Best AI Tools for EB-1A Eligibility and Evidence Gaps (2026)

Disclosure: Immigration Copilot is our product. Competitor details come from each vendor's public pages as of October 2026.

Two searches keep sending attorneys to this page: which AI tool can assess EB-1A eligibility and find the gaps, and which platforms track progress toward each criterion. Both questions assume a tool can answer something it mostly cannot. It can sort and check evidence. Whether a person is extraordinary is a legal conclusion.

This guide splits the job into the parts software can do and the parts it can't, sorts the available tools into four categories, and gives you a checklist for vetting any vendor before it touches a client file.

Eligibility Means Two Tests, and Counting Is Only the First

An EB-1A eligibility assessment is a two-step legal analysis, and most tools only attempt the first step.

The regulation sets the bar. Under 8 CFR 204.5(h)(2), "extraordinary ability means a level of expertise indicating that the individual is one of that small percentage who have risen to the very top of the field of endeavor." Section 204.5(h)(3) then requires evidence of "sustained national or international acclaim," shown either by "a one-time achievement (that is, a major, international recognized award), or at least three of the following" ten criteria. Section 204.5(h)(4) lets the petitioner submit comparable evidence when those standards "do not readily apply to the beneficiary's occupation."

The Ninth Circuit split the review in two. In Kazarian v. USCIS, 596 F.3d 1115 (9th Cir. 2010), the court held that "neither USCIS nor an AAO may unilaterally impose novel substantive or evidentiary requirements beyond those set forth at 8 C.F.R. § 204.5" at the criteria stage, and placed questions like whether a petitioner is at the very top of the field in a later "final merits determination."

USCIS adopted that structure. The USCIS Policy Manual, Volume 6, Part F, Chapter 2 tells officers to first determine "which evidence submitted by the petitioner objectively meets the parameters of the regulatory description," and then to "evaluate all the evidence together when considering the petition in its entirety for the final merits determination." The same chapter says plainly that "objectively meeting the regulatory criteria in the first step alone does not establish that the person in fact meets the requirements."

So a real assessment answers five questions:

1. Which documents does the client actually have?
2. Which criterion does each document plausibly support, and does it meet the regulatory wording?
3. Are at least three criteria met (or is there a one-time major international award)?
4. Taken together, does the record show sustained national or international acclaim at the top of the field?
5. What evidence is missing, and what would close each gap?

Software is good at questions 1, 2 and 5. It can help organize question 3. Question 4 is the attorney's, and the final merits denial patterns show what happens when a petition treats it as automatic.

What AI Can Check From Documents, Criterion by Criterion

Some criteria are mostly a document check. Others hinge on comparisons no document makes for you. The table below shows what an AI system can reasonably verify from the exhibits for each of the ten criteria, and the color shows how much still depends on attorney judgment and outside evidence.

The pattern in that table matters for any vendor demo. Extraction and attribution are document problems, and good software handles them well. Every red or amber row also asks a comparative question (recognized by whom, distinguished compared to what, high relative to which peers), and the answer usually sits outside the case file. For the regulatory text and evidence lists behind each row, see the EB-1A criteria reference.

One trap shows up in C5 and C8 more than anywhere else. Expert letters describe two people: the client and the writer. A tool that lifts "30 years of experience" or "editor of a leading journal" out of the writer's own paragraph and files it under the client's achievements has produced a false positive that a USCIS officer will catch on the first read.

What ChatGPT, Claude and Gemini Can and Cannot Do Here

General chatbots are the tool most attorneys already have open. They are useful for a first pass at eligibility, as long as you know where the answer stops being reliable.

Here is a typical prompt an attorney or a self-assessing professional might use:

You are an immigration attorney reviewing an EB-1A profile.
Field: computational biology. Current role: senior scientist at a
pharmaceutical company.
- Best Paper Award at an international bioinformatics conference (2024)
- 14 peer-reviewed papers, 1,200 citations total
- Reviewer for two journals, 40 completed reviews
- Profiled once in a trade magazine
- Base salary $310,000
For each of the 10 criteria in 8 CFR 204.5(h)(3), say whether the
profile likely meets it, what evidence would prove it, and what is
missing. Then give a final merits view under Kazarian. Do not assume
facts I have not given you. Mark anything uncertain as UNKNOWN.

A model will return a tidy criterion-by-criterion answer. The output is often a decent starting checklist. The problems sit underneath it.

