A familiar ritual now plays out online. Someone reads a bland article or a suspiciously smooth email and asks: was this written by AI? Detection tools are consulted, telltale phrases are listed, and a verdict is delivered. Then everyone moves on, having learned nothing about whether the thing was any good.
Updated September 2026: we added links to Google’s guidance on AI content and to the evidence on how unreliable AI detectors are.
I think we are asking the wrong question.
Readers never cared about the tool
Nobody asks whether an article was typed in Word or Google Docs, or whether the author used spellcheck, a thesaurus or an editor. We judge writing on what it delivers: Is it accurate? Does it tell me something useful? Can I trust where the information came from?
AI does not change those standards. It changes how easy it is to produce text that fails them.
The real problem is cheap, empty text
The genuine worry behind “is this AI?” is volume. It has never been cheaper to produce large amounts of plausible, generic writing that says nothing new, repeats what is already everywhere and sometimes gets facts wrong. That was possible before AI, through content farms and copy-paste rewriting. AI simply removed the cost.
Google’s guidance for site owners makes the same point. Its policies target content made primarily to manipulate search rankings, whatever produced it, while stating that appropriate use of AI is not against its guidelines. The test is helpfulness and quality, not authorship method.
Detection is a dead end
AI text detectors are unreliable. OpenAI withdrew its own detector in 2023, citing its low rate of accuracy. They produce false positives, and research has found they are more likely to flag writing by non-native English speakers; in one Stanford study, seven detectors labeled 61% of essays by non-native speakers as AI-generated. Students have been wrongly accused. As models improve and people edit AI drafts, the signal gets weaker still. Building policy on a test that cannot reliably tell the difference is unfair and ultimately futile.
Better questions to ask
- Is it accurate? Are claims supported, and can I check them?
- Is it original? Does it add experience, analysis or information I cannot find everywhere else?
- Who stands behind it? Is there an accountable person or organization?
- Is it honest about process where that matters, such as research, journalism and education?
That last point deserves emphasis. Transparency still matters. A student submitting work in a course that bans AI assistance is breaking a rule, and a news outlet should tell readers how it uses automation. But the purpose of disclosure is to protect trust, not to shame a tool.
Where AI genuinely helps writers
Used thoughtfully, AI can help people research faster, structure messy ideas, tighten sentences and catch errors. It can help non-native speakers express expertise they already have. None of that makes writing worse. Publishing whatever the machine produces, unchecked and unedited, does.
The standard we hold ourselves to
At Token & Tell, the question we ask of every piece is whether it is accurate, useful and clearly explained, and whether our editors stand behind it. That standard would be the same whatever tools existed. It is also the only one readers have ever really cared about.
Sources
- Google Search’s guidance about AI-generated content, Google Search Central, February 2023
- New AI classifier for indicating AI-written text, OpenAI, updated July 2023
- AI detectors biased against non-native English writers, Stanford HAI, 2023
