Using AI to Draft an EB-1A RFE Response: Attorney Workflow

An RFE is not a letter. It's a structured legal document with two distinct parts: boilerplate language about extraordinary ability standards, and specific findings identifying what the officer found deficient. Most unsuccessful responses address the boilerplate. The winning responses address the findings.

Claude is good at the former. It has no access to your record, your client's documents, or the officer's actual concerns. Used correctly as a drafting partner with attorney judgment driving strategy and verification, it cuts first-draft time dramatically. Used naively as an oracle, it writes fluent prose that addresses the wrong problem.

Before You Open Claude: Read the RFE Correctly

The 87-day clock starts on the date printed on the RFE notice, not the date it was delivered, not the date it was opened. Mark this date immediately. If the deadline falls on a weekend or federal holiday, it extends to the next business day. COVID-era flexibility ended March 2023; there are no automatic extensions.

An RFE has a standard structure:

1. Introductory language citing the petitioner and petition number
2. Standard extraordinary ability language (boilerplate: apply minimal attention here)
3. Criterion-specific findings (where the officer identified deficiencies)
4. Standard closing

The criterion-specific findings section is the only part that matters for your response.

Before involving any AI tool, do this yourself:

• Read each specific finding
• Identify which criterion each finding targets
• Classify each finding as: evidence gap (something missing), framing failure (evidence filed but not connected to the legal standard), or legal argument failure (officer applied wrong standard)

This classification determines your strategy: add new evidence, reframe existing evidence, or make a legal argument. AI cannot make this call. It will draft text regardless of whether the strategy is correct. The attorney must decide.

Stage 1: AI Parses the RFE (10 Minutes)

The first AI stage is parsing. Paste the full RFE text into Claude and ask it to:

1. Separate boilerplate language from specific findings
2. Identify which criterion each finding concerns, citing the specific 8 CFR subsection
3. Flag the response deadline based on the notice date
4. Produce a structured finding-by-finding list

The prompt:

You are an immigration attorney reviewing an EB-1A RFE. The RFE is pasted below.

1. Identify the date on the notice (not the delivery date).
2. For each criterion-specific finding (not boilerplate), extract:
   - Which criterion is being challenged (C1 through C10)
   - The officer's specific objection in their own words
   - Whether this appears to be an evidence gap, a framing failure, or a legal argument issue
3. List the CFR subsection that governs each challenged criterion.
4. Do not address the boilerplate extraordinary ability language as a specific finding.

[PASTE RFE TEXT]

What you get back is a structured deficiency map. Review it against the actual RFE. Officers sometimes bury multiple distinct objections in a single paragraph; Claude may miss them. This is a first-pass tool.

The parsed output becomes the skeleton for the rest of the workflow.

Stage 2: Evidence Mapping (20 Minutes)

With the deficiency map in hand, the next stage is mapping your existing evidence against the challenged criteria.

Provide Claude with:

• The parsed deficiency list from Stage 1
• Your exhibit list (file names, exhibit labels, one-line descriptions of what each document is)
• The client's basic profile (field, career level, notable achievements)

Ask Claude to:

• Map each existing exhibit to the criteria it supports
• Identify gaps: what evidence the RFE requires that is not in the current exhibit list
• Flag evidence that was filed but appears to have been disregarded or misread

This mapping often reveals that the issue is framing, not substance. A salary comparison that was filed as a standalone exhibit without a cover letter explaining how to read it against BLS OES percentiles, for example, may have looked ambiguous to the officer. The evidence exists; the argument connecting it to the criterion was missing.

The gap analysis should produce two lists:

1. Evidence gaps that can be filled within the response window (30-87 days is tight but workable for most)
2. Framing gaps where existing evidence needs to be reconnected to the legal standard in the brief

Stage 3: Evidence-Gathering Priority (Days 1–14)

Before any drafting begins, the response needs new evidence where gaps exist.

