Share

As more people use AI to generate written content, college admissions officials are seeing AI-generated admission essays, teachers are seeing AI-generated research papers, and more companies and nonprofits are using generative AI to create both written and visual marketing material. Some of these uses are inauthentic (as in the case of the AI-generated college admissions essay and the marketing materials) or even unethical (as in the case of the plagiarized paper), and to address the rise of such cases, more people are turning to tools like AI Detector to help identify written material that doesn’t have a personal, human origin.

AI and the Problem of Authenticity

One of the problems with inauthenticity in written content is that, in cases like those above, it prevents the reader from achieving their objectives in accessing that content. The teacher or professor needs to assess the student’s actual knowledge and possibly their writing skills; they will not be able to do so accurately based on a paper that has been created by a large language model (LLM) such as ChatGPT.

The admissions official needs to get a sense of the person who has submitted the college essay, collecting from their own words an understanding of their values, expertise, hopes, and dreams; the admissions officer cannot learn this from an AI-generated essay.

The customer or potential business partner needs genuine knowledge of the character and philosophy of the company or nonprofit they are investigating. They will not be able to gauge this from AI-generated marketing materials. Even though the materials may not lie to them, they do not provide the organization’s authentic voice.

There are other cases, too, where AI-generated content creates issues for other consumers of written material. For example, a busy parent looking up the news in the morning needs to trust the accuracy of the article they’re reading, but if the article has been generated entirely by AI and hasn’t been adequately fact-checked, it may include information that is out of date or even false. Large language models are still hallucinating content regularly, referencing “facts,” quotations, resources, and anecdotes that simply don’t exist.

The Growing Need for AI Detection

For all these reasons, the rise of AI-generated content presents challenges to a broad array of industries, including education, marketing, and journalism, all of which place a premium on originality, trust, and authenticity. As LLMs have become increasingly sophisticated, detecting the differences between human-generated and AI-generated content is also becoming more difficult.

New tools like AI detectors offer solutions to this challenge by analyzing written copy, returning a report specifying how likely it is that the material is AI-generated, and often identifying the sources of the copy. These tools can flag inauthentic content quickly, allowing admissions officers, educators, chief editors of newspapers and newspaper readers, business partners, and legal teams—all those who are relying on the originality of what they’re reading—to know when they are being offered only a simulacrum of what they seek and when they can trust the source.

The Curious Case of the Declaration of Independence

Both AI generation and AI detection are new technologies and are evolving at a rate that can feel difficult to keep up with. Early AI detection tools faced difficulty providing accurate results. In one famous case, a copywriter ran the Declaration of Independence through an AI detection tool and discovered that, according to the tool, this founding document of the United States was 98.51% AI-generated—an obvious impossibility for a document authored in 1776.

Concerns have also been raised over the past year about “false positives” in detecting AI-generated college papers. False positives can have severe consequences for students falsely accused of cheating, but as AI detection tools become more complex, cases of such errors are increasingly rare. Content specialists who have declared that AI detectors “don’t work” may have been overstating the matter; what appears to be the case is that AI detectors work most of the time.

The Declaration of Independence flagged a false positive precisely because it isn’t an authentic example of modern writing. AI detectors are trained to analyze texts for low burstiness and low perplexity. A document with “low burstiness” shows little variation in the length of its sentences (something formal eighteenth-century writing and AI actually have in common). Low perplexity means that a document’s diction is predictable. Its word choice and vocabulary feel “canned” by contemporary standards; it may use cliches or overuse “business speak.”

An AI Arms Race

Each new LLM produces written text that is less canned and more natural than the last (though still not human). For this reason, LLMs and tools like AI detectors are locked in a technological “arms race.” Because AI detection tools are trained on the currently available generation of large language models, the release of new LLMs can catch them briefly off guard. AI detection tools are constantly improving to keep up with the advancement of generative AI engines.

What Does the Future Hold for AI Detection Tools?

While it may be easy to laugh at generating a false positive with the Declaration of Independence, the reality at the outset of 2025 is that AI detectors are becoming widespread in their use, more accurate, and more trusted than in the past, with some detectors famously boasting scores of 98% confidence.

Two-thirds of teachers now report that they use such tools regularly. Journalists and publishers are using AI detectors on submitted materials to check the originality of anything public-facing and preserve their organizations’ trustworthiness and credibility. Brands are relying on AI detection tools to help their message remain unique and personal so that they can connect with their audience in an authentic way. In the corporate and legal sectors, AI detectors are becoming an essential step before approving written copy, verifying that important documents are original, consistent, and free from AI-generated hallucinations or inaccuracies.

As both AI generation and AI detection improve, perhaps the best advice anyone can take to heart is “stay on your toes.” Don’t become complacent with your current strategy for detecting and responding to AI-generated content. Technology becomes outdated swiftly, and when choosing an AI detector, you will want one that is up-to-date, highly accurate, and produced by an organization committed to its product’s ongoing improvement and iterability.