USCIS uses several specialized document processing systems, not a single universal ‘robot’ that approves or denies petitions. Each system handles specific tasks: document classification, matching person data from different registries, searching for text patterns. Understanding what these systems actually do with your submission helps you avoid myths about AI detectors and focus on what matters.

What USCIS Systems Don’t Do

The first important clarification: no USCIS system makes the decision to approve or deny a petition. It doesn’t judge whether evidence is good or whether the applicant is truthful. Systems extract information, classify it, link data from different sources, and identify potential matches.

It makes sense to distinguish between two concepts. Legal texts in similar cases do often coincide completely: the legislation is the same, case law is the same, and this is normal. Systems shouldn’t confuse standard legal language with a sign of fraud.

Moreover, some systems only work with specific immigration forms or programs. A claim like ‘this AI checks all USCIS cases’ without specifying which system and its functions doesn’t make sense.

Evidence Classifier: Sorting Documents by Type

The first system that encounters your documents is Evidence Classifier (DHS 16). It receives scanned evidence and determines what each page contains: a passport, marriage certificate, form I-589, bank statement, or another recognized document class.

Then the system applies metadata tags—a simple label of the document type. An officer, instead of manually flipping through hundreds of pages, gets a link to the relevant page. USCIS explicitly describes Evidence Classifier as a full-featured ML service that tags individual pages and creates internal metrics.

Importantly: the system doesn’t judge the quality of evidence. It simply tells the officer: there appears to be a passport here, a Marriage Certificate there.

Intelligent Document Processing: Breaking the Package Into Parts

The second system, Intelligent Document Processing, disassembles the uploaded package into individual documents and classifies them. A large scan stops being a shapeless pile: form separate, passport separate, bank statement separate, marriage certificate separate.

All of this is linked to a specific benefit request, because in the US immigration is a privilege that is requested and may be granted. USCIS has already expanded this functionality to forms I-765, I-131, I-824, related to requests for additional evidence.

Result: the officer sees not chaos, but a structured package where each document is tied to a specific part of your case.

Person-Centric Identity Services: Matching Records About a Person

When USCIS works with more than just the PDF you submitted today, Person-Centric Identity Services (PCIS) comes into play. The system tries to understand which records in different USCIS databases belong to the same person.

It matches name, date of birth, A-Number, Social Security number, passport data, biometrics, and other identifiers. For example, in one system you have A-Number 140, in another 765, somewhere an old address, somewhere biometrics. All of this might sit in separate records.

The system calculates a Confidence Score—the probability that two records belong to the same person. USCIS sets a high threshold: 0.98 out of 1.0. If confidence is insufficient or there’s an error signal, the record goes to manual review.

Result: the officer sees not just today’s form, but a complete immigration history assembled from linked records on a timeline. The system helps identify discrepancies: for instance, if you stated different names or birth dates in different places. But it doesn’t decide whether to approve your petition or deny it. Its task is to correctly identify the person.

Text Analytics: Searching for Recurring Text Patterns

The Text Analytics Data Science Sentence Similarity Model works directly with text. It scans text sections of applications and searches for recurring language patterns—similarities.

This system can indeed find potential matches, and USCIS uses it in the context of detecting possible fraud, national security issues, or public safety threats. The system is already operational and falls into the ‘high impact’ category.

But there’s a fundamental point. USCIS explicitly states: the system doesn’t make decisions, it doesn’t say whether this person is lying, it simply shows a pattern. Results must be reviewed by trained employees, because the system can mistakenly see matches where there are none, or miss real matches.

The main error people often make: confusing Sentence Similarity with an AI detector (like a ChatGPT detector). These are different tasks. If two texts contain similar chunks, the system helps detect this, but it doesn’t follow that it knows who wrote the text: a human, AI, lawyer, paralegal, or someone else. It searches for similarity, not the author’s intent.

What You Should Really Focus On

Of all the published systems, applicants should care not about the words ‘artificial intelligence,’ but about four specific questions: what does the system extract, what does it match, what does it classify, and what exactly does the officer see afterward?

The more a case is built on actual events, real projects, and verifiable documents, the less sense it makes to try to write text in an ‘AI-friendly’ way—changing passive voice, word order, or using gerunds hoping the system won’t understand. These are empty manipulations.

Instead, focus on whether documents actually support the facts stated in your case. Whether one part of your submission contradicts another. Whether you are genuinely the author of what you call your own work. That’s much more interesting and important than any word tricks.

Inter-Agency Data Sharing

Integration and data sharing between agencies is a standard process. For example, you submitted an immigration petition, but years ago you received a tourist visa denial. This is an important fact in the applicant’s history, and the officer should see and understand the reason for the denial.

USCIS systems can receive information from databases of other agencies, so your history of interaction with American authorities is much more complete than you might imagine based on a single submission. All of this serves one purpose: to give the officer the most complete picture.