AI Nexus Letters: What Actually Holds Up at the VA

We analyzed 295,756 Board of Veterans Appeals decisions to understand one thing: what makes a private medical opinion actually hold up, and what gets it thrown out. Every one of those decisions is a public record published by the VA.

We are not going to walk you through the spreadsheet. What matters is what we learned, because it is the same standard your letter will be judged against.

The VA does not weigh all opinions equally

The Board does not simply note whether a private opinion exists. It evaluates the opinion’s probative value using the framework from Nieves-Rodriguez v. Peake (2008). Three questions decide it:

• Is it factually accurate?
• Is the reasoning fully articulated, not just asserted?
• Is it based on sound medical principles applied to this veteran’s record?

This is the part most people miss. A letter that fails those questions is not neutral. When the Board explicitly rejects an opinion, that rejection can end up strengthening the denial rationale. A weak letter can leave a veteran worse off than sending nothing at all.

Why the Board rejects medical opinions

The same failure modes come up again and again, in the Board’s own language:

• Speculative. Hedging language such as “could have,” “may be,” or “possible.” Warren v. Brown: “could have been” is not probative. Tirpak v. Derwinski: “may or may not” is not probative.
• Conclusory. An opinion stated without explanation. A conclusion without reasoning carries no probative value regardless of who signed it.
• No rationale. The Board needs to see why, not just that someone qualified agrees.
• Did not consider contrary evidence. Failing to address alternative causes or contradictory findings in the record.
• Did not review the file. An opinion not grounded in the veteran’s actual records.

Where AI-generated letters fail

Every failure mode above is a default behavior of AI text generation. AI hedges, which reads as speculative. It states conclusions without building reasoning from specific records, which reads as conclusory. And it has not reviewed anyone’s claims file, because it has never seen one.

An AI-generated nexus letter is text that resembles a medical opinion. The VA’s framework does not reward resemblance.

What actually holds up

Across everything we read, the letters that carried weight had the same things in common:

• A qualified medical professional
• Who actually reviewed the veteran’s specific records
• And applied sound medical reasoning
• To explain why this condition is connected to this veteran’s service
• While addressing the alternative explanations a reviewer will think of

AI has legitimate uses here. It can help a veteran prepare a personal statement, understand medical terminology, or organize their thinking before they talk to someone. What it cannot do is stand in for a physician who has read the file and is willing to put their name on the reasoning.