hour gets case type, jurisdiction, key facts and urgency collected, preliminary conflict-check inputs run, and a consultation scheduled with the right attorney.

03

Document review: AI speeds the reading, attorneys keep the judgment

AI does not replace legal judgment, but it accelerates the reading step. It can summarize depositions by witness and topic, extract key clauses from contracts, flag unusual terms against your standard positions, and organize discovery materials into issue folders for attorney review.

Opinion 512 gives a practical standard for how much checking is enough: it depends on the tool and the task. Its example is a lawyer who tested a tool's contract summaries against a manually reviewed subset and found them accurate; that lawyer would not necessarily need to re-read every document. Build that test into your rollout. Pick a sample, have a paralegal do it by hand, compare results, and write down the error types you found.

For large productions, technology-assisted review in platforms such as Relativity or Everlaw is a mature, separate category. Generative tools are better at the next step: explaining what a document set says, building chronologies, and drafting deposition outlines from the key documents.

  • Depositions
  • Contracts
  • Discovery
  1. AI first passReading step
  2. Key clausesExtracted
  3. Unusual termsFlagged
  4. Attorney reviewLegal judgment
Document review acceleration: AI speeds the reading step by summarizing depositions, contracts and discovery, extracting key clauses and flagging unusual terms, while legal judgment stays with the attorney.

04

Can AI draft discovery responses for a law firm?

Yes, as a first draft that the attorney reviews and certifies. Discovery responses are one of the best drafting targets in a litigation practice. Requests arrive as numbered lists, most objections come from a library the firm already uses, and the facts sit in the client questionnaire, the pleadings, and the documents. A drafting workflow can split the requests, match each one to standard objections and to responses from similar matters, pull supporting facts with citations to source documents, and assemble a draft in your format.

What it must not do is replace the reasonable inquiry. Under Federal Rule 26(g), the attorney's signature certifies the responses were formed after a reasonable inquiry, and interrogatory answers are made under oath by the client. Courts criticize boilerplate general objections, and Rule 34 requires stating whether responsive materials are being withheld on the basis of an objection. State rules differ, so the objection library should be organized by jurisdiction.

  1. Parse the set: split each interrogatory, request for production, and request for admission, and put the response deadline on the calendar.
  2. Match each request to your objection library and to responses from comparable matters in the same jurisdiction.
  3. Pull facts from the client questionnaire, pleadings, and produced documents, citing the source for every factual statement.
  4. Flag requests that need a client answer, a privilege call, or attorney judgment instead of guessing.
  5. Route the draft to the responsible attorney, then to the client to verify interrogatory answers.
  6. Save the signed final back to the library so the next draft starts from approved language.

05

Client updates and after-hours inquiries

Clients call to ask what is happening because nobody told them. A status-update workflow can draft a plain-language note when something changes on a matter: a filing, a hearing date, an offer, a document request. Each note says what happened, what comes next, and what the firm needs from the client. The responsible attorney or paralegal approves it before it sends, which also supports the communication duty under Rule 1.4.

After hours, an assistant can answer logistics (office hours, how to send documents, consultation scheduling), collect intake facts, and escalate urgent situations such as an arrest, a served complaint, or a protective order. It should say plainly that it is not a lawyer, that no attorney-client relationship exists yet, and that it cannot give legal advice. Our Angelone Homes case study is a home builder, not a firm, but it shows the same front-desk pattern at volume: an assistant named Ava handles 200+ messages a week and hands off to people.

06

Billing narratives and what Opinion 512 says about fees

Time entries written on Friday from memory are vague, and clients cut vague entries. AI can draft narratives from the record of the work: calendar events, documents edited, emails sent, and docket activity, mapped to UTBMS task codes for LEDES invoices and checked against each client's outside counsel guidelines for block billing, vague descriptions, and staffing limits. Each timekeeper approves their own entries.

Opinion 512 is direct about fees. Hourly billers must bill actual time, which includes time spent prompting a tool and reviewing its draft, but not the hours the tool saved. Charging the same flat fee for work that now takes far less time may be unreasonable under Rule 1.5. Lawyers may not bill clients for learning a tool they will use regularly. An AI feature built into everyday software is usually overhead, while a per-use third-party charge incurred for a specific client's matter can ordinarily be billed as an expense, with the basis explained to the client in advance.

07

Knowledge search across your own work product

Most firms have years of briefs, research memos, deal documents, and clause variations in a document management system such as NetDocuments or iManage, findable only by people who remember the matter name. A retrieval system (often called RAG) indexes that work product and answers questions with links to the source documents: how did we handle this indemnity in prior deals, what did our memo conclude about that statute, which motions have we filed before this judge.

Two requirements are specific to law. Search must inherit ethical walls and matter-level permissions, because Opinion 512 flags the risk of client information reaching screened people inside the firm. And searching your own documents is not legal research: authority still gets checked in Westlaw or Lexis, and every citation in a filing gets read by a person. Courts have sanctioned lawyers for filing AI-invented cases, and some judges require disclosure of AI use in standing orders.

