What You Will Learn

• Spot clusters of AI "tells" that indicate machine-written text
• Recognize AI-favorite words and uniform sentence rhythm
• Distinguish human specificity (dates, names, phrasal verbs) from generic AI prose
• Understand the limits of automated AI detectors and avoid false positives
• Edit or prompt AI to match your personal writing fingerprint

Study Guide

Introduction: The Skill You Didn't Know You Needed

Here's a strange thing that's happening everywhere right now. You open an email from a colleague and something feels off. You read a product review and you can't quite place why it sounds hollow. You skim a LinkedIn post and it's so polished, so perfectly structured, that you start wondering if a person actually wrote it. You're not imagining things. The text you're reading might have been generated by artificial intelligence.

This course is a field guide for the era we're living in. I'm going to teach you how to spot AI-written English text with a level of accuracy that surprises most people. Not by using a software tool or a website, but by training your eyes and ears to recognize the subtle patterns that large language models leave behind in every sentence they produce. These patterns are consistent, they're measurable, and once you learn to see them, you can't unsee them.

Why is this skill worth your time? Because you're already being exposed to AI text daily, whether you realize it or not. Student essays, job applications, customer reviews, news articles, internal company memos, dating profiles. AI is writing more of them than you might think. If you're an educator, you need to know. If you're a hiring manager, you need to know. If you're a communications professional, you need to know. If you're just someone who reads things and wants to know what's real, you need to know.

Now let's get into the actual mechanics.

The Foundational Ideas Before We Dive Into the Tells

There are three core principles you need to understand before we start analyzing text. Everything else in this course builds on these ideas.

Principle number one: think in clues, not proof.
There is no single word, no single sentence structure, no single test that proves a text was written by AI. A human can write "delve." A human can write "it's worth noting." A human can write a list of three things. The detection game works differently. You're looking for clusters. When five or more indicators show up in the same text, the probability that a machine wrote it becomes extremely high. One clue is interesting. Five clues is a smoking gun.

Principle number two: AI has an accent, and that accent is smoothness.
Think about how spoken accents work. You meet someone and within a few sentences you can often tell where they're from or where they learned English. AI has something similar. It's trained to produce text that is clear, polite, balanced, and safe. When every sentence is perfectly formed, when every statement is inoffensive, when every point is explained with the same thoroughness, the text develops a uniform quality. This isn't a flaw from the AI's perspective. It's the result of optimization. But this smoothness becomes a fingerprint. Just like a spoken accent reveals where a speaker comes from, the accent of mechanical polish reveals that text came from a machine.

Principle number three: the training data problem.
Large language models are trained on enormous amounts of text. Scientific papers, academic research, legal documents, tourism brochures, corporate reports. These sources have a distinct vocabulary and rhythm. The models absorb those patterns and reproduce them. This is why AI favors words like "delve," "tapestry," and "testament." These words exist in formal written registers far more often than in everyday human conversation. The AI isn't trying to sound robotic. It's just faithfully reproducing the patterns it was trained on.

Keep these three principles in mind. They're the foundation for everything that follows.

Tell #1: The Favorite Words

AI models have a statistically measurable preference for certain vocabulary. These aren't wrong words. They're not even unusual words. But they show up with a disproportionately high frequency in AI-generated text. When you start looking for them, you'll see them everywhere.

Here are the biggest ones:

"Delve" - meaning to explore or examine deeply.
"Let's delve into this report further." Human beings almost never say this in real life. They say "let's look into it" or "let's dig into this." But AI loves it. The word "delve" became roughly ten times more common in scientific papers after large language models became widely available. That wasn't because scientists suddenly changed their vocabulary. It's because AI was writing their papers.

"Tapestry" - meaning a mix or rich combination.
"The city is a tapestry of cultures." Nobody talks like this. You might read a poetic travel essay and see "tapestry" once in a decade. AI uses it daily.

"Boast" - meaning to have or feature.
"The hotel boasts three pools and a spa." Humans say "the hotel has three pools." The word "boast" in this sense is a tourism-brochure word that AI picked up from its training data.

