01

Why immigration is the form-heaviest practice

A typical family-based green-card case generates: I-130, I-485, I-693, I-765, I-864, I-944, plus optional I-131 and supplemental forms. A typical asylum case generates: I-589, I-918 if applicable, supporting affidavits, country-conditions documentation. A typical removal-defence case generates: EOIR-28, motions to terminate or to continue, applications for relief, voluminous evidentiary submissions.

The forms have specific instructions, specific supporting-document requirements, and specific signature and notarisation rules. They are also revised periodically; using a superseded form revision is a rejection-by-USCIS basis.

Three implications for an AI implementation:

  • The marginal hour saved on form-assembly is a meaningful percentage of total case time.
  • Errors in form-assembly cost real time and sometimes real cases (rejections, RFE responses, missed filing windows).
  • Form-revision tracking is a real workflow cost that AI can absorb cleanly.

Immigration practice is, in this sense, the closest to a transactional document-assembly practice that the consumer-facing bar has. The AI thesis follows from the workflow shape.

02

Where AI fits without sanctions risk

The strongest AI workflows in immigration practice are the ones that touch the form-assembly side of the case rather than the legal-judgement side.

Structured intake plus form pre-population. A client-intake form (online or paper) captures the data that will populate the federal forms. AI converts the intake into populated USCIS / EOIR / DOS forms. The lawyer reviews each form before filing. Time saved per case: two to six hours, depending on case complexity.

Document-checklist generation. For a given relief and a given case posture, AI produces the supporting-document checklist. The lawyer reviews and adjusts. The AI here is essentially encoding the firm’s institutional knowledge into a checklist that a paralegal can run, and the lawyer reviews exception cases.

Translation of routine vital-records documents. Birth certificates, marriage certificates, divorce decrees, school records. The translations are formalistic and the certification language is settled. AI can produce the translation; a certified translator (often the same firm or an in-house bilingual paralegal) certifies. Tools like DeepL Pro or specialist legal-translation products plus AI post-editing have collapsed the cost of routine translation.

RFE-response first drafts. A Request for Evidence from USCIS is structured: USCIS asks for specific evidence; the firm responds with the evidence and a cover letter explaining how the evidence satisfies the RFE. AI can produce the cover-letter first draft from the RFE text and the firm’s document file; the lawyer revises.

Country-conditions documentation organisation. Asylum and withholding cases run on country-conditions reports. AI can extract relevant excerpts from State Department country reports, USCIRF reports, and academic literature, organised by claimed-harm category. The lawyer’s legal analysis sits on top of this organised evidence.

03

Where AI breaks: legal judgement and asylum-declaration voice

The two AI uses that tend to fail in immigration practice both involve legal judgement that AI cannot reliably perform.

Categorical-eligibility and discretionary-relief analysis. Whether a client is eligible for a particular form of relief turns on a complex interaction between immigration history, criminal history, family relationships, and the specific statutory and regulatory framework. AI tools that claim to do this analysis tend to be confident and wrong in roughly the same proportions a junior lawyer would be confident and wrong, but without the junior lawyer’s capacity to ask for help. The defensible practice is to use AI to organise the facts and produce checklists; the eligibility judgement is a lawyer task.

Asylum declarations and supporting affidavits. Asylum declarations are sworn statements written in the client’s voice about the events that drove them to seek protection. They are often the central evidence in the case. The voice has to be the client’s, not the lawyer’s and certainly not the AI’s. Translation, structure, and editing assistance are appropriate; voice is not. Immigration judges have started flagging declarations whose voice does not match the client’s testimony at hearing.

04

Translation: where AI has changed the economics most

The cost of competent translation has dropped meaningfully with the current generation of AI tools, and it has dropped most for the documents immigration practice handles most often.

Routine vital-records translations (birth, marriage, divorce, school records) used to cost $35-$75 per document at most certified-translation services. Modern AI tools (Google Translate Cloud, DeepL Pro, Microsoft Translator with the legal corpus, plus AI post-editing in the firm’s general AI) produce defensible translations at near-zero marginal cost. The certification step still requires a human translator’s signature; the substantive translation is largely AI.

What does not work the same way:

  • Asylum declarations, where voice and nuance are evidence.
  • Foreign court documents and judgments, where the legal system’s terminology requires a translator who understands the source jurisdiction.
  • Religious or cultural texts where the meaning depends on context the AI does not have.

For a typical family-based immigration practice, AI-assisted translation drops a $200-$500 per case cost to roughly $40-$100. Across a typical case volume, this alone justifies a meaningful portion of the AI-stack budget.

05

The recommended stack

For a typical solo or small immigration practice:

  • Practice management with immigration-specific tooling. Docketwise, INSZoom, or Cerenade are the immigration-specific options; Clio Manage or MyCase work with immigration form add-ons. (See the 2026 PMS guide.)
  • Form-assembly AI. The form-population workflow built into the practice-management platform (Docketwise and INSZoom both have mature implementations). Alternatively, AI-driven form populators that integrate with structured-intake products.
  • General-purpose enterprise AI. ChatGPT Enterprise or Claude for Work for RFE-response drafting, country-conditions extraction, and supporting-document drafting.
  • Translation. DeepL Pro plus AI post-editing in the general-purpose AI for routine documents. Specialist certified translators for non-routine documents and asylum work.
  • Legal research. Bender’s Immigration Bulletin (LexisNexis) for the citation-of-record source; CoCounsel or Lexis+ AI for AI-assisted research, with the standard verification step.
  • Country-conditions corpus. Direct subscriptions or institutional access to State Department reports (free), USCIRF reports (free), Refworld (UNHCR), and country-conditions services. AI sits on top of this corpus, not the other way around.