No evidence verification. The model only sees the bullet points. It cannot tell whether the Best Paper Award has published selection criteria, whether the trade magazine counts as major media, or whether the 40 reviews are documented by the journals. It evaluates a description of the evidence, which is a different thing from the evidence.

Confident criteria fit. Language models tend to fill gaps. Asked whether $310,000 is a high salary, a chatbot may say yes without any wage data for the occupation and location, because nothing in the prompt forces a comparison. The "mark anything uncertain as UNKNOWN" line helps, but it does not guarantee the model will follow it. On legal questions the error rate is measurable: Stanford RegLab and HAI researchers reported that general-purpose chatbots "hallucinated between 58% and 82% of the time on legal queries," and that even purpose-built legal research tools produced incorrect information more than 17% of the time (Stanford HAI, May 2024).

No final merits judgment worth relying on. The model can recite the Kazarian framework. It cannot compare your client to the actual top of computational biology, because it does not know who else is in the field at that career stage, and it will rarely say so.

Confidentiality. Pasting a client's real name, employer and salary into a consumer chatbot raises real ethics questions. ABA Formal Opinion 512 (July 29, 2024) states that "a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool" when the tool is self-learning, and boilerplate engagement-letter consent does not satisfy it. Anthropic says consumer Claude chats are used for training only if the user chooses to allow it, and that a BAA is offered only for commercial services such as the API and Enterprise plans. Google tells Gemini Apps users that a subset of chats is read by human reviewers and asks them not to enter confidential information they would not want a reviewer to see. OpenAI's data-use help pages blocked automated access when we checked, so verify ChatGPT's current terms on your plan directly.

Four Kinds of EB-1A Assessment Tools in 2026

Tools that claim to assess EB-1A eligibility fall into four groups. They differ less in model quality than in what they take as input.

Our recommendation, once documents exist, is the fourth group, and within it Immigration Copilot: it reads every exhibit, keeps letter writers' credentials with the letter, and shows strong, thin and empty criteria per case. It is not a forms or case management system. Before documents arrive, the free questionnaire tools in group 1 are enough.

1. Free self-assessment quizzes and calculators

These are the tools that rank for "EB-1A eligibility check." They ask questions, or take a CV, and return a score per criterion. Two examples, in their own words:

• eb1a.app describes an AI assessment that reviews "achievements, publications, and impact against all 10 USCIS criteria" and asks users to "provide your name, field, employer, and any public profiles. No documents needed."
• eb1avisa.ai advertises a check "in under 60 seconds" that maps achievements against the criteria from a CV and selected milestones, states it is "not a law firm" and "not legal advice," and lists a $47/month tier with an evidence vault and builder tools.

What they are good for: a quick screen of which criteria might be in play, especially for professionals deciding whether to call an attorney. What they cannot do: look at a single exhibit. Accuracy depends on how honestly and precisely the person answered. A user who calls a company hackathon prize a "national award" gets a C1 score the evidence will not support.

Our own EB-1A Case Strength Evaluator belongs in this group too. It asks the attorney to mark the evidence a client has, criterion by criterion, and returns a likelihood per criterion, the strongest combination to argue and the evidence still missing. It is a questionnaire, and we say so.

2. General-purpose chatbots

ChatGPT, Claude and Gemini, covered above. More flexible than any quiz, and able to read documents you upload, but with no built-in criterion structure, no attribution checks and no source citations unless you build them through prompting. The AI in legal practice hub covers tier selection and ethics setup.

3. Case management platforms with AI features

These run the practice: intake, forms, deadlines, billing. Their AI features target admin work. Docketwise, for example, advertises a Document Assistant that "captures client and case data straight into your forms" and a Writing Assistant that "edits, refines, and translates your client communications." Neither feature is described as EB-1A criteria evaluation or evidence gap analysis on its public homepage. Useful software. Different job.

4. Evidence-based petition platforms

These take the client's documents as input. Immigration Copilot sits in this group and is described in its own section below. DraftyAI's homepage says to "upload your case evidence and Drafty AI will prepare your petition, motion, letters, RFE responses," with "client-detail masking before anything reaches an AI provider" (DraftyAI). Its homepage does not advertise per-criterion eligibility scoring or gap analysis. Visalaw.AI's homepage describes immigration-specific extraction, research with citations and drafting, and does not publicly advertise EB-1A eligibility assessment. For a wider comparison of petition software, see best EB-1A petition software.