08

What ABA Formal Opinion 512 asks of your firm

The ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on July 29, 2024. It applies existing Model Rules to generative AI rather than creating new ones, and it covers competence, confidentiality, client communication, candor toward the tribunal, supervision, and fees.

ABA opinions are not binding. Your state's rules, ethics opinions, and court rules control, so use the list below as a checklist to review with ethics counsel or your firm's general counsel, not as a legal conclusion.

  • Competence (Rule 1.1): understand what your tools can and cannot do, and review output to a level that fits the tool and the task.
  • Confidentiality (Rule 1.6): get informed client consent before putting information relating to the representation into a self-learning tool. Boilerplate engagement-letter language is not informed consent.
  • Communication (Rule 1.4): disclose AI use when the client asks, when outside counsel guidelines require it, when it affects the fee, or when output will influence a significant decision.
  • Candor and meritorious claims (Rules 3.1, 3.3, 8.4(c)): verify every assertion and citation before it reaches a tribunal.
  • Supervision (Rules 5.1 and 5.3): adopt a firm AI policy, train lawyers and staff, and vet vendors' security, confidentiality, and breach-notice terms.
  • Fees (Rule 1.5): bill actual time, explain AI charges in advance, and do not bill for learning the tool.

09

Buy, configure, or build: choosing AI tools for your firm

Start with what you already pay for. Practice management platforms such as Clio, MyCase, Filevine, and Smokeball offer AI features for intake, drafting, and summaries, and legal-specific assistants such as Harvey, CoCounsel, and Lexis+ AI cover research and drafting. If a feature inside your current system handles the workflow, that is usually the cheapest path.

A custom build makes sense when the workflow crosses systems: intake forms that feed a conflicts database, a calendar, and an engagement letter, or discovery drafting that needs your objection library and matter documents together. Most focused internal tools land between $15K and $75K for a first production version, and a focused first version can ship in about five weeks with weekly demos. Whatever you choose, ask every vendor the same questions before client data goes in.

  • Is our data used to train models, and can that be excluded by contract?
  • Where is data stored, for how long, and which subprocessors touch it?
  • Does the tool respect matter-level permissions and ethical walls?
  • Is there an audit log of who asked what and what the system returned?
  • Will you notify us of a breach or of legal process seeking our data?

Related services

Related use cases

Proof in the wild

Continue reading

Questions

Yes, if it produces a first draft and nothing more. AI can split interrogatories, requests for production, and requests for admission, match them to your objection library, and pull facts from the client questionnaire and documents with citations. The signing attorney still owns the reasonable inquiry that Federal Rule 26(g) certifies, the client verifies interrogatory answers under oath, and objections should be specific to each request rather than copied boilerplate.

It depends on the tool and the information. ABA Formal Opinion 512 says informed consent is required before inputting information relating to the representation into a self-learning generative AI tool, and that boilerplate engagement-letter language is not enough. Using a tool for idea generation without client information does not require consent. State rules and opinions vary, so confirm the standard with ethics counsel in your jurisdiction.

Not for hours you did not work. Opinion 512 says hourly billers must bill actual time, including time spent prompting and reviewing the draft. A flat fee may become unreasonable if a tool makes the work much faster, and lawyers cannot bill for learning tools they use regularly. Per-use third-party charges for a client's matter can ordinarily be passed through as expenses when the basis is explained in advance.

It should not be. Configure it to collect facts, answer logistics, and schedule consultations, and to decline questions like 'do I have a case?' or 'what should I do?'. It should state that it is not a lawyer and that no attorney-client relationship exists yet. Keep the transcripts secure, because the firm owes duties to prospective clients under Rule 1.18 even when it declines the matter.

Most start with AI features inside their practice management system, such as Clio, MyCase, Filevine, or Smokeball, plus a legal research assistant such as CoCounsel or Lexis+ AI. Firms with a specific bottleneck, like high-volume intake or repetitive discovery, sometimes add a custom workflow that connects those systems. Choose based on the workflow you need fixed, not the length of the feature list.

Treat every citation as unverified until a person has read it in Westlaw, Lexis, or the official reporter. Use retrieval tools that link answers to source documents, add a cite-check step before any filing, and note in the file which drafts were AI-assisted. Check your courts' local rules and standing orders as well, since some judges require disclosure of AI use in filings.

They can if the tool indexes every document for every user. Opinion 512 specifically flags the risk of client information reaching screened lawyers inside the firm. Require the system to inherit matter-level permissions from your document management system, test it with a screened user account before rollout, and keep an audit log of searches and answers.

Sources and further reading

Next step

Pick one law firm workflow and make it measurable

Bring a sample intake form, a discovery set, or a month of billing entries. We'll map the review points and systems involved, then scope a first version your ethics counsel can check.

AI for Law Firms: Intake, Discovery, Drafting & Ethics