"Testament" - meaning evidence or proof.
"Her success is a testament to hard work." Again, not something a normal person says in conversation. It's a formal, literary word that AI loves.

"Pivotal" - meaning crucially important.
"This is a pivotal moment." Humans say "this is a big deal" or "this matters." AI says "pivotal."

"Realm" - meaning area or domain.
"In the realm of politics." This is a cliché in AI writing.

"It's important to note" - a filler phrase.
"It's important to note that prices may vary." Why is it "important to note"? Why not just note it? Because AI uses hedging phrases to sound polite and objective.

"Navigate the complexities of" - meaning deal with complicated situations.
"We help clients navigate the complexities of tax law." This phrase is pure AI. No human has ever spoken these words in casual conversation.

Here's the caveat that matters. These words are not automatically bad. A human can use any of them. But when you see several of them clustering in one text, or when the same one appears repeatedly, your suspicion should go up. The word "delve" appearing twice in a paragraph is a massive red flag.

An example of what this looks like in practice:
"Let's delve into the realm of modern marketing. In today's world, brands must navigate the complexities of digital platforms. This is a pivotal moment for the industry. The success of any campaign is a testament to the team's ability to adapt. The market is a tapestry of diverse consumer behaviors."
That's six AI-favorite words in five sentences. No human writes like that.

Tell #2: Uniform Sentence Rhythm

Human writing has natural rhythm. A short, punchy sentence. Then a longer, meandering one. Then a fragment for emphasis. This variation mirrors the way we actually speak. We pause. We rush. We trail off. We emphasize.

AI doesn't do this. AI aims for balance and consistency. The result is a flat, uniform cadence where every sentence is roughly the same length. It feels like a metronome. Tick. Tick. Tick. The text is grammatically fine, but it's monotonous.

Example A - this is likely AI:
"The weather was lovely that morning. The birds were singing in the trees. The children played happily in the park and everyone enjoyed the warm sunshine."

Example B - this is likely human:
"Beautiful morning, the park was already full by 9:00. Kids everywhere, the birds going mad in the trees like they knew something we didn't. First proper sun of the year."

Do you see the difference? Example A has three sentences, all about the same length, all grammatically complete, all polite. Example B has a fragment ("Kids everywhere"), a longer sentence with a weird observation, and a short final line. There's also a colloquial phrase ("going mad in the trees") and a specific detail (9:00, "first proper sun").

Here's a practical technique. When you're reading a text, imagine the sentence lengths as lines on a chart. AI produces a flat, even line. Human text shows peaks and valleys. This visualization works surprisingly well.

Another example. AI version: "The company has experienced significant growth over the past year. The team has expanded to include new talent. The product has evolved to meet customer needs." Three sentences, identical length, identical structure. Human version: "We grew. A lot. New people, new roles, and honestly, a bit of chaos. The product went through a total redesign and somehow the customers stuck with us."

The human version has one-word sentences. It has "honestly." It has "somehow." It sounds like a person talking. AI text sounds like a person reading a speech they didn't write.

Tell #3: The Rule of Three

There's a reason people like things in threes. It's a rhetorical device that's been used for centuries. "Blood, sweat, and tears." "Lock, stock, and barrel." "Location, location, location." List of three feels complete and satisfying.

AI has learned this. But it's applied the rule so often that it's become a tell. AI will produce lists of three adjectives, three examples, three benefits, three reasons, three everything. Even when the natural number of items is two, or four, or five.

Here's a classic example:
"The course is comprehensive, engaging, and accessible for learners of all levels. It offers flexibility, structure, and support throughout the journey."

That's two lists of three in two sentences. The first sentence has three adjectives. The second has three nouns. This is the AI signature.

Compare with a human version:
"The course is genuinely good. Doesn't baby you and a real person answers your questions, usually the same day."

No list of three. No polished structure. Just an actual opinion with a specific detail.

Now, a single list of three is not suspicious. Humans use them all the time. But when you see multiple triads in a short text, especially in close proximity, that's a strong indicator. Three triples in three sentences? That's not human writing. That's a pattern that a model learned.

