Can AI improve your immigration approval? Experts dissect the dos and don’ts

At the same time, a growing number of Indian students, professionals and entrepreneurs are increasingly relying on ChatGPT, Claude, Gemini and other AI tools to write personal statements, support letters, recommendation letters and visa petitions. While these tools can save time and improve drafting, immigration and AI experts say using them carelessly can introduce factual errors, create inconsistencies and, in serious cases, expose applicants to allegations of misrepresentation.

Financial Express Digital exclusively spoke to immigration attorney Kevin J Andrews, immigration and AI expert Divyanshi Chandra, and US entrepreneur and immigration expert Nikin Tharan. Their message was clear: AI can be an extremely useful assistant, but it cannot replace judgment, evidence and human verification.

What AI tools is USCIS using?

According to immigration attorney Kevin J Andrews, USCIS’s Evidence Classifier system does not replace immigration officers. “The government’s Evidence Classifier system does not replace the officer. It flags, organizes, prioritizes, and compares a case against vast historical records,” he said.

However, he said the way the system organises a file can influence outcomes. “If the Evidence Classifier doesn’t pick up a document in the filing, the officer often relies on that determination without checking the file, and issues an RFE asking for evidence already in the record.”

Andrews said USCIS uses these systems for fraud detection, case triage and even drafting notices. He added that visa vetting increasingly extends beyond the initial application. “The State Department screens visas and keeps vetting after a visa is issued, where your social media becomes part of the record,” he added. 

“These systems are part of a growing infrastructure that keeps the vetting ongoing long after the benefit is granted,” he said. 

Why are immigration lawyers talking about same-day Requests for Evidence?

Several immigration attorneys have reported unusually fast Requests for Evidence, or RFEs. Andrews said he has not personally seen same-day RFEs but believes patterns in some notices suggest officers themselves may be relying on consumer AI tools. “I am inferring that from the text and the patterns of the notices, not stating it as official practice.”

He said the same hallucination and inconsistency issues that affect applicants can also enter the adjudication process. “In other words, the same hallucination and inconsistency problems that get applicants in trouble also reach into the adjudication when an overworked officer generates unreliable output with an unsanctioned tool,” he added. 

Which visa categories face greatest scrutiny?

According to Andrews, employment-based immigration categories are under the most pressure because they contain the largest amount of information available for cross-checking. “The categories with the most cross-referenceable data are the most exposed, because there is simply more to check against.”

He said denial rates in several categories have risen sharply. “EB-2 NIW denials now run close to forty percent, and EB-1A faces markedly heightened scrutiny,” he added. 

H-1B petitions face separate challenges because of higher fees, stricter wage rules and processing difficulties, while F-1 visa applicants face greater scrutiny during visa issuance and vetting. “The common thread is data. The more records exist that can be compared, payroll, job postings, prior filings, public profiles, the more surface area a system has to find a mismatch,” he said. 

Are Indians using AI for visa applications?

Nikin Tharan said the increase has been irregular over the past year. “Many applicants are using ChatGPT, Claude, and similar tools because these visa applications are expensive, confusing, and document-heavy,” he said. 

AI makes it easier to draft recommendation letters, petition summaries and personal statements. However, he warned that polished writing does not automatically create a stronger immigration case.

Divyanshi Chandra said many people focus too much on technology and too little on information. “The bigger issue is whether they are giving the tool accurate information in the first place,” she said. 

What mistakes do applicants make most often?

All three experts said the biggest problem is using AI to sound impressive instead of using it to tell an accurate story.

Andrews said AI often invents facts. “AI will produce a confident case citation, a fake award, or a publication that does not exist,” he said. 

He said that several lawyers in the US have already faced sanctions for submitting fabricated AI citations in court. He also warned about generic boilerplate language, contradictions between dates and job histories, unsupported claims and application of incorrect legal standards.

Chandra said many applicants provide only two or three lines of information and expect AI to draft complete legal documents. “Unsurprisingly, the output ends up being generic because the input itself was generic,” she said. 

She also warned that people often overestimate AI’s ability to exercise legal judgment. “AI works best as a support tool that enhances efficiency, while lawyers remain responsible for accuracy, legal reasoning, and final decision-making,” she added. 

Tharan said many strong Indian applicants make the opposite mistake. “One common mistake I see among Indian applicants is trying to compensate for lack of confidence with overly grand language,” he said. 

He said AI often turns ordinary facts into exaggerated claims. “The strongest petitions are not the ones that sound the most dramatic. They are the ones that are specific, credible, and well-supported,” he added. 

Can USCIS detect ChatGPT-written documents?

Experts said there is no evidence that USCIS runs a reliable system capable of determining whether ChatGPT wrote a document. “There is no public evidence that USCIS runs a reliable ‘this was written by ChatGPT’ detector,” Andrews said.

The real issue is not whether the document was written by AI. “They get flagged because unsupervised AI produces generic, templated language that looks like thousands of other filings,” he said. 

He said AI-generated documents also contradict public records and invent specifics.

Nikin Tharan agreed. “The most obvious signs are vague adjectives, repetitive structure, inflated language, and a lack of concrete details,” he said.  He said phrases like “globally recognized leader,” “transformative impact,” and “groundbreaking contributions” without evidence often weaken credibility.

What is the single most important instruction applicants should give AI?

Andrews said there is one rule above everything else. “Do not invent anything. Every citation, fact, date, award, publication, or claim must come from a source I have given you, or one you can point me to so I can check it. If you cannot verify something, stop and tell me, and label it unverified. Treat checking as a separate step from writing,” he said. 

He called this philosophy the foundation of responsible AI use. “Hallucination is like bacteria on hands. You need to wash after every use,” he said. “Verification is always a separate step from creation,” he said. 