The AI's role here is narrow: given your gap list and the deadline, help prioritize what to pursue first. It can rank evidence types by expected impact per criterion, flag which types typically take 2-3 weeks to obtain (updated Google Scholar citation counts, new expert letters, salary survey data), and identify which gaps can be addressed by reframing existing evidence versus requiring new documents.

Attorney judgment: which gaps can actually be filled in time, and which gaps are better addressed with the strongest legal argument you can make from existing evidence.

The most common evidence worth pursuing for each challenged criterion:

C5 (original contributions): Named external organizations that adopted the work, current citation analysis (not the one from the original filing; if the petition was filed 18 months ago, the citation count has changed), and new expert letters from practitioners outside the employer who describe field-wide adoption specifically.

C8 (critical role): Third-party media coverage of the organization itself (not just the petitioner), analyst reports, investor names in startup contexts, and industry rankings that establish the organization operates at national or international scale.

C9 (high salary): Three data sources for the percentile claim: BLS OES, a salary survey, and a public platform like Levels.fyi or salary.com. The percentile (not just "above average") must be stated explicitly.

Step 2: The final merits section needs comparative evidence. Who else is in this field at a senior level? What do their metrics look like compared to the petitioner? This is the hardest evidence to obtain and the most often omitted from original petitions.

Stage 4: Criterion Argument Drafts (Days 15–45)

This is where AI delivers the highest time return.

For each challenged criterion, give Claude:

• The specific RFE finding (Officer's exact language)
• The correct legal standard (the 8 CFR subsection and what it requires)
• Your existing evidence plus any new evidence obtained in Stage 3
• One-sentence descriptions of what each exhibit establishes

Ask for a 300-500 word argument section that:

1. Quotes the RFE finding directly
2. States the correct legal standard
3. Argues that the record, including new exhibits, satisfies the standard
4. Cites each exhibit by label and explains specifically what it establishes and how that addresses the RFE deficiency

The prompt for C5:

Draft a 350-word argument section for Criterion 5 (original contributions of major significance).

RFE finding: [paste the exact language]

Legal standard: 8 CFR 204.5(h)(3)(v) requires evidence of "original scientific, scholarly, artistic, athletic, or business-related contributions of major significance in the field."

USCIS Policy Manual Vol. 6, Part F, Ch. 2 clarifies that contributions must be "widely implemented throughout the field" or have "remarkably impacted or influenced the field"; employer-specific impact is insufficient.

Evidence available:
- Exhibit F (Google Scholar report): [brief description]
- Exhibit G (expert letter, Prof. Smith, Stanford): [brief description]
- Exhibit H (company press release showing adoption): [brief description]

Write the section so it: (1) cites the RFE finding directly, (2) states the correct legal standard with the CFR cite, (3) explains what each exhibit establishes, and (4) argues that the record as a whole satisfies the standard. Do not address issues not raised in the RFE. Write as a legal brief, attorney-voice.

The 300-500 word output is a first draft. Your job:

• Verify every factual claim against the actual exhibit
• Confirm the CFR citation is accurate
• Add specific context the AI couldn't know (the organization's market position, how the citation velocity compares to field norms)
• Cut anything that doesn't directly address the specific RFE finding

Attorney time: 30-45 minutes per criterion to prompt, review, and edit, versus 90-120 minutes to draft from scratch.

Stage 5: Kazarian Step 2 Final Merits (Days 40–55)

The Kazarian framework requires that even after meeting three or more criteria at Step 1, the record must support a holistic finding that the petitioner is "in the very top percentage of people in their field." (Kazarian v. USCIS, 596 F.3d 1115 (9th Cir. 2010))

When the RFE challenges Step 2 only (not individual criteria), the response must answer this comparative question directly. A list of criteria met does not answer it. Bullet points of achievements do not answer it. The officer already accepted those in issuing an RFE instead of a denial.