Indicative monthly stack budget per attorney: $250-$500 for a non-asylum-heavy practice; $400-$800 for asylum-heavy or removal-defence practice with deeper country-conditions tooling.

06

Implementation playbook (six weeks)

Immigration implementations move quickly because the workflow is well-understood and the tools are mature.

Week one: workflow audit. Pick three current cases at different stages: a family-based green-card application, an RFE response in progress, and an asylum case. Walk through the time spent on each. The point is to understand where AI will help and where it cannot.

Weeks two and three: practice-management decision. Choose between immigration-specific PMS (Docketwise, INSZoom) and horizontal PMS (Clio, MyCase) with immigration add-ons. The right answer for most firms below ten attorneys is the immigration-specific option; horizontal PMS becomes more attractive as the firm grows beyond immigration into family or other practices.

Week four: AI layer. Add the general-purpose enterprise AI subscription. Start using it for RFE-response drafts and country-conditions extraction. Build a one-page firm style-guide for AI use that addresses asylum-declaration voice, translation certification, and citation verification.

Week five: translation workflow. Establish the routine-translation workflow: which AI tool, which post-edit step, which certifier. Document the workflow.

Week six: training and go-live. Two-hour training session for all firm members. Cut over to the new workflow on every new case. Existing cases continue with the legacy workflow until they close.

07

ROI: form-density math

Immigration ROI is dominated by form-assembly and translation savings.

For a representative family-based green-card case, the firm spends:

  • Four to eight paralegal hours on form preparation and document assembly
  • Two to four attorney hours reviewing forms and supporting documents
  • $200-$500 on translations of vital records
  • Variable additional time on RFE responses (when issued)

AI tooling, on the workflows that work, saves:

  • Three to five hours of paralegal time per case (form-population AI)
  • Roughly $150-$400 per case on translation
  • One to two hours of attorney time on RFE responses

For a one-attorney firm running 100 cases per year, this is roughly $40,000-$70,000 of annual cost savings against a $10,000-$15,000 stack budget. ROI in the first quarter.

The savings flow to the firm as margin in flat-fee work (the dominant immigration billing model) and as capacity in hourly-billed work. Either way, the lawyer cannot bill time AI did under Op. 512’s Rule 1.5 analysis.

08

Bar-rules and ethics flags specific to immigration

The general ABA Op. 512 framework applies. The immigration-specific overlays:

  • Detained-client representation. Clients in immigration detention have limited communication channels, no easy access to client-portal AI tools, and elevated confidentiality risks (detention-facility staff, monitored calls). AI workflows that depend on the client’s active participation outside detention need a non-AI fallback for the detention period.
  • Country-of-origin retaliation risk. For asylum and persecution-based cases, the client’s country of origin may have an interest in the client’s submissions becoming known. AI vendors with data-residency in jurisdictions adjacent to the country of origin pose a different risk than vendors based in the U.S. or EU. The DPA review for these cases is more cautious than for non-persecution cases.
  • USCIS’s own data practices. Submissions to USCIS are subject to USCIS’s own retention and use policies, which include law-enforcement information-sharing under specific circumstances. AI tools that pre-screen submissions before filing should not be confused with private legal advice; the submission becomes USCIS’s document the moment it is filed.
  • UPL by AI. Pure-AI immigration tools marketed directly to consumers (without a lawyer in the loop) raise unauthorised-practice-of-law concerns in many states. Practising immigration lawyers using AI inside the firm are not in this zone, but the line between ‘AI-assisted lawyering’ and ‘AI-as-lawyer’ is regulator-watched.

Frequently asked.

Can I use AI to fill out USCIS forms?

Yes — this is the strongest AI use case in immigration practice. Form-population AI built into Docketwise, INSZoom, and the immigration modules of Clio and MyCase will populate USCIS forms from structured intake data. The lawyer reviews each form before filing. The defensible workflow saves 3-5 paralegal hours per case.

Can I use AI to write asylum declarations?

Not the prose. Translation, structure, and editing assistance are appropriate; the voice has to be the client’s. Immigration judges have started flagging declarations whose voice does not match the client’s testimony at hearing. The defensible workflow uses AI for translation of supporting documents and for organisation of country-conditions evidence; the declaration itself is co-drafted with the client.

Is AI translation acceptable for vital-records documents?

For routine documents — birth certificates, marriage certificates, school records — AI translation followed by certified-translator review is now the standard. For asylum declarations, foreign court judgments, and any document where legal terminology or voice carries weight, a specialist human translator is still required.

What about consumer-tier AI for immigration work?

Same answer as for every other practice area after United States v. Heppner: consumer-tier AI exchanges are not protected by attorney-client privilege and the privacy policies permit training on prompts. Enterprise-tier AI (with the data-protection addendum executed) is the floor for any client-document work. Consumer-tier may be acceptable for general legal research that does not include client identifiers.

· END ·

Citations and further reading.

Dan Hughes — portrait of the founder of IXSOR, an AI implementation consultancy for legal practitioners

Dan Hughes is the founder of IXSOR. Twenty years in the UK, twenty-five in North Carolina. Ex-BBC. Ex-Apple. Lifelong technologist. And, most importantly: not an attorney. He writes about legal AI from the operational and infrastructure side, where the rules meet the machines. Reach: [email protected]. More by Dan →