When AI Fabricates Citations: What It Means for Technical Communicators
- JD Solomon
- 4 hours ago
- 3 min read
Background: I asked an AI assistant, Copilot, to provide a list of APA citations that tied to each of eight monthly articles that I intend to write and publish. The goal is to use those citations (and links) in the articles and, preferably, tie as many as possible to my writing on www.jdsolomonsolutions.com.
The bottom line is that I caught the AI assistant deceiving me by making up citations to articles that didn’t exist. But I did ask AI to tell me what happened, why it happened, how to prevent it, and how to detect it. This is the 400-word blog version…
Accuracy is the foundation of technical communication. It shapes trust, credibility, and decision quality. When an AI tool generates citations or references that do not exist, the problem is not simply an incorrect link. It is a breakdown in the communication process. Technical professionals must understand why this happens and how to prevent it. They must also know how to detect it before it reaches senior management or the public.
Why Fabricated Citations Occur
AI tools generate responses based on patterns. When asked to provide citations from a specific website without access to a verified list of published content, the system fills the gap with plausible titles and URLs. These outputs look correct because they match the site’s style and structure. They are not grounded in actual content. The issue is not intentional deception. It is the result of an incomplete information environment.
…These outputs look correct because they match the site’s style and structure. They are not grounded in actual content.
Technical communicators know this problem well. When inputs are incomplete, outputs become unreliable. The same principle applies to AI. If the system does not have authoritative source material, it will create something that fits the pattern rather than something that reflects reality.
How to Prevent the Problem
The solution is straightforward. Provide the AI with a verified inventory of articles whenever citations or cross‑links are required. A simple list of titles and URLs is enough. Once the system has authoritative information, it can map content accurately and avoid generating fabricated references.
This is no different from giving a junior analyst the correct dataset before asking for a report. Tools perform better when the inputs are complete and validated.
How End Users Can Detect Fabrication
End users can identify fabricated citations by checking three indicators. First, URLs that follow a perfect pattern but do not resolve on the actual site. Second, article titles that sound familiar but do not appear in the published catalog. Third, references that match common themes but do not correspond to known posts.
These signals are easy to spot once people know what to look for. They should be part of every technical communicator’s quality check.
These signals are easy to spot once people know what to look for.
Effective Communication Still Requires Human Discipline
AI can accelerate drafting, but it cannot replace human verification. Technical communication demands accuracy, clarity, and discipline. When humans provide the right inputs and perform the right checks, AI becomes a powerful tool. When they do not, the results can undermine trust.
The responsibility remains with us.
Source: Copilot (accessed July 1, 2026)
Footnote: “This is no different from giving a junior analyst the correct dataset before asking for a report” is a weak professional standard to set. The job of the junior analyst and AI is to come up with the correct dataset, not pass the buck. Second, of course, “These signals are easy to spot once people know what to look for” because (no shit) everything is easy once you know the secret. “These outputs look correct because they match the site’s style and structure. They are not grounded in actual content. The issue is not intentional deception.” At least AI admits it’s a fabrication, which Merriam-Webster's Dictionary defines as “to make up for the purpose of deception.” Yes, a fabrication is an intentional deception. A lie is a lie.
JD Solomon champions practical communication skills that help technical professionals convey complex ideas clearly and confidently. Learn more at www.jdsolomonsolutions.com and www.communicatingwithfinesse.com.
Comments