While AI Hallucinations Grab Headlines, Human and Structural Errors May Still Be the Bigger Drafting Risk
Everyone is talking about AI hallucinations—and rightly so.
Courts have issued sanctions. Ethics opinions have multiplied. The profession is being reminded, correctly, that technology can assist but never replace professional judgment.
Yet the quieter, more persistent risk may be the one we create ourselves.
A prominent consumer credit defense attorney, Joseph Kelemen, Founder of the Western New York Law Center, put it this way: “The only reason so many legal documents seem to work is that no one ever reads them—but we do, and the errors we find often decide the case.” Most documents escape immediate scrutiny. When they do not—during litigation, regulatory review, a contested probate, or even a careful client audit—the consequences can be swift and expensive.
It is tempting to think of document errors as failures of care. In practice, they are more often a function of process. Legal documents are long, structured, reused, and repeatedly modified. Across disciplines, research shows that complex documents routinely contain errors even after careful review. Human reviewers catch many of them, but not all, and the remaining errors tend to be small, subtle, and persistent. In legal work, those are precisely the ones that matter.
The issue is not the first draft. It is what happens after.
Documents are copied forward from prior matters, adjusted to fit new facts, and revised under time pressure. Names are updated, dates are changed, clauses are modified, and defined terms are carried across sections. Each of these steps is entirely reasonable. Each also introduces a new opportunity for inconsistency. The risk is not the first change. It is the tenth, when a small variation slips through and remains embedded.
Consider a large estate planning package. It may run hundreds of pages, contain tens of thousands of words, and depend on hundreds—sometimes thousands—of variables that must align perfectly across the document set. Every reference, every relationship, every distribution must remain consistent from beginning to end. When handled manually, this depends on repeated editing. Even for a careful attorney, fatigue becomes a factor, and small inconsistencies begin to appear. A defined term may shift slightly. A clause may not be updated everywhere it appears. A reference may no longer align with the section it points to. Individually, these issues are minor. Collectively, they are not.
We recently worked on a 31-page form containing more than 1,500 variables. That level of detail might suggest complexity, but the experience of using it was quite different. The person completing the form answered just 82 tightly structured questions. From those answers, the system handled the rest, adjusting pronouns and plurals, assembling lists, applying conditional provisions, and ensuring that names, dates, and defined terms were used consistently throughout. What would otherwise require hours of repetitive editing was reduced to a single, controlled input process.
What changed was not the document. It was the method, from several hours entering and re-entering hundreds of individual variables.
Instead of writing, editing, fixing, and rechecking, the process became one of defining information once and allowing it to be applied everywhere it was needed. The lawyer’s effort shifted from repeated intervention to initial design. Judgment was applied at the beginning, where it has the greatest impact, rather than on the fly, where it is most vulnerable to oversight.
This is the point at which rules-based document automation earns its place in a disciplined practice. It does not replace attorney judgment. It enforces the logic and language that the attorney has already determined to be correct. By reducing the need for repeated manual editing, it reduces the primary source of inconsistency in complex documents.
The benefit is not merely speed, although the time savings are often substantial. It is consistency. When information is entered once, controlled centrally, and applied systematically, the opportunity for divergence drops sharply. Errors do not disappear, but they stop multiplying.
Our recent Word Warrior™ piece resonated because it named a familiar frustration: Word’s numbering works—until it doesn’t. The same principle applies here. Manual document assembly works, until the document is actually read closely.
When that moment comes, small inconsistencies become visible. Sometimes they are inconsequential. Sometimes they are not.
The question is not whether errors can occur. It is whether the process allows them to propagate.
The most reliable way to reduce that risk is straightforward. Build the document correctly once, and eliminate the need to keep fixing it.
This is where Doxserá, our professional-grade document automation system, excels—delivering enterprise-level control without requiring enterprise infrastructure, all within Microsoft Word.
See also:
How Many Errors Are Acceptable?