Chandra suggested beginning with a detailed instruction: “I am an Indian applicant applying for a [specific visa category] to [country]. Use only the facts I provide. Draft a personal statement that clearly explains my purpose of travel, financial circumstances, professional or academic background, ties to my home country, and intention to comply with all visa conditions. Do not make assumptions or add information that has not been provided.”

She said successful visa statements are not necessarily the most eloquent ones. “They are the most credible,” she added. 

Tharan recommended a different prompt: “Rewrite this section to be clearer and more concise. Do not add any new facts. Do not exaggerate. Preserve the original meaning. Make the language specific, evidence-based, and credible for a USCIS officer,” 

He also recommended asking AI to identify vague or unsupported statements with the prompt: “Review this draft and identify any sentences that sound vague, exaggerated, generic, or unsupported by evidence. For each issue, explain why it may weaken credibility and suggest a more specific alternative.”

What should applicants tell AI not to do?

Chandra said applicants should explicitly tell AI to avoid assumptions, emotional language and unsupported claims. She suggested a prompt: “Use only the facts I provide. Avoid assumptions, exaggeration, emotional language, generic statements, and unsupported claims. If additional information is required, ask for clarification rather than filling gaps on your own.”

She said phrases such as “It has always been my dream” may sound persuasive but immigration applications are evaluated on facts and evidence.

Andrews also advised applicants to prohibit AI from filling information gaps. “If it is missing a fact, it should ask you or leave a clearly marked blank, not manufacture one.”

Why should applicants focus on workflows instead of prompts?

Andrews said he spent the first few months of his AI journey searching for “magic prompts.” He eventually reached a different conclusion. “Search for better workflows, not better prompts,” he added. 

He advised applicants to build a small searchable library containing Administrative Appeals Office decisions, policy memoranda, USCIS handbooks and practice materials. “It pulls the model off its guesses and onto real text,” he added. 

He also suggested running several independent research passes across multiple models. “Where they agree, you are on firmer ground. Where they disagree, you have found the gray area that needs your own judgment,” he said. 

He uses stronger models to design workflows and cheaper models to execute them once they become stable.

He also said that academics map their citation networks through Google Scholar and use browser-based assistants to inventory code repositories and portfolios to identify concrete evidence of impact. “AI features are advancing faster than our ability to master the features,” he said. 

How should applicants personalise AI-generated drafts?

“Do not try to make AI text sound human. Make the document true, and make it yours,” Andrews said.

He advised applicants to replace stock phrases with actual projects, real numbers and genuine recognition. “A document built from your real facts reads as yours because it is yours,” he added. 

Chandra recommended giving AI as much information as possible. “The more authentic information you provide, the more authentic the final document will feel,” she said. 

Tharan said applicants should ask four questions before submitting any document. “Is this true? Is this specific? Can I prove it? Does this match the evidence?” he said. 

“If the answer is no, that sentence should be rewritten or removed,” he said. 

Should prompts differ for different visa categories?

Andrews said every category requires a different approach. “For EB-1A, metrics matter. Before 2024, the letters needed to be read like the back of a novel. Now they read like the back of a baseball card. Stats over narrative, metrics over adjectives,” he said. 

For H-1B support letters, he said consistency is everything because the documents must align with labour applications, job descriptions and wage levels. For F-1 personal statements, he had a different recommendation. “Use AI to organize and tighten. Never let it write the statement for you,” he said. 

“A reader can feel an empty one. Talk, don’t type,” he added. 

How far can AI realistically take applicants who cannot afford lawyers?

Andrews said AI can help considerably but has limits. “AI can help very much with analysis and very little with judgment,” he said. 

It can help applicants understand criteria, organise evidence, draft personal statements and translate documents. However, legal judgments remain beyond its capabilities. “What it cannot do is the part that decides outcomes,” he added. 

He also warned about misrepresentation. “A material misstatement can become a lifetime bar from future immigration benefits,” he said. 

Tharan agreed. “AI can take them part of the way,” said Tharan.  He said it can improve organisation and writing but cannot reliably determine whether evidence satisfies legal standards.

Could your conversations with AI become evidence?

Andrews raised privacy concerns. He referred to the federal case United States v. Heppner, where a court held that a person’s chats with a consumer AI platform were not privileged and could potentially be used against him. “When you pour your case into a free chatbot, you may be building a discoverable record of your own doubts,” he said. 

He said lawyers may provide something beyond expertise. “A lawyer may be buying protection for the work itself,” he added. 

Which AI tools do experts recommend?

Andrews said the choice of tool matters far less than methodology. “The free and low-cost tiers of the major assistants are surprisingly capable,” he said.  He said he regularly uses Claude, Perplexity and Gemini and occasionally uses ChatGPT.

Chandra said Claude works particularly well for longer documents, while ChatGPT is useful for brainstorming and refining drafts. She also recommended Gemini for users already working in the Google ecosystem.

“Ultimately, the quality of the final document depends as much on the user’s instructions and review process as it does on the AI itself,” she said. 

Advice of experts for applicants

All three experts returned to the same principle. “Verification is always a separate step from creation, and you never file anything you cannot personally swear is true,” Andrews said.

Chandra said, “Don’t ask AI to write a convincing visa document. Give it the information it needs to tell your story accurately and credibly.”

Tharan added, “Use AI as an editor, not as the architect of your case.”

Disclaimer: This article is intended solely for informational purposes and should not be construed as legal or immigration advice. Financial Express does not endorse or discourage the use of artificial intelligence tools in preparing visa or immigration documents. Immigration laws and application requirements vary based on individual circumstances. Applicants should consult qualified immigration attorneys or authorised experts before submitting any application or making immigration-related decisions.