Can you really tell if something was written by AI? In this episode, I dig into why AI detection tools are far less reliable than most people think, what the latest research actually says, and why provenance, transparency, and human judgement might be the smarter path forward.
The Uncomfortable Truth About AI Detection
The short answer to “was this written by AI?” is: we can’t know for certain. And that’s a genuinely uncomfortable place to be when decisions about grades, jobs, and reputations are on the line.
I came across a really fascinating piece of research journalism by Reuters journalist Marina Adami, who went deep into the academic literature on AI detection tools. Her findings — and the broader research — paint a pretty sobering picture.
Why AI Detectors Fall Short
Current AI detectors are inconsistent. Some studies show reasonably strong performance on unedited AI text, but accuracy drops significantly when that text has been edited, paraphrased, or blended with human writing. That’s a massive limitation in the real world, where almost no AI-generated content goes unedited.
The false positive problem is also serious — and frankly, alarming. Non-native English speakers and people writing technical content are being flagged as AI writers when they’re not. Think about what that means for a student submitting an essay or a professional applying for a job.
Here’s the core issue: a detector doesn’t actually “know” who wrote something. It estimates the statistical likelihood that the text matches patterns of machine-generated language. That’s a signal — not proof. And when the stakes are high, that distinction matters enormously.
So What’s the Alternative?
This is where provenance tools come in, and I think this is the more promising direction. Instead of trying to guess after the fact, provenance systems document how content was created in the first place — through metadata, watermarking, or verified workflows that show where a text originated.
Voluntary certifications of human authorship are part of this same shift. Rather than detection, the focus moves to disclosure and traceability. It’s a fundamentally different — and more honest — approach.
But provenance tools bring their own challenges:
- Will creators actually use them consistently?
- Can different platforms and systems talk to each other?
- What happens when someone opts out?
The Real Question Is About Trust
In an environment where AI can draft, edit, translate, and mimic style almost instantly, certainty is getting harder to find. For those of us working in public sector communications, this matters — whether you’re publishing content, reviewing submissions, or setting policy on AI use.
The real choice isn’t between blind trust and perfect detection. It’s between weak inference, better documentation, and human judgement. And right now, human judgement — informed by context and transparency — is still our most reliable tool.
If you want to build your AI literacy and skills, I can help
📚 Grab AI Playbook on Amazon, the perfect podcast companion.
📰 Subscribe to Human in the Loop, my LinkedIn Newsletter
🎓 Take the CPD Accredited AI Playbook Diploma
📞 Book a call at digitaltraining.ie
Sources
A human wrote this, believe me: Reuters Institute article by Marina Adami on AI detection tools and their limitations. https://reutersinstitute.politics.ox.ac.uk/news/human-wrote-believe-me
Between human and AI: assessing the reliability of AI text detection tools: PubMed study examining the reliability of AI text detection tools. https://pubmed.ncbi.nlm.nih.gov/38265047/
Can we trust academic AI detective?: PMC article on accuracy and limitations of AI-output detectors. https://pmc.ncbi.nlm.nih.gov/articles/PMC12331776/
Testing of detection tools for AI-generated text: Springer study on the performance of AI detection tools. https://link.springer.com/article/10.1007/s40979-023-00146-z
Key Statistics
– AI detector accuracy drops significantly when text has been edited, paraphrased, or mixed with human writing.
– False positives are a serious problem, particularly for non-native English speakers and technical writing.
– Some studies show reasonably strong detection performance on unedited AI text, but reliability falls in real-world conditions.
About the Podcast Host
AI and Communications Specialist | Author | Podcaster | International Speaker
Joanne Sweeney is a recognised leader in digital marketing and digital transformation, specialising in the practical application of AI to enhance communication strategies. The founder of Digital Training Institute, Joanne, empowers leaders and communications professionals to leverage AI responsibly and effectively, ensuring transparency, trust, and tangible results.
She completed the Harvard Kennedy School Executive Programme in AI and Public Policy this summer before launching her fourth book AI Playbook, in September. Joanne has also published two editions of Public Sector Marketing Pro, which were preceded by her first title, Social Media Under Investigation Law Enforcement and the Social Web.
Her books have become go-to resources for governments, public sector and policing agencies, as well as marketing and communications professionals, all navigating the challenges of the changing Internet.
Joanne is the host of the AI SIX Podcast, a daily six-minute podcast that provides actionable insights, expert interviews, and practical advice to help professionals navigate the new world of work.
