How to Spot AI Writing in 60 Seconds

The 60-Second Method at a Glance

Set a one-minute timer. Read the passage once at normal speed, then run these six passes. You are not grading grammar — machine text has flawless grammar. You are counting tells.

Those six passes map, not by coincidence, to the five things a good slop detector measures: vocabulary, cliché, structure, diversity, and substance. The manual version is slower and more forgiving. The automated one is instant — we will get to that. If you want each of these turned into a number you can compute yourself — em dash density, sentence-length variance, cliché counts — the companion reference on the signs of AI writing with reproducible thresholds gives all twelve.

1Substance: Does It Say Anything?

Start here, because it is the single most reliable tell and the hardest to fake. After reading a paragraph, try to state one specific, concrete thing you now know that you did not before. A name. A number. A date. A cause. A trade-off.

AI text is engineered to sound informative while committing to nothing. It gestures at authority — "studies show," "experts agree," "research suggests" — without a study, an expert, or a citation you could check. It describes the shape of an answer instead of giving one.

Read them again. The first paragraph survives being deleted; nothing is lost. The second cannot be deleted without losing the fact. That is the difference between filler and information.

When this signal lies: Genuine beginner or corporate writing can also be vague without any AI involved — a nervous intern padding a report reads the same way. Vagueness alone is a yellow flag, not a verdict. Weigh it with the passes below.

2Vocabulary: The Giveaway Words

Certain words barely existed in everyday prose and then, after late 2022, showed up everywhere. This is not folklore — it is measured. In a study of 14.2 million PubMed abstracts from 2010 to 2024, researchers Dmitry Kobak, Rita González-Márquez, Emőke-Ágnes Horvát and Jan Lause found that the vocabulary that spiked after ChatGPT's release was "almost entirely style words," not scientific terms. The word delves appeared at roughly 25 times its pre-ChatGPT frequency; showcasing and underscores jumped about nine-fold. Their conservative estimate: at least 10% of 2024 abstracts were written with LLM help, and up to 30% in some subfields (Kobak et al., 2024).

So scan for the usual suspects. A single one proves nothing. A cluster is loud.

If a 300-word post contains "delve into a multifaceted tapestry of robust, seamless solutions," you are done in ten seconds. Real people do not talk like that under deadline.

When this signal lies: Academics, consultants, and marketers used some of these words long before AI, and a writer who read the same viral "AI words" lists you did may now avoid them on purpose. Frequency and clustering matter more than any single word. Note, too, that after "delve" was called out publicly in early 2024, its use in new papers dropped — the tells drift as models and writers adapt.

3Clichés: The Empty-Phrase Count

Move from single words to whole phrases. AI leans hard on connective tissue that fills space without adding meaning. You have read all of these a thousand times:

• →Openers: "in today's fast-paced world," "in the digital age," "in a world where," "picture this."
• →Filler: "it's important to note that," "when it comes to," "at the end of the day," "let's dive in."
• →Fake authority: "studies have shown," "experts agree," "the data speaks for itself."
• →Pseudo-wisdom: "the key is to find balance," "true growth comes from within," "self-care isn't selfish."

One cliché is a lazy human. A pile-up — an opener cliché, then a filler phrase, then a fake-authority phrase, all in the same short passage — is the rhythm of generated prose. Count them. Three or more in a couple of paragraphs is a strong signal.

That is four clichés and zero information in one sentence. No human editor lets that through; a model produces it as a default.

When this signal lies: Speeches, ad copy, and motivational content lean on cliché by design and predate AI by a century. Judge clichés by density inside otherwise-ordinary writing, not by their mere presence.

4Structure: The Rule of Three and the Contrast Tic

Now look at the skeleton instead of the words. Two structural habits give models away.

The first is the rule of three on autopilot: everything arrives in tidy triplets. "Efficient, scalable, and reliable." "Plan, execute, and optimize." Humans use threes sometimes; AI reaches for them constantly, because balanced lists are statistically safe.

The second is the "it's not just X, it's Y" reversal and its cousins — "the result? Transformation." "But here's the thing." "It's not about the destination; it's about the journey." These are engagement-shaped templates. One is fine. When a piece keeps snapping into the same contrast frame paragraph after paragraph, you are watching a pattern generator, not a person thinking.

The human sentence is lopsided, specific, and slightly funny. The AI sentence is architecturally perfect and says nothing.

When this signal lies: Skilled copywriters use these exact structures on purpose, and they work. Structure is a supporting signal — pair it with a failed substance check before you lean on it.

5Diversity: Do the Sentences Breathe?

Read the passage for rhythm, ignoring meaning entirely. Human writing is uneven. We write a long, winding sentence that piles up clauses, and then a short one. We start with "But." We fragment. For emphasis.

AI writing tends toward uniform cadence: sentence after sentence landing in the same 15-to-20-word range, each a complete, balanced clause, each starting with a subject or a transition word. It reads smooth in a way that becomes, oddly, monotonous — the prose equivalent of a metronome. Watch especially for stacks of paragraph openers like "Furthermore," "Moreover," "Additionally," "Ultimately." Real writers vary how they connect ideas; models overuse a small set of formal transitions.

When this signal lies: Technical documentation, legal text, and translated writing are legitimately uniform and formal. Uniform rhythm outside those genres is the flag — not uniformity as such.

6Voice and Punctuation: Opinions, Mistakes, and the Em Dash

The last pass is about what is missing. Human writing carries a fingerprint: a real opinion someone could disagree with, a first-hand detail no search returns, an actual mistake, a joke that risks not landing. AI, tuned to be agreeable, hedges every claim and takes no position you could argue with. If you cannot find one sentence the author would defend in a fight, that absence is the tell.

Punctuation is the other half, and it needs care. Through 2025 the em dash — this mark — became the internet's favorite "AI tell," to the point that OpenAI's Sam Altman acknowledged the quirk in November 2025 and said ChatGPT would finally stop using em dashes when users asked it to (TechCrunch, via Techmeme). A high density of em dashes joining clauses is a real signal — but it is also where manual detection gets people hurt.

An em dash is not evidence. Plenty of careful human writers love it. Use it as one input among six, never as a ruling.

When this signal lies: All the time, on its own. Punctuation and voice are the passes most likely to produce a false positive, which is the whole reason the method above is six passes and not one.

Putting It Together: Two or More, Not One

Here is the rule that separates a fair read from a witch hunt: no single pass convicts. Any one signal — a couple of buzzwords, a rule of three, a dash — has an innocent explanation. What you want is convergence: three or four passes failing at once in the same short passage. Hollow substance plus a buzzword cluster plus a cliché pile-up plus metronomic rhythm is the fingerprint.

Notice what you measured across all six passes: vocabulary, cliché, structure, diversity, and substance. Whether a machine typed it never came up. You judged the writing on quality — the only judgment that holds up, because a lazy human and a careless prompt produce the same slop.

Do It Automatically in One Paste

The manual method is worth having — you cannot always paste a private email into a website. But when you can, running six passes by hand is slow, and your buzzword memory is not as complete as a dictionary of 200-plus flagged words and phrases.

That is what our detector automates. Paste any text and check it in the free, in-browser slop checker — nothing is uploaded, the analysis runs locally on your machine. You get a 0–100 score across the same five dimensions this guide walks through, with the exact words and phrases that dragged the score down highlighted in place, so you see why, not just a number. It is the six-pass checklist, run in a second, with the receipts shown.

To keep building the manual eye, the sister guide on how to spot AI slop across images, video, and text widens the lens to every medium, and the Slop Taxonomy names the flavors — Generic, Pseudo-Insight, Fake Authority — so you have a shared vocabulary for what you see.