Detecting AI-Generated Content
How Forensic Accountants and Business Valuation Professionals Can Give AI a “Run for its Money”
As AI proliferates, business valuation and forensic accountants will need to distinguish between human-authored and AI assisted/generated content. Significantly, AI generated content, as of the date of writing, is not accurate. In this article, the author examines what the above will mean to forensic accountants and business valuation analysts.
With the rise of content generated by artificial intelligence (AI), it makes sense that there would also be an increase in online tools that can detect AI-generated content. What does this mean for professionals in the forensic accounting and business valuation fields?
As AI writing tools like ChatGPT gain popularity, distinguishing between human-authored and AI-assisted content (and/or charts and images) will become necessary; not just for upholding professional standards but also for ensuring the trustworthiness of reports that inform big business decisions and litigation outcomes. However, forensic accountants and business valuation professionals challenge these AI tools’ accuracy due to the technical styles many professional reports are written in.
How AI Detection Tools Work
AI detection tools are becoming the lie detectors of the digital age. They analyze text to figure out whether it was fully written by a human, partially generated by AI, or fully AI-generated. These tools use various methods (such as examining linguistic patterns, statistical models, and machine learning techniques) to spot the telltale signs of AI writing. Some of these tools are even able to tell the difference between different AI models, i.e., whether something was written by ChatGPT, Gemini, Claude, or another model.
One of the most useful features of AI detection tools is their scoring system, which provides a percentage rating to indicate how likely it is that a piece of content was AI-generated. For instance, a tool might look at a report and declare it “75% likely to be AI-written.” This scoring system can be helpful for forensic accounting and business valuation professionals who depend on the accuracy and credibility of their written reports, or who are challenging the reports of other experts.
How Accountants are Like AI: Precise, Reliable, and Slightly Less Conversational at Parties
Forensic accountants and business valuation professionals are no strangers to sounding a bit like robots themselves. (Let’s face it, we have all heard the jokes.) Accountants have always had a reputation for being the “robots” of the business world long before ChatGPT came along. Forensic accountants and business valuation experts are often so highly educated and so focused on accuracy that their writing can come across as methodical, technical, and yes, sometimes a bit dry—much like AI.
The challenge is that these professionals are naturally inclined to write in ways that could seem AI-assisted, even when it is not AI-assisted. So, when an AI detection tool flags their work as “suspiciously robotic,” it is difficult to say whether or not that is actually the case when the author is from the litigation support or valuation fields.
Jokes aside, it is important for professionals in these fields to recognize that their writing could be misidentified as AI-generated. This awareness can help professionals in the field refine their style to better reflect their unique voice and distinguish their work from the predictable patterns of AI.
The Mechanics Behind AI Scoring
AI detection tools score content by analyzing several factors:
- Linguistic Patterns: AI-generated text often follows specific patterns, such as repetitive phrases or overly formal language that lacks variability. In contrast, (non-accountant) human writing tends to be more varied and nuanced.
- Sentence Structure and Complexity: AI models like ChatGPT can produce complex sentences, but they may miss the subtle human touch, especially in how ideas connect or flow.
- Predictability and Perplexity: AI-generated content can be overly predictable, often choosing the statistically “safest” next word rather than the most contextually fitting one. Detection tools measure this “perplexity” to gauge how much the text aligns with human writing.
- Stylistic Features: These tools also look at things like punctuation, paragraph structure, and word choice. But since forensic accountants and valuation professionals often write in a precise, technical manner, AI detection tools may not be able to correctly spot the difference between a skilled professional’s work and AI output yet.
Why it Matters
For forensic accounting and business valuation professionals, knowing whether AI was used to create content is increasingly important. Accurate detection ensures that reports and analyses are based on genuine expertise, not just algorithmic patterns. This is crucial when the findings of those reports influence high-stakes decisions, such as testimony to judges and juries.
- Professional Integrity: Being able to distinguish between human and AI-generated content helps maintain the credibility of reports, ensuring they truly reflect the professional’s expertise.
- Source Verification: In forensic accounting, verifying the authenticity of documents is critical. AI detection tools can help ensure that key documents are genuinely human-authored, supporting the validity of the analysis.
- Adapting Writing Styles: By understanding how AI-generated content is identified, professionals can adjust their writing styles to maintain their unique professional voice and avoid unintended AI-like patterns.
What if an Opposing Expert’s Report is AI-Generated?
The rise in AI-generated content brings a new question to expert witnesses: What happens when an opposing expert’s report is substantially AI-generated? While AI detection tools can give us clues, the real challenge begins when this fact is uncovered in litigation.
One of the first issues to address is whether the expert’s reliance on AI affects the credibility of their testimony. Expert witnesses are relied upon for their specialized knowledge and the application of that knowledge to real-world financial problems. If a report turns out to be substantially AI-generated, the expert may find themselves in a tricky position, especially when it comes to justifying their findings in depositions.
