Can AI Draft Discovery Requests?

AI can produce interrogatories, requests for production, requests for admission, and draft responses or objections in seconds. The harder question is whether that output is tied to the operative pleadings, the client’s documents, the governing rules, and prior work that survived the challenge.

Problems arise when polished output is mistaken for usable discovery. A general-purpose tool can return a cleanly numbered set in a minute, but fluent wording does not show that the questions fit the matter.

A weak discovery request is worse than a slow one. It signals the case theory to the other side, invites motion practice under the rules that govern discovery, and can give away the ground it was meant to protect. Speed was the easy part all along. The risk sits in the drafting choices a fast tool makes without knowing the case. In this article you will learn what grounds a discovery request, the certification rule that makes a lawyer's signature mandatory, and how litigation teams draft this way on real matters.

Grounded Drafting is the Real Test

Producing text and producing a usable request are different tasks. Any capable model can generate 40 numbered interrogatories from a one-line prompt. Whether those interrogatories help the case depends on something the prompt usually leaves out: the matter itself. The real test of generative AI for litigation is whether each request traces to the case in front of you and whether a lawyer can verify it before it goes out.

Each request instrument does a different job. Interrogatories put questions to the other side, requests for production call for documents, and requests for admission ask a party to concede specific facts. The responding side answers and objects to those same requests. Both directions carry the same requirement. A qualified lawyer must review and sign every AI-drafted request before it is served.

That requirement is easy to lose sight of when the output looks finished. AI for legal drafting produces clean formatting and confident phrasing that read as quality, and for litigators evaluating AI for their practice, the polish is exactly what hides the gaps. The sections below break the grounding into four sources, put a number on what ungrounded requests cost, and connect the work to the rule that keeps the signature in human hands.

The Hidden Cost of Overbroad Discovery Requests

Every ungrounded request can create downstream costs that appear weeks after the initial time savings. Those costs include additional meet-and-confer rounds, motions to compel or for protective orders, narrowed or stricken requests, missed production deadlines, and unnecessary disclosure of case strategy.

Each downstream cost traces to a drafting failure. An interrogatory that covers every year and every custodian invites a proportionality objection because discovery must fit the needs of the case. A set that exceeds the numerical limit may be cut down, often after a dispute over how the requests should be counted. A request that extends beyond the pleadings can reveal strategy without producing useful information.

Consider an interrogatory asking the opposing party to identify every communication about a product over a 10-year period when the complaint covers only 18 months. Opposing counsel objects on proportionality grounds, the parties brief the dispute, and the court narrows the request to the relevant period. The associate then spends a week defending language that a tool generated in a second, and a meet-and-confer that should have taken one call takes three.

None of those costs appears in the minute it took to generate the text. They appear in associate hours, motion practice, partner time, and strategic information disclosed without a corresponding benefit. Drafting time is only one part of the total cost. A tool that saves a minute but creates a week of motion practice has cost the team far more than it saved.

AI does not eliminate those consequences. They decline only when the work is tied to the matter and verified before service.

The Four Anchors of a Grounded Discovery Request

A grounded discovery request answers to four sources. If one is missing, the request starts to resemble boilerplate and becomes more vulnerable to objection. When all four are present, the request is more likely to serve its intended purpose.

1. The operative pleadings and the claims at issue

Start with a specific allegation, claim, or affirmative defense in the live pleadings. That connection helps establish relevance: a breach claim involving a single supply contract should produce requests about that contract and its performance, not the client’s unrelated business.

2. The client's own facts and documents

The client’s records, contracts, and correspondence supply the custodians, dates, and document types that make a request specific. Naming the three custodians who ran the deal is more likely to reach relevant material than asking for all documents relating to the subject matter.

3. The governing rules and proportionality

Next, test the request against Rule 26(b)(1), the 25-interrogatory limit under Rule 33, and any jurisdiction-specific caps or local rules. A set built to the count and scaled to the amount in controversy is more likely to withstand proportionality and numerosity objections than one that ignores those limits.

4. The organization's prior requests that survived challenge

Prior requests that survived challenge provide another source of tested language. Reusing the phrasing and structure of proven interrogatories and document requests gives the team a stronger starting point than drafting from a blank page.

This is the point where the source material matters more than the wording of any prompt, and it is where a legal AI platform built for grounding shows its value against a general-purpose tool. Harvey grounds discovery drafting in the client's own document set through Vault, in the organization's prior requests and standards through Harvey Playbooks and Knowledge, and in the governing rules the drafter names. In practice, that means a litigator drafts from the facts in the client's own documents, so each request traces to the record. That source-level grounding gives the reviewing lawyer a clear record for verification.

Rule 26(g) Certification for AI-Drafted Discovery Requests

Every discovery request in federal court carries a signature, and that signature means something specific. Under Rule 26(g) of the Federal Rules of Civil Procedure, an attorney who signs a discovery request makes a certification. After a reasonable inquiry, the signer certifies that the request is consistent with the rules, is not interposed for an improper purpose, and is neither unreasonable nor unduly burdensome given the needs of the case. Rule 26(g) is the discovery counterpart to Rule 11 and mandates sanctions when the certification is unfounded.