The Step 2 section needs three components:

1. A defined field (not "science"; the specific subfield where "very top" is being claimed)
2. A characterization of who else operates at the senior tier of that field
3. A comparative argument placing the petitioner in the top tier with specific data

Give Claude the comparative evidence you have (citation percentile data, salary percentile, award selectivity rates, publication venue rankings) and ask for a 400-500 word Step 2 section that:

• Defines the field narrowly and explains why that definition is appropriate
• Identifies relevant comparators at the senior tier
• Uses the specific comparative data to argue the petitioner is in the very top percentage
• Cites 8 CFR 204.5(h)(2) directly

The existing Kazarian framework article covers the legal substance in full. The AI's role is drafting the application of that framework to your specific record.

Stage 6: Cover Letter and Exhibit Index (Days 55–70)

The administrative packaging of an RFE response affects how quickly an officer can navigate the brief. A disorganized package requires more officer time to process; more time creates more opportunities for unfavorable interpretation.

Cover letter components:

• Receipt number, petitioner name, beneficiary name, form type (I-140), RFE date
• One-line statement of what the response contains
• Summary of each argument section (one sentence per criterion addressed)
• Note that the original RFE is included on top per USCIS instructions

Exhibit index: each exhibit listed by number, label, and one-sentence description of what it is and what it establishes.

Claude handles both well. Provide the exhibit list and your brief sections; ask for a cover letter in USCIS format and a formatted exhibit index. Review the output for accuracy: the receipt number, dates, and exhibit descriptions require attorney verification.

Submission logistics: the RFE specifies the address for response. Do not send to the original filing address unless the RFE says to. Submit one complete package, not multiple mailings.

Stage 7: Compliance Review (Day 85)

Before assembling the final package, run a structural compliance check.

Give Claude:

• The full parsed deficiency list from Stage 1
• The completed brief with all sections

Ask it to verify:

• Every RFE bullet point from Stage 1 has a corresponding argument section in the brief
• Every cited exhibit exists in the exhibit index
• Every exhibit in the index is cited somewhere in the brief

This check catches the most common structural failure: a brief that addresses four of five RFE issues but misses one because it was buried in the original parsing.

Attorney review: read the compliance check output against the actual RFE notice. The AI can miss issues it did not correctly parse in Stage 1.

What AI Still Cannot Do

The workflow above produces a faster, well-structured response. It does not substitute for attorney judgment on four things:

Strategy selection. Whether to respond, withdraw and refile, or appeal to the AAO: this requires probability assessment based on the specific officer's adjudication pattern, the available evidence, the priority date situation, and client circumstances. No AI tool assesses case-specific approval odds reliably.

Citation verification. Every case name, CFR subsection, and AAO decision ID must be verified by the attorney against a primary source before it goes into the filing. Use the AAO non-precedent decisions search tool for AAO cites; use Google Scholar for case law; use eCFR for regulatory text.

Evidence accuracy. What the brief says about each exhibit must match what the exhibit actually says. The attorney reads every exhibit that is cited.

Privilege protection. Consumer AI plans (Claude.ai Personal, Pro) allow Anthropic to use conversation data for training unless manually disabled in privacy settings. Handle client-identifying information according to your bar's guidance on AI tools. Enterprise plans with data processing agreements present a different situation. ABA Formal Opinion 512 (July 29, 2024) requires attorneys to understand the confidentiality implications of any AI tool before use.

The chatgpt-claude-prompts article provides specific tested prompt language for each RFE stage. The RFE response guide covers the complete substantive framework for each criterion. The RFE prevention playbook documents the eight patterns that generate RFEs in the first place.

For practices handling 20+ RFEs per year, the time return is approximately 60 reclaimed attorney-hours per month. More important than the time: the structural consistency of the brief improves when every response goes through the same seven-stage process, regardless of which attorney handles it.

Immigration Copilot automates the evidence-to-criterion mapping and brief generation stages for new petitions, reducing RFE rates before the notice arrives. See how it works.