Think Immigration: Through the Eyes of a USCIS Adjudicator

Every immigration practitioner knows what is filed with U.S. Citizenship and Immigration Services (USCIS). Few know what USCIS sees.

Today’s adjudicator no longer reviews only the forms and supporting documents submitted with an application or petition. Immigration officers at USCIS may encounter information drawn from multiple government databases, prior immigration filings, digitized records, fraud-detection systems, and other agency resources assembled behind the scenes.

In this Think Immigration post, AILA member Angelo Paparelli interviews former USCIS asylum adjudicator Joshua Perez about the adjudicator's view of an asylum filing. Using asylum adjudications as a starting point, the discussion explores a broader reality affecting family-based, employment-based, humanitarian, and naturalization cases alike: the growing gap between what applicants submit and the much larger universe of information available to government decision makers.

Joshua reveals the data available to asylum officers offering a rare glimpse into that reality. Drawing on publicly available government records, he explores how certain analytical tools present information to adjudicators reviewing asylum applications. His observations raise broader questions extending far beyond asylum adjudications and touching nearly every corner of the U.S. immigration system.

Much of what is known about these systems comes from documents immigration lawyers have little reason to read. Privacy impact assessments, system of records notices, and federal AI use case inventories are published to satisfy transparency obligations rather than to guide practices, and they seldom surface in the Policy Manual or in the regulations that shape daily work. The result is a body of public information about how cases are reviewed, published openly and circulated almost nowhere.

As USCIS continues to modernize its systems and connect previously siloed sources of information, understanding the adjudicator's perspective may become just as important as understanding the substantive law. Comprehending that informational environment is increasingly important for lawyers seeking to anticipate questions, explain outcomes, and advocate effectively on behalf of their clients.

Beyond the Application: What USCIS Asylum Officers Actually See

Practitioners increasingly ask whether USCIS uses analytical tools to review asylum applications on Form I-589 and supporting declarations before an interview. Publicly available USCIS documents provide at least a partial answer, including a description of information presented to officers during case review.

What digital tools are USCIS asylum officers using?

Pangaea Text is listed in the DHS AI use case inventory as the Text Analytics Data Science Sentence Similarity Model (inventory identifier DHS-130), and described in DHS/USCIS/PIA-085, the Privacy Impact Assessment for Pangaea Text, published January 2021 and never updated. It analyzes the narrative portions of the Form I-589, including Part B, any supplementary statement, and the certificate of translation, applying rules and algorithms to flag patterns that may point to fraud, national security, or public safety concerns. A FOIA complaint filed by Refugees International calls the application layer Asylum Text Analytics, a name the assessment never uses.

Pangaea Text is one piece of something larger. The assessment describes Pangaea as a suite of data-driven applications built to give USCIS “actionable intelligence" for operational decisions, with Pangaea Text as the component applied to asylum narratives. USCIS committed to updating the privacy impact assessment as it brought additional tools in the suite into use. No update has appeared since January 2021, so what else the suite contains, and what those tools do, is not public.

The public record does not describe a system that approves or denies cases. Rather, USCIS describes a tool that analyzes text, identifies patterns, and presents information for review by human officers. The practical significance lies not in automated decision-making but in how information is framed for the human adjudicator.

What does the asylum officer actually see?

Per the privacy impact assessment, the report "displays a copy of the application where portions of the application have been highlighted to indicate the algorithms and rules-identified patterns."

Not a score, not a risk rating. Nothing describes a number reaching the officer's screen. What reaches it is your client's own declaration, with passages highlighted, pulled up by A-number. Biographic details, including attorney information, come from two other USCIS systems, Global and eCISCOR.

In practical terms, two versions of the same document exist. Counsel filed one. The officer opens another, with a portion marked. The text is identical; the presentation is not. Counsel sees only the first.

The significance of that description extends beyond asylum adjudications. Across USCIS benefit programs, officers increasingly operate within digital case-management systems that assemble information from multiple sources. Although the specific tools differ among programs, the broader question remains the same: What information accompanies an application or petition by the time it reaches the adjudicator?

What is being matched?

The inventory is more specific. The model looks for similarity and relevancy in text to find matches among documents in the system. So the unit of analysis is not internal to one declaration; it is language shared with other filings. The entry adds that the tool stays agnostic about whether a pattern means anything.