Assessment is a one-time snapshot. Once the case is open, the question changes to tracking: we compared spreadsheets, case management and evidence coverage views in software to track EB-1A criteria progress.

A Checklist for Vetting Any EB-1A Assessment Tool

Run any tool through these questions with a real, anonymized case file. A demo with the vendor's sample data proves very little.

Criterion mapping to actual exhibits

• Does every criterion finding name the specific exhibit behind it?
• Can you click through to the passage the finding rests on?
• Does the tool separate the client from letter writers, co-authors and people with similar names?
• What happens with an unsigned draft letter or a duplicate? Is it flagged or silently counted?

Final merits reasoning

• Does the tool say anything about the second Kazarian step, or does it stop at "4 of 10 criteria met"?
• If it does discuss final merits, does it cite the record, or does it produce generic language about "sustained acclaim"?

Gap list with suggested evidence types

• Are gaps listed per criterion?
• Does each gap name a document type to collect (selection criteria for an award, circulation data for a publication, a wage survey for salary)?
• Does it distinguish an empty criterion from a thin one?

Source citations

Every factual claim the tool makes about the client should carry a citation to an uploaded document. If the tool cites regulations, those citations should point to the regulatory text or the Policy Manual, not to the vendor's blog.

Data security and terms

A BAA is a HIPAA instrument, and most immigration files contain no health information, so its absence alone is not disqualifying. What you need in writing: no training on your client files, encryption at rest and in transit, isolation between firms, role-based access, and a retention policy. Map those terms to the informed-consent analysis in ABA Formal Opinion 512 before uploading anything.

How Immigration Copilot Approaches EB-1A Gap Analysis

Immigration Copilot is a petition platform for EB-1A and O-1 attorneys. The assessment comes from the documents, not from a questionnaire. Here is what happens to a case file, as described on our features page and homepage:

1. Every document is read and classified. The attorney or client uploads PDF, Word, JPG and PNG files, scans and photos included. Each one is classified, summarized and checked for who it is about. A recommender's CV is filed as the recommender's background, and a namesake in a newspaper is set aside.
2. Facts come with quotes. Every fact is stored with the exact words from the document, and code compares the quote to the source text. A quote that is not there is dropped. Each document is read four times with different instructions, three reads must agree, and splits and unsigned drafts come to the attorney as flags.
3. Evidence is mapped to criteria. Verified facts are linked to the EB-1A criteria they support (or the O-1A and O-1B criteria for O-1 cases). The evidence map works as a graph that connects each document to the facts it proves and each fact to the criteria it supports, across all documents in the case. Context documents connect to the letter they back up and never count as the client's achievement.
4. Gaps surface per criterion. The map shows strong, thin and empty criteria before anyone writes a paragraph, and the gaps tell the attorney which documents to ask the client for.
5. The attorney decides. The attorney picks which criteria to argue. The draft that follows includes a final merits section written against the Kazarian framework, and a validation pass gives each section an RFE risk rating with the reason.

The platform does not decide whether a client qualifies. It assembles the record, shows which criteria the exhibits support and where they are thin, and turns open questions into a short list of decisions for the attorney. Files are encrypted at rest and in transit, every query and file path is scoped to the firm and case, and documents go to AI models through business APIs whose terms exclude training on customer data.

What to Do Before You Pick a Tool

A short decision path, based on where you are with the client:

• Prospective client, no documents yet. A questionnaire tool is enough. Use it to decide whether the consultation is worth booking, and treat the score as a hypothesis.
• Engaged client, documents arriving. You need evidence-level mapping. A chatbot on a business tier with written no-training terms can help if you prompt it carefully and check every citation by hand. A petition platform such as Immigration Copilot does the mapping and attribution for you.
• Record looks complete. Turn to the final merits question. No tool answers it for you. Read the final merits denial patterns and test the record against the "small percentage at the very top" standard in the Policy Manual.

Whatever you choose, keep three habits. Check the tool's criterion findings against the exhibits yourself. Keep client-identifying data out of any tool whose written terms you have not read. And never let a criteria count stand in for the second Kazarian step.

To see which criteria a client might support before the documents arrive, run the free EB-1A Case Strength Evaluator. It takes about three minutes, needs no signup, and returns a likelihood per criterion, a recommended combination and the evidence still missing. When the documents are in, book a demo to see Immigration Copilot build the evidence map from a real case file.