How to Use AI to Draft an RFE Response

Learning how to use AI to draft an RFE response is the single biggest workflow upgrade immigration attorneys have made in the past five years. Done right, it turns a four-hour drafting session into thirty minutes—without giving up legal accuracy, USCIS-compliant formatting, or your firm's voice. Done wrong, it produces generic boilerplate that weakens your response and your credibility.

This guide walks you through the full workflow we use at Drafty AI: how to deconstruct an RFE, prompt an AI tool with the right context, review and edit the output, and sign off with confidence. It is written for practicing immigration attorneys, not consumers. By the end, you will have a repeatable process you can apply to any RFE type—H-1B specialty occupation, I-140 ability to pay, I-130 bona fide marriage, asylum credibility, and everything in between.

What Is an RFE — and Why the 87-Day Clock Matters

A Request for Evidence (RFE) is USCIS's way of saying "we cannot approve your petition on the current record." The officer has identified one or more deficiencies in the evidence and is giving the petitioner a chance to cure them before issuing a denial. The stakes are high: a weak RFE response typically results in a denial, not another RFE.

Standard RFE response deadlines (87 days and variations)

Most RFEs carry an 87-day response deadline from the date of issuance, measured from the date printed on the notice—not the date you received it. A handful of case types and specific situations use shorter windows (30 or 33 days for some notices of intent to deny, 12 weeks for certain asylum matters), so always read the deadline line on page one of the notice before you do anything else.

The 87-day window sounds generous. It is not. Between gathering fresh evidence from the client, coordinating with experts, drafting the legal argument, assembling exhibits, and leaving room for mail delivery back to the service center, most attorneys are working against a real 60-day timeline. This is the central reason AI RFE response drafting has become standard practice: the time saved on the first draft is time you get back for client calls, expert coordination, and evidence sharpening.

What USCIS is actually asking for when they issue an RFE

Every RFE does three things in its body, and reading them in this order is the fastest way to orient yourself:

1. Cites the legal standard. The officer reproduces the statute or regulation that governs the benefit sought (for example, 8 C.F.R. § 214.2(h)(4)(iii)(A) for H-1B specialty occupation).
2. Summarizes the evidence already submitted. The officer lists what you filed and, often, why it was insufficient.
3. Requests specific additional evidence. This is the operative section. Every response must address each bullet point here, in order, on the record.

If your response does not map to the request list point-by-point, the adjudicator has to hunt for your answers. Adjudicators under volume pressure do not hunt. They deny.

Why AI Is the Right Tool for RFE Response Drafting

The real time-cost of manual RFE drafting

A typical H-1B specialty occupation RFE response runs 15 to 40 pages, including the legal argument brief, supporting letters, and exhibit list. From scratch, a senior associate can spend three to five hours on the first draft alone, with another two to four hours in edit cycles before it leaves the office. Across a firm handling even a modest RFE volume, this is hundreds of attorney-hours per quarter spent on documents that follow highly repeatable structural patterns.

Where AI excels: pattern recognition, evidence matching, formatting

AI models trained on (or appropriately prompted with) immigration-law context are genuinely good at four tasks that dominate the first draft:

• Recognizing the RFE issue category from the language USCIS uses. "A specialty occupation requires" signals one template; "the petitioner has failed to establish ability to pay" signals another.
• Mapping evidence to arguments. Given a list of exhibits, AI can draft the argument section that references each exhibit in the right order with the right legal framing.
• Enforcing USCIS formatting conventions. Headings, exhibit tabs, response-by-issue structure, and citation format follow patterns AI can produce reliably once shown.
• Producing a first draft faster than a human can type. On the order of thirty seconds for a structured 20-page response once the prompt is set.

What AI still cannot do

AI does not practice law, and an RFE response is legal practice. The attorney retains every judgment call that matters:

• Strategy. Whether to argue the regulatory standard, the factual record, or both. Whether to concede a weak point to strengthen the stronger one.
• Credibility assessment. Whether a client declaration supports or undermines the petition.
• Expert selection and coordination. AI can draft the expert letter request; it cannot decide which expert to engage.
• Final sign-off. Every word that leaves the office is your signature, your bar number, and your professional responsibility.

For the ethical framing of where AI fits in a law practice, the ABA's guidance on AI in legal practice and the Florida Bar's AI ethical guidelines are the clearest starting points.

How to Use AI to Draft an RFE Response: The 6-Step Workflow

The workflow below is the one we teach new hires at Drafty AI and the one our customers run on thousands of RFEs per month. It is designed to produce a response an attorney can sign in one pass after AI-assisted first drafting.

