Access to an immigration lawyer is a practical barrier as well as a legal one. The U.S. has roughly one immigration attorney for every five hundred people who could benefit from one, while pro bono and clinic capacity are limited. Many applicants must interpret documents in another language and work against unfamiliar deadlines. A carefully scoped assistant can help them read and organize information before professional review.
We have worked on this with two nonprofit partners and one clinic network over the last year. The projects require close attention to what the assistant can explain, when it should refer someone to a person, and who is accountable for reviewing errors.
Tasks an assistant can support
Read a notice and translate what it means
USCIS notices, NTAs, EADs, and hearing notices can be difficult to interpret, and mistakes can have serious consequences. A grounded model can help identify a document and explain its text in plain language. Deadlines and next steps still need verification, with a clear route to professional help when the document or case is unclear.
Walk through a form, in the user's language
Forms such as I-589, I-130, and N-400 contain detailed questions, instructions, and exceptions. An assistant can walk through them in plain language and flag uncertain answers for human review. In our pilot, first-pass accuracy on practitioner-graded forms moved from the low 60s to the mid 80s, leaving the lawyer with a more complete draft to review.
Triage and route
Clinic intake teams need to identify applicants who require urgent legal help and those who need routine information. An assistant can collect answers to standard triage questions in supported languages and organize them for staff review. The clinic remains responsible for assessing urgency and assigning the next step.
What an AI legal assistant absolutely should not do
- arrow_rightMake legal recommendations on close questions. An assistant needs clear boundaries between general information and advice about an individual's case, with qualified review when the distinction matters.
- arrow_rightPredict case outcomes. Outcome prediction in immigration is unreliable and the harm of being wrong is extreme. The model can describe what historically has happened in similar cases. It should not tell a user what will happen in theirs.
- arrow_rightReplace the clinic intake interview for any case in active proceedings. Anyone with a removal date, a credible-fear interview, or active appeals needs a human, not a chatbot.
- arrow_rightEncourage forum-shopping or filing strategies based on judge or court statistics. Even if the data exists, the practice corrodes the system.
Why deployment context matters more than model choice
Decide whether the assistant serves applicants directly or supports clinic staff. A direct-to-consumer tool needs conservative scope, clear limits, and accessible referrals. A clinic tool can support a broader workflow when a named person reviews the output and remains accountable for the case.
We have focused on clinic-facing tools because a named reviewer can assess uncertain outputs and decide what happens next. That accountability needs to be designed into the workflow, including how the applicant reaches a person when the assistant is wrong.
Tools to watch and red flags to avoid
When evaluating a tool, ask for its methodology, known limitations, retrieval sources, and accuracy results. Review its relationships with legal-aid providers and its data retention and deletion policies. Claims such as 'AI lawyer' or 'guaranteed approval' need particular scrutiny because applicants may mistake them for professional advice or a promised outcome.
An assistant can help an applicant read and organize documents in their own language, including outside clinic hours. Building that support well requires reliable sources, careful evaluation, and access to a person when needed. Judge the tool by whether it helps the applicant reach the right next step.
