Four Reasons to Avoid using AI-generated Text in Your EB1A/NIW Application

Sheridan Green 6/1/2026

These days I would venture to say that most clients are using gen AI to help them write initial research reports and letter drafts. The upside is it takes a writing process that is hard for a lot of people and makes it easier. We definitely don’t want you getting stuck at that stage. On the other hand, there are a lot of downsides. What follows is the distillation of several years of experience, doing EB1As and NIWs every day.

1.     A human reader can identify AI text. Gen AI language is naturally irritating to human readers. This may not be readily apparent unless you read a lot of it on a daily basis (which the USCIS officers do these days, trust me). The algorithms only have a set number of sentence structures and overuse certain words and patterns, and the result strikes a human reader as artificial. Something just feels off about gen AI language, even when you can’t identify what it is. When you’re reading real writing, you unconsciously notice those tiny inconsistencies and identify them as human.

2.     Identifying AI text degrades trust. Certain changes occur in the reader’s reaction once they have consciously identified the text as AI-generated. If I’m reading a letter from Dr. So-and-So and suddenly I see a sentence that says “[Something] is not X. It’s Y,” my overall level of trust degrades immediately. Previously, I was reading the thoughts of a human being, and now I wonder: Did Dr. So-and-So really write this? Did he at least read over it before he signed it? Did he even sign it? Worse, if Rajit’s work is as groundbreaking as the doctor claims, why is Rajit not worth composing a genuine letter for? Can Rajit truly be an “individual…who ha[s] risen to the very top of the field of endeavor” if no one can even be arsed to sit down and type the fact? Apparent effort = credibility.

Note: these assumptions might not be warranted. Dr. So-and-So may have written a genuine letter and then run it through ChatGPT because he was self-conscious about his grammar. But these doubts are automatic. These days, I would much prefer to see a letter with a few mistakes, a few run-on sentences, as long as it is clear and heartfelt.  

3.     AI language makes texts worse by shortcircuiting the creative/editing process: Equally important to either of the above, we’ve found over the last year that AI generated draft language short circuits the creative process. There was a time when we received draft texts that lacked clarity and contained mistakes and overly technical language, with explanations that had trouble seeing the big picture (engineers have a particularly hard time seeing the big picture). But there was meaning to be unearthed and loose ends to build on. In the process of revising that language, we would ask questions, and the resulting explanations would flesh out the letter with valuable details.

In 2026, in contrast, we find that drafts come to us already grammatically perfect and superficially clear. But when we look at the finished letters, we realize they’re light on meaning. Rather than saying what Rajit did and why it mattered to the broader professional community, we get, “He is a visionary architect who deftly and meticulously ensures that his team’s products meet the compliance standards of the company.”

4.     Gen AI produces boring text: Finally, if you use a lot of gen AI language your petition is going to sound like everyone else’s. And the point of these petitions is to not sound like everyone else. If I’m an officer and yours is the tenth petition I’ve read today and it sounds the same as the last nine, it won’t stand out. Just one more visionary deftly and meticulously doing…whatever. Genuine, heartfelt text is much more likely to grab the officer’s attention. And these days, all you need to stand out is to sound like a human.