Credibility
An expert who heavily relies on AI tools to generate a report might face scrutiny over how much of the report reflects their personal expertise. Opposing counsel could ask questions like, “Did you personally write this analysis, or did AI generate it?” and “What specific judgments did you make, separate from the AI’s output?” These questions may challenge the expert’s authority or weaken their conclusions.
Integrity
In depositions, attorneys and/or judges may focus on uncovering how the expert used AI. For example, “Can you walk us through your decision to use AI, and how that affected the report’s conclusions?” or “Did you review and verify the AI’s output for accuracy?” The expert might be unable to fully explain their report if they relied too heavily on AI, raising doubts about the integrity of the expert’s analysis.
Ethical and Legal Implications
Adverse parties may argue that presenting an AI-generated report without proper disclosure violates ethical standards or disclosure rules. It may lead to Daubert and/or other admissibility challenges to exclude all or portions of the report, for example, in situations where it is found that the expert passed off the AI’s work as their own without adequate disclosure. In extreme cases, a court might disregard the report altogether if it determines that the expert’s involvement was minimal.
Deposition and Trial Ramifications
If the AI-generated nature of a report is discovered during a deposition, the follow-up questions could become increasingly targeted. “What portion of this analysis is truly yours?” or “Can you explain this conclusion without referring to the AI-generated report?” These are scenarios where the expert could potentially struggle.
How Accurate are AI Detection Tools?
The author tested three free AI detection tools by providing each with the following AI-generated paragraph:
“This comprehensive report, along with the attached schedules, outlines my detailed financial analysis, findings, and conclusions, as well as the data and assumptions utilized in reaching these determinations. Included with this report is a summary of my professional credentials, highlighting my extensive background, previous publications, and instances of expert testimony provided.”
And this human-written paragraph:
“This expert report and accompanying schedules (in Exhibit B) present my financial analysis, findings, conclusions, data, and assumptions considered in forming them. My professional profile, including details of my prior publications and testimony, accompanies this expert report (in Exhibit C).”
Here are the findings:
- Grammarly’s AI detector[1] correctly determined the AI-generated paragraph above was written by AI and correctly determined the human-written paragraph above was written by a human.
- ai’s AI detector,[2] which boasts that it is 99.8% accurate and 100% free to use, incorrectly determined that the AI-generated paragraph above was “likely to be written by a human,” and could not analyze the human-written paragraph (as it is less than 50 words long).
- ZeroGPT’s AI detector[3] incorrectly determined that the AI-generated paragraph above was human-written. It also correctly determined that the human-written paragraph was written by a human.
Embracing the Future of AI Detection in Professional Writing
As AI continues to develop, the tools that detect it will also advance. For forensic accounting and business valuation professionals, staying updated on these tools is essential; not just to safeguard written works but to ensure that the integrity and authenticity of the field remain intact.
Beyond protecting one’s own work, professionals must also be prepared to evaluate the credibility of opposing experts’ reports, particularly if they suspect significant AI involvement. An AI-generated report could undermine the expert’s credibility in deposition and courtroom settings, leading to tough questions about the expert’s personal contribution to the analysis.
With these tools, professionals can continue to deliver work that is distinctly human, crafted with expertise, experience, and that personal touch and expert thought that no AI can quite replicate … yet.
[1] Grammarly’s AI Detector: https://www.grammarly.com/ai-detector?utm_source=google&utm_medium=cpc&utm_campaign=21545053911&utm_content=709620349807&utm_term=run%20my%20paper%20through%20ai%20detector&target=&targetid=kwd-2281794560399&adgroup=169269628281&device=c&matchtype=b&placement=&network=g&extension=&clickid=EAIaIQobChMIjfKc4_ukiAMVwg6tBh2cjjZ-EAAYASAAEgIsWfD_BwE&gad_source=1&gclid=EAIaIQobChMIjfKc4_ukiAMVwg6tBh2cjjZ-EAAYASAAEgIsWfD_BwE&gclsrc=aw.ds
[2] Phrasly.AI’s AI Detector: https://phrasly.ai/ai-detector?gad_source=1&gclid=EAIaIQobChMIjfKc4_ukiAMVwg6tBh2cjjZ-EAAYAiAAEgLfhfD_BwE
[3] ZeroGPT: https://www.zerogpt.com/
Miranda Kishel, MBA, CVA, CBEC, MAFF, supports the completion of business valuations, business calculations, forensic accounting, economic damages, and asset tracing engagements, focusing on in-depth financial analysis including modeling, forecasting, research, and report preparation. Her previous experience lies in small business consulting, commercial lending, accounting, real estate development, and economic development.
Ms. Kishel may be contacted at (218) 404-4877 or by e-mail to mkishel@litcpa.com.