The description has softened too. The October 2022 version, preserved in the ACT-IAC use case library, called this plagiarism detection in asylum and withholding cases; the 2025 version drops the phrase.

Is the tool deciding anything?

No, and every document says so. The inventory entry calls it "merely a research tool" that "does not make predictions, recommendations, or decisions." The assessment requires a human step: under PIA Section 2.4, if the system detects a pattern, an asylum officer, an FDNS immigration officer, or both must manually review whether the output can support an adjudicative or investigatory determination. Note who is in that loop: the adjudicator is not downstream of a fraud unit, but one of the reviewers.

Influence is the harder question, and the same entry concedes it. Under potential impacts it calls the system a decision support tool. It answers yes to the high-impact question, the OMB category for AI whose output serves as a principal basis for decisions with legal or significant effect. Its classification field reads Classical/Predictive Machine Learning.

The more important question may not be whether software makes decisions, but whether information highlighted before an officer begins reviewing a file affects how evidence is perceived. A document arriving with flagged passages is not necessarily read in the same way as one presented without annotations. Any influence occurs not through an automated decision, but through the human review that follows. That matters for assessments of credibility. Under the REAL ID standard, the adjudicator weighs inherent plausibility under the totality of the circumstances. Plausibility has no external benchmark; it is measured against what the reader already expects.

Does the client know?

No, and USCIS says so. PIA 4.3 identifies the risk that applicants may not know their information is evaluated this way, mitigated only by the assessment's own publication.

What about errors?

The inventory concedes a small risk of false positives and false negatives, mitigated by manual review. No figure is published, and small is the agency's characterization, not a finding. PIA Section 2.5 notes that OCR can misrecognize characters; below 90 percent confidence the interface shows a banner reading "Scanned Text May Contain Errors."

What should I do now?

Start with the fact that you drafted the declaration. The model matches text across filings, so the overlap it surfaces may be with your own other clients: a template structure, a standard country conditions passage, a translator's stock certificate. The public record does not say whether reused language is treated as an indicator. Note whether credibility questioning fixed on particular phrases rather than events, and whether an RFE demanded documents already submitted. Consider requesting through FOIA the complete A-file, including FDNS records, and do not assume a standard production captures system-generated material.

On correcting the record, PIA Section 7.2 provides that corrections must be made in the source systems rather than the report, that Privacy Act amendment reaches only citizens, permanent residents, and Judicial Redress Act beneficiaries, and that everyone else may visit a field office. PIA 7.4 concludes that there "is no risk associated with redress."

What is still unknown?

Significant questions remain unanswered. Publicly available records do not disclose the specific patterns being identified, the system's actual error rate, whether independent validation studies have been conducted, or how generated reports are used once a case leaves the asylum office. Those unknowns may ultimately prove as important as the information already disclosed. Two FOIA suits are pressing for answers, one squarely and one broadly: Refugees International v. USCIS, 1:24-cv-03559 (D.D.C.), and Pangea Legal Services v. USCIS, 1:24-cv-02809 (D.D.C.) (the latter organization has no relation to Pangaea Text).

None of this changes the substantive law. An asylum claim still rises or falls on the same elements, and the same evidence still has to be gathered and presented. What changes is the practitioner’s picture of the environment in which the evidence is read. That picture is worth having even when it does not point to a particular filing, argument, or objection.

Whatever additional information emerges through litigation or future disclosures, one point is already clear. Immigration lawyers can no longer assume that officers see only the application package submitted by counsel. Understanding the digital context in which cases are reviewed is becoming an essential part of effective advocacy.

Joshua Perez Garcia is a former U.S. Citizenship and Immigration Services (USCIS) asylum adjudicator who also served as a Refugee Officer and Humanitarian Parole Officer. During his tenure with USCIS, he conducted credibility interviews and adjudicated asylum, refugee, and humanitarian parole cases under federal immigration laws and regulations. His writing focuses on immigration adjudications, administrative decision-making, and the evolving role of technology in government case review.

The authors thank Jonathan Micale, Former USCIS Operational Executive, 25-year Immigration Practitioner, and Consulting Principal at Integra Strategic Advisors for his editorial review of this blog post.