Step 1 — Read and deconstruct the RFE

Open the notice. On a single page (or in a single memo field in your case management system), write down three things: the deadline, the legal standard(s) cited, and the itemized evidence request list. This is the AI's input brief.

Identifying USCIS issue categories

Every RFE maps to one or more of a finite set of issue categories. For H-1B: specialty occupation, employer-employee relationship, beneficiary qualifications, LCA compliance. For I-140: ability to pay, extraordinary ability criteria, specific skill. For I-130: bona fide relationship, valid marriage, supporting documentation. For asylum: credibility, nexus, well-founded fear. Naming the category in your own words before you prompt the AI sharpens the output dramatically.

Step 2 — Organize source evidence by issue

For each bullet point in the RFE's request list, collect the exhibits that will support your response. Label them provisionally (Exhibit A, B, C…). If an exhibit does not yet exist—a new expert letter, a revised employment contract, an updated financial statement—note the gap and assign it to a team member with a due date well inside the 87-day window. AI cannot draft around evidence that has not been gathered.

Step 3 — Prompt the AI with case facts, issue, and evidence

The quality of an AI RFE response is almost entirely a function of prompt quality. A thin prompt ("write an RFE response for an H-1B specialty occupation case") produces generic boilerplate. A structured prompt produces a document you can edit in one pass.

A reusable prompt template for AI RFE response drafting

Use the structure below. Fill in the bracketed fields for each case. For a deeper treatment of prompt engineering for legal work, see our free AI prompting guide for immigration lawyers.

Role: You are an experienced immigration attorney drafting an RFE response for filing with USCIS.

Case context: [Petition type, beneficiary, petitioner, key dates, prior filings.]

RFE issue(s): [Copy the itemized USCIS request list verbatim.]

Legal standard cited: [Quote the statute or regulation from the notice.]

Available evidence: [List exhibits with a one-line description of each.]

Firm style: [Formal, plain-English, third-person references to the beneficiary, section headings in title case.]

Task: Draft a complete RFE response brief. Open with a one-paragraph summary. Organize the argument section by issue, addressing each RFE bullet in order. Reference each exhibit where it supports the argument. Close with a request for approval. Do not invent evidence or cite sources not in the provided list.

Step 4 — Review AI output against USCIS requirements

The first read-through is not an editing pass—it is a compliance pass. You are checking four things before you touch prose:

• Does the response address every bullet in the USCIS request list, in order?
• Does it cite the exact legal standard USCIS cited?
• Are exhibit references correct and complete?
• Is the formatting (headings, tabs, response-by-issue structure) USCIS-conforming?

If any answer is no, fix the structural gap first. Prose quality is meaningless on a structurally non-responsive brief.

Step 5 — Edit for tone, legal accuracy, and firm style

Now read it as a lawyer. AI draft prose tends toward three specific failure modes: it hedges when it should assert, it asserts when it should hedge, and it uses generic legal filler ("as a matter of law," "it is well-established that") where a concrete citation would do more work. Replace hedging with confident language where the record supports it. Replace overreach with attributable statements where it does not. Delete filler and replace with citation or specific evidence references. This step typically takes 15–45 minutes depending on response length and is the step that converts an AI draft into an attorney work product.

Step 6 — Final attorney proof and sign-off

Read it twice, out loud if possible. Confirm every exhibit is attached. Confirm the signature block, bar number, and G-28 are correct. Confirm the cover letter matches the brief. Your signature is the last thing the file gets before it leaves the office, and nothing else in the pipeline replaces your review.

Worked Example: AI-Drafted RFE Response for an H-1B Specialty Occupation Challenge

To make the workflow concrete, here is how it plays out on one of the most common RFE types immigration attorneys see.

The RFE excerpt (what USCIS asked)

The officer wrote: "The petitioner has not established that the position offered to the beneficiary meets the definition of a specialty occupation under 8 C.F.R. § 214.2(h)(4)(iii)(A). The petitioner is therefore requested to submit evidence that the position normally requires a minimum of a bachelor's degree or its equivalent in a specific specialty."

The prompt we sent

The prompt followed the Step 3 template. Case context identified the petitioner (a 45-person biotech firm), the beneficiary (an M.S. in biomedical engineering), and the position (Senior Research Associate). Available evidence included the job description, the industry O*NET entry, three letters from comparable employers in the industry confirming similar degree requirements, and an expert opinion letter from a professor in the field.