Tell #4: The Diplomat

AI has no lived experience. It has no opinions. It has no stake in any argument. It's trained to be helpful, polite, and harmless. So it hedges. It avoids taking a clear stance. It presents both sides of an issue and leaves the decision to you.

This diplomatic tone is a dead giveaway, because humans are not like this. Humans have opinions, biases, pet peeves, and preferences. We say things like "option two is better, trust me" or "don't do it, I made that mistake once." We commit.

Common diplomatic constructions that AI uses:
"It depends on your individual circumstances and goals."
"There are both advantages and disadvantages to this approach."
"The right choice ultimately depends on what you value most."
"It's worth considering both perspectives."
"Each option has its own unique benefits."

Here's what AI diplomat sounds like:
"It's worth noting that there are both advantages and disadvantages to this approach. The right choice ultimately depends on your individual circumstances and goals."

Here's what a human sounds like:
"Honestly, option two. Option one sounds nice on paper, but you'll regret it within a month. Ask Martha what happened when she tried that."

The human example is direct. It has a name. It has a consequence. It has a past event. AI can't reference "Martha" because Martha doesn't exist in its training data. It can't say "you'll regret it" because it doesn't know what regret is.

Here's another human marker that's interesting. The word "usually." Humans use "usually" all the time because our experiences are messy and inconsistent. "It usually rains in England." "The train is usually late." "She's usually pretty good at this." AI doesn't use "usually" because it doesn't have actual experience. It has statistical patterns, so it tends to make absolute or perfectly balanced statements. When you see "usually" in a text, especially in a context that sounds like personal experience, that's a sign of a human writer.

Tell #5: The "Not Just X, It's Y" Formula

There's a specific sentence shape that AI uses over and over. It's the contrastive formula. It looks like this:

"It isn't just X, it's Y."
"It's not about X, it's about Y."
"This isn't only X, it's also Y."
"Not only does X, but Y."

This formula is used to create a sense of depth and insight. It sounds like the writer is revealing a deeper truth. But AI uses it so frequently that it's become one of the most recognizable fingerprints of machine writing.

Example of the formula in action:
"Remote work isn't just a cost-saving measure, it's a rethink of how teams operate. The benefits aren't only financial. People report deeper focus and far fewer interruptions."

Now the human version:
"Remote work has been great for us, mostly. Deep focus days are unbeatable. Onboarding juniors, though, genuinely harder, and nobody's cracked it yet."

Notice that the human version doesn't use a dramatic contrast structure. It just says what it means, with a caveat. The human version also has a specific detail ("onboarding juniors") and a verb phrase that's natural ("cracked it").

A single use of "not just X, it's Y" is not conclusive. It appears in professional journalism and good persuasive writing. But when the formula shows up multiple times in one text, or when it's combined with other AI tells, it becomes a major fingerprint. Look for it. You'll start seeing it everywhere.

Tell #6: The Over-Explainer

AI assumes that the reader hasn't been paying attention. It introduces what it's going to say, says it, and then summarizes what it just said. It's a triple-layered redundancy that's extremely rare in human writing. Humans trust the reader. AI doesn't.

Here's the AI version:
"In summary, the key takeaway is that consistency beats intensity. This means that rather than working hard occasionally, you should show up regularly, which, as mentioned above, is the foundational foundation of progress."

Notice the "In summary" at the beginning. Then "the key takeaway is." Then "this means that." Then "as mentioned above." The idea is stated four times in four different ways. No human would do this. It's like a teacher who keeps saying "do you understand?" after every sentence.

Here's the human version:
"Show up regularly. That's it really. Consistency beats intensity every time and nobody wants to hear it because it's boring advice."

The human says it once, says it's boring, and moves on.

Markers to look for in this tell:
"In summary" followed by a restatement.
"This means that..." following the main point.
"As mentioned above" appearing close to the original statement.
The same idea repeated in slightly different words within a short passage.

When you see these patterns, the text is probably written by a machine that was designed to compensate for a reader who fell asleep halfway through.

Tell #7: Suspicious Perfection and Small-Print Details

This is where things get interesting. The seventh tell isn't about a single word or phrase. It's about a cluster of subtle indicators that relate to perfection and inconsistency.

