You just read an article that felt… off. Perfectly structured, strangely generic, oddly confident about everything and specific about nothing. Was it written by AI? As synthetic content floods the internet — news sites, product reviews, student essays, social media posts — knowing how to check if an article was written by AI has become essential literacy. Here’s your practical guide to detection: the tells, the tools, and the limits of what you can actually determine.
I’ll be honest upfront: there’s no foolproof method. As AI writing improves, detection gets harder, and anyone promising 100% accuracy is selling something. But you can get good enough to catch most AI content most of the time — and more importantly, you can develop the critical reading habits that matter far more than any detection tool.
How to Check if an Article Was Written by AI: The Tells
Generic, hedged language. AI writing loves empty filler phrases like “it’s important to note” and “it’s worth mentioning.” It hedges constantly: “may,” “might,” “could,” “potentially.” It favors grandiose abstract nouns and over-polished transitions. Human experts are specific and opinionated; AI is comprehensive and noncommittal. If the article says everything and nothing simultaneously, be suspicious.
Perfect structure, zero soul. AI articles are often impeccably organized — intro, three sections, conclusion — with no rough edges, no digressions, no personality. Human writing has texture: an odd metaphor, a personal aside, a strong opinion, an imperfect transition. Perfection is the tell.
Vague specifics. “Studies show” (which studies?), “experts agree” (which experts?), “recent data indicates” (what data?). AI invents the shape of evidence without the substance. Real reporting names names, cites sources, provides numbers you can verify.
No original reporting or experience. “I tried this and…” / “When I interviewed…” / “In my twenty years…” — AI can’t do these. Articles without any first-person experience, original quotes, or on-the-ground detail are candidates for synthetic origin.
Repetitive phrasing and rhythm. AI falls into patterns: similar sentence lengths, parallel structures repeated endlessly, the same transitional phrases. Read a paragraph aloud — human writing has rhythm variation; AI writing often has a metronomic quality.
Confident errors. AI states false facts with total assurance — invented statistics, nonexistent studies, wrong dates. If you spot one confident falsehood, the whole piece becomes suspect. Verify a specific claim or two; it’s the fastest test.
Emotional flatness. AI can describe emotions but struggles to convey them. Articles about genuinely moving or infuriating topics that read like encyclopedia entries are suspicious. Human writers bleed onto the page, at least a little.
AI Detection Tools: What They Can and Can’t Do
Dozens of tools claim to detect AI writing: GPTZero, Originality.ai, Turnitin’s AI detector, and others. Here’s the honest assessment:
- They work best on long, pure-AI text. A 2,000-word article generated in one shot? Detectors catch it reliably. That’s their sweet spot.
- They struggle with short text. Under 300 words, accuracy drops dramatically. Don’t trust verdicts on paragraphs.
- They struggle with edited AI text. Human-edited AI writing — the most common real-world case — fools most detectors. Light editing drops detection rates sharply.
- They produce false positives. Non-native English writers, formulaic academic writing, and simple clear prose all trigger false AI flags. This has caused real harm — students falsely accused of cheating. Never treat a detector score as proof.
- They’re in an arms race they’re losing. As AI models improve, detection gets harder. The long-term trend favors generators, not detectors.
Use detectors as one signal among many, never as a verdict. A “98% AI” score on a long, generic, unsourced article is meaningful corroboration. A “60% AI” score on anything is noise. And always consider the source: an anonymous content farm publishing 50 articles daily is AI-generated regardless of what any tool says.
The More Important Skill: Critical Reading
Here’s what I actually want you to take away: detecting AI authorship is less important than evaluating content quality. The real questions aren’t “was this written by AI?” but:
- Who published this, and what’s their incentive? This question protected readers long before AI existed, and it still works.
- Does it cite verifiable sources? Real claims have real citations. Follow them.
- Does it align with trusted outlets? Extraordinary claims need corroboration from sources with editorial standards.
- Is it trying to make me feel or think? Content engineered for outrage or fear deserves extra skepticism — whether human or AI in origin.
These questions catch bad human writing too — which is the point. AI detection is a subset of media literacy, not a replacement for it. Our guide to how AI is changing news covers the newsroom side, and our fake image detection guide covers the visual equivalent. Together, they’re your synthetic-media defense kit.
The Cybersecurity and Infrastructure Security Agency publishes guidance on identifying synthetic media and misinformation that’s worth reviewing — this is recognized as a security issue, not just a curiosity.
There’s also a broader lesson here about information literacy in the AI age. As synthetic content multiplies, the skills that matter most aren’t technical detection tricks but classic critical thinking: who published this, what’s their incentive, does it cite verifiable sources, does it align with what trusted outlets report? These questions protected readers long before AI existed, and they still work. Teach them to kids, practice them yourself, and apply them especially to content that triggers strong emotions — that’s when skepticism matters most. The Cybersecurity and Infrastructure Security Agency publishes guidance on identifying synthetic media and misinformation that’s worth reviewing. Detection tools will keep evolving, but critical thinking is the one detector that never becomes obsolete.
The Bottom Line
To check if an article was written by AI: look for generic hedged language, perfect-but-soulless structure, vague unsourced claims, absence of original reporting or experience, repetitive rhythms, and confident errors. Use detection tools as corroborating signals, never as proof — they struggle with short text, edited text, and produce false positives. But prioritize the deeper skill: critical reading. Who published it? What’s their incentive? Are claims sourced and verifiable? Those questions predate AI and outlast every detection tool. In an internet filling with synthetic content, the goal isn’t to identify every AI article — it’s to never be fooled by a bad one, whoever (or whatever) wrote it.