The AI draft (first-pass output)

In about thirty seconds, the AI produced a 12-page brief structured as: (1) one-paragraph summary asserting the position meets all four criteria under 8 C.F.R. § 214.2(h)(4)(iii)(A); (2) a legal standard section; (3) four argument sections, one per regulatory criterion, each referencing the relevant exhibits; (4) a request for approval. Every exhibit in the input list was referenced. The regulatory citation was correct. No fabricated sources appeared.

Attorney edits and the final response

The attorney made three categories of edits. First, the industry-standard argument was strengthened by quoting directly from the employer letters instead of summarizing them. Second, the degree-equivalency argument was trimmed because the beneficiary's M.S. made extensive equivalency analysis unnecessary. Third, the closing request was reworded to match firm style. Total edit time: 28 minutes. Total drafting time including the first-draft generation: 35 minutes. For comparison, the same attorney estimated the manual version at 4 to 5 hours.

Common Mistakes When Drafting RFE Responses with AI

Over-relying on generic template phrasing

AI defaults to safe, generic legal language. On an RFE response, generic language is a liability: the officer has already rejected the generic record. Every paragraph should reference specific evidence from this case. If a paragraph could apply to any H-1B RFE, cut it or rewrite it against your exhibits.

Missing USCIS-specific formatting and exhibit conventions

USCIS adjudicators expect a specific structure: cover letter, brief organized by issue, exhibit tabs, exhibit list. AI tools not trained on immigration filings will default to generic legal-memo formatting. Always review against your firm's template and USCIS's published guidance before the response leaves the office.

Skipping citation and source verification

General-purpose AI tools occasionally produce citations to cases or regulations that do not exist, or that exist but say something different from what the draft claims. Every citation in an AI-drafted response must be verified against the source before filing. Purpose-built legal AI tools reduce this risk substantially but do not eliminate the attorney's duty to verify.

Drafty AI's RFE Response Studio: Purpose-Built AI for Drafting RFE Responses

The workflow above works with general-purpose AI. It works better with an AI tool built specifically for immigration drafting. RFE Response Studio is the module in Drafty AI designed for exactly this use case.

How RFE Response Studio differs from ChatGPT

Generic AI tools are trained on the open internet. They are generalists. RFE Response Studio is trained and prompted against USCIS adjudication patterns, immigration-specific regulation citations, and real RFE structures. This shows up in three practical places: issue-category detection from the raw RFE text, exhibit-to-argument mapping, and USCIS-compliant formatting that requires no post-edit pass.

Built-in USCIS and DHS formatting compliance

The output includes the expected structural elements out of the box: cover letter, issue-by-issue brief, exhibit list, proper citation style, and headings that match USCIS convention. Attorneys who adopt the tool typically spend zero time on formatting after the first week.

Evidence-to-issue matching for immigration attorneys

You upload or describe your exhibits once. RFE Response Studio maps each exhibit to the RFE issues it supports, drafts the argument section accordingly, and flags any RFE bullet that has no corresponding evidence—surfacing evidence gaps before the brief is filed rather than after a denial.

Frequently Asked Questions About AI RFE Response Drafting

How much time does AI save on RFE drafting?

The consistent pattern across attorneys adopting AI for RFE drafting is a 60–80% reduction in first-draft time. A response that historically took 4–6 hours of attorney time typically moves to 45–90 minutes, concentrated in the review-and-edit phase. The savings compound across RFE volume: a firm handling 20 RFEs a month reclaims roughly 60 attorney-hours monthly. For the broader picture of how immigration attorneys are reallocating this time, see our guide to 113 time-saving tips for immigration attorneys.

Is it ethical to use AI for RFE responses?

Yes, with the usual duty-of-competence framing: the attorney must understand what the tool does, verify its output, protect client confidentiality, and retain decision authority. Both the ABA and the Florida Bar have published AI guidance aligned on these principles. Using AI to draft an RFE response is professional practice; delegating the attorney's judgment to AI is not.

What RFE types does Drafty AI support?

RFE Response Studio covers the full range of employment-based (H-1B, L-1, O-1, EB-1, EB-2, EB-3), family-based (I-130, I-751, K-1), humanitarian (asylum, U visa, T visa, VAWA), and waiver case types (I-601, I-601A). It also drafts notice of intent to deny (NOID) responses and motions to reopen or reconsider. To see the full feature set or start a free trial, visit the Drafty AI homepage or our free AI tools for immigration attorneys.

If you take one thing from this guide: the attorneys who are winning the most time back from AI are not the ones using the most advanced tools. They are the ones with the most disciplined workflow around the tool. Get the workflow right, and the tool choice follows.