Rule 26(g)’s key requirement is reasonable inquiry. A litigator can make that certification when each request can be traced to the record, and the inquiry becomes more efficient when the draft points to the relevant pleading, document, and rule. It becomes more difficult when the tool never had access to the file.

An AI draft is therefore a starting point, not the final word. The lawyer must verify it against the sources, confirm that it fits the rules and the case, and sign it. Responsibility remains with the attorney because Rule 26(g) places the certification — and any resulting sanctions — on the signer.

Grounding also keeps that verification quick enough to do on every request. Harvey ties each drafted request to the source document the reviewing attorney can open and check, which is what the reasonable-inquiry standard asks for. Research supports the point. According to the RSGI report, The Accelerating Impact of Legal AI, grounding AI in an organization's own documents and matters reduces the risk of unsupported output and produces answers a lawyer can trust.

The professional-responsibility rules point the same way. The American Bar Association's Model Rule 1.1 Comment 8 ties competent representation to an understanding of relevant technology, which now includes the AI that drafts a request. Competence here means knowing what the draft contains. The duty to supervise the work applies to an AI draft the way it applies to an associate's, and the reviewing lawyer stands behind every request served. The lawyer who signs answers for the result because Rule 26(g) places that responsibility on the signing attorney.

Drafting Discovery Requests and Responses With AI

On a live matter, the work starts with the record. A litigation team begins from the operative pleadings and the client's documents, identifies the claims and defenses that discovery needs to reach, and drafts targeted interrogatories and requests for production against them. The team checks the draft against proportionality and the numerical limits, tightens anything overbroad, and routes the set to a litigator for review and signature.

The loop stays tight because each step feeds the next. When the first draft is tied to the record and scaled to the rules, it requires less rewriting, less negotiation, and less time to verify. Legal workflow automation earns its place when it tightens this loop, from the first grounded draft to the litigator's signature. The associate spends the saved hours on strategy and the substance of the case, and the partner signs a set built to survive the challenge.

Discovery runs both ways, and the responding side takes as much time as the propounding side. When the other side serves requests, AI can draft the responses and objections, grounded in the client's documents and the claims actually in play, and the same review and signature apply. Drafting these requests and responses is one part of AI for legal discovery. Reviewing incoming productions for relevance and privilege is the other, a discipline of its own that lives on the review side of discovery.

In practice, this runs where litigators already work. Harvey operates inside Microsoft Word and connects to the client's documents and the matter record. Harvey Agents plan a drafting task, pull the relevant facts, and return a cited, review-ready draft the litigator can check line by line. Lynn Pinker Hurst & Schwegmann, a Band 1 Chambers-ranked litigation boutique, uses Harvey across its litigation work and reports saving more than eight hours per lawyer each week.

None of this touches collection, hosting, or the mechanics of a production. Litigation management software organizes the matter file, calendar, and docket, while grounded drafting is a separate function. The team's job here is drafting the requests and responses and grounding them in the record, and the litigator's judgment stays in the loop from the first draft to the signature.

The Standard for AI-Drafted Discovery Requests

Producing discovery text has become easy. What distinguishes a request worth serving from boilerplate is a clear connection to the matter and attorney verification. The Four Anchors identify the sources that provide that connection, the downstream-cost analysis shows what happens when they are missing, and Rule 26(g) keeps responsibility with the signing lawyer. Together, those elements define the standard for an AI-drafted discovery request.

This is the work Harvey is built for. Harvey drafts discovery requests grounded in the matter's own record, pulling facts from the client's documents in Vault and returning each request with citations a litigator can open and verify. The result is a faster first draft and a stronger request, grounded in the record a litigator can verify.

See what grounded discovery drafting looks like on your own matters and book a demo with our team.

Frequently Asked Questions

Can AI draft interrogatories and requests for production?

Yes. AI drafts interrogatories under Rule 33 and requests for production under Rule 34, and it can draft requests for admission under Rule 36 as well. Its usefulness depends on grounding each request in the pleadings, the client’s documents, and the governing rules, followed by attorney review under Rule 26(g).

Are AI-drafted discovery requests admissible or proper to serve?

Discovery requests are not evidence, so admissibility is the wrong test. A request is proper to serve when it fits the Federal Rules of Civil Procedure, respects proportionality and the numerical limits, and carries an attorney's Rule 26(g) signature after a reasonable inquiry. An AI draft that a qualified lawyer verifies and signs meets that standard.

Does a lawyer have to review AI-drafted discovery requests?

Yes. Rule 26(g) requires an attorney to conduct a reasonable inquiry and certify that each discovery request is consistent with the rules and not unduly burdensome. Because that certification carries potential sanctions, the lawyer — not the AI tool — must verify and sign the request before service.

Can AI draft responses and objections to discovery requests?

Yes. AI can draft answers and objections to interrogatories, requests for production, and requests for admission, grounded in the client's documents and the claims in play. Reviewing incoming productions for relevance and privilege is a separate task on the review side of discovery. As with any draft, a qualified lawyer verifies and signs it before service.