Can ChatGPT Write Your Nexus Letter?

Veterans are doing it. Reddit threads about it. There are even websites now that will generate a "nexus letter" from AI for free. Upload your records, get a letter, take it to your doctor to sign.

It sounds like it should work. The AI can cite medical literature. It can use the right legal phrases. It can produce something that looks exactly like the nexus letters that cost $800 to $1,500.

But looking like a nexus letter and surviving VA scrutiny are different things.

We analyzed 295,756 Board of Veterans Appeals decisions from 2021 through 2025 to understand what makes medical opinions succeed or fail. The data is clear on what the BVA values and what it throws out.

The three reasons AI-generated letters fail at the VA

1. They're conclusory

The BVA's word, not ours. A "conclusory" opinion states a conclusion without explaining why. "It is my opinion that this condition is related to service" with no specific reasoning tied to the veteran's actual records.

AI is good at generating confident-sounding conclusions. It's bad at building reasoning from specific clinical findings in specific records it hasn't actually reviewed. The BVA threw out opinions for being "conclusory" hundreds of times in our dataset.

2. They use speculative language

"Could have been caused by," "may be related to," "it is possible that." These phrases appear constantly in AI output because AI hedges by default. The BVA specifically flags these as reducing probative weight. Warren v. Brown established that "could have been" language is not probative. Tirpak v. Derwinski said the same about "may or may not."

In our analysis, "speculative" was the third most common reason the BVA rated a medical opinion as not probative.

3. They're not signed by a physician who actually reviewed the records

This is the big one. A nexus letter requires a licensed medical professional to review the veteran's actual records, form an independent medical opinion, and sign it. Under Nieves-Rodriguez v. Peake, the BVA evaluates three things: was it factually accurate, was the reasoning fully articulated, and was it based on sound medical principles applied to this specific veteran?

AI can produce the format. It cannot review records, form a medical opinion, or take professional responsibility for the conclusion. If a veteran submits an unsigned AI letter, it's a lay statement. If a physician signs an AI letter they didn't actually review, they're putting their license behind reasoning they didn't develop from records they didn't read.

Where AI can actually help

To be fair, AI isn't useless in the claims process. It can help you organize your thoughts for a personal statement. It can help you understand medical terminology in your records. It can give you a rough sense of whether your condition might qualify for secondary service connection. As a research and drafting tool for lay statements, it has real value.

And if you use AI to draft something and then take it to your own doctor who genuinely reviews your records, edits the reasoning, and signs it as their own opinion, that can work. At that point your doctor is doing the medical work. The AI helped with formatting. That's a legitimate use.

But submitting an AI-generated letter on its own, unsigned, or rubber-stamped by a provider who didn't actually review your records? There is zero data in 295,756 BVA decisions that suggests this helps your claim. None. We looked.

The bottom line

For a medical opinion to carry weight at the VA, a licensed physician has to sign it. That's not a technicality. It's the whole point. The BVA evaluates whether a qualified medical professional reviewed your specific records and explained, with sound reasoning, why your condition is connected to your service.

AI can help you prepare. It cannot replace the physician. The reasoning has to come from your records. The signature has to mean something. And right now, there is no evidence that AI-generated nexus letters improve outcomes at the Board of Veterans Appeals.