First, the American English default.
Most AI models are trained primarily on American English. If you have a British colleague who has always written "colour," "organise," and "apologise," and one day they suddenly start writing "color," "organize," and "apologize," you know something is up. That email was almost certainly written or heavily edited by an AI. Australian English is even rarer as an AI default. If someone who normally writes "realise" and "the program" suddenly writes "realize" and "the programme," be suspicious.

Second, chatbot phrases.
Certain phrases from AI chat interfaces are leaking into emails, documents, and even printed receipts. Look for "Certainly," "Great question!" "I hope this helps!" and "Is there anything else I can assist you with?" These are phrases that come from the AI response interface. When you see them in a human communication, it usually means the person pasted the AI response without editing.

Third, flawless grammar and the absence of error.
No human writes perfect English. Everyone makes typos. Everyone omits a comma. Everyone sends lowercase text messages. Everyone occasionally produces a sentence that doesn't quite work grammatically. When a message arrives with absolutely perfect grammar, formal sentence structure, and impeccable punctuation, it's worth questioning whether a human wrote it. Especially if the sender doesn't normally write that way.

Here's a strange cultural development. Some people are now deliberately including spelling mistakes and informal abbreviations in their writing to prove they're human. "Definately" instead of "definitely." "TBH" and "gonna" in professional emails. This is a sign of the times. Perfection has become suspicious, and small errors have become a marker of authenticity.

Fourth, a sudden change in writing style.
This is the strongest single clue. If a colleague's messages suddenly become grammatically perfect, structured in polished paragraphs, and free of their characteristic quirks and typos, they've probably delegated their writing to an AI tool. People don't suddenly get much better at English overnight. They don't suddenly stop making their usual mistakes. When you see a dramatic change in someone's written voice, someone else is writing their words.

Real-World Examples: Applying the Framework

Now that you know the seven tells, let's practice. I'm going to show you four examples and walk through the analysis.

Example 1: A LinkedIn-style post that's AI-generated.
"10 years in software taught me one thing. Growth isn't just about talent, it's about consistency. In today's fast-paced tech landscape, you have to embrace challenges, seek feedback, and stay curious. Every setback is a lesson. Every lesson is an opportunity."

Let's analyze this. First, we have the "isn't just X, it's Y" formula: "Growth isn't just about talent, it's about consistency." Second, we have a rule of three: "embrace challenges, seek feedback, and stay curious." Third, we have "In today's fast-paced tech landscape," which is a classic AI opener. Fourth, the whole thing is aphoristic and generic. No specific company. No specific story. No specific numbers. This is AI.

Example 2: A social media post that's human-written.
"Hot take, stand-up should be async. Nobody needs to hear nine people say no blockers at 9:00 a.m. We moved ours to Slack in March and honestly, best decision this year. Meeting hours down 40%. I counted."

This is human. Why? Because it has a specific date ("March"). It has a specific number ("40%"). It has a personal reference ("I counted"). It has a blunt opinion without hedging. It has varied sentence lengths, including a fragment. And it mentions a specific detail ("nine people say no blockers") that no AI would invent.

Example 3: A restaurant review that's AI-generated.
"The restaurant offers a pleasant atmosphere and a diverse menu. The staff were friendly and attentive throughout our visit. While the prices are slightly higher than average, the overall experience justifies the cost. It's worth noting that weekend reservations are recommended."

Analysis: All the sentences are roughly the same length. The language is polite but lacks any real opinion. There's no mention of the food. The author doesn't say what they ordered. A human reviewer would almost always name a dish. They'd say "the lamb was overcooked" or "the pasta was amazing." They'd mention a specific interaction with a waiter. This review says nothing. It's AI.

Example 4: A workplace message that's human-written.
"Quick heads up. Staging is down again. Same Redis thing as last Tuesday. DevOps are on it. ETA 1 hour. We'll post here when it's back. 🙃"

This is human. It has shared history ("same as last Tuesday"). It has an emoji that carries emotional weight (the upside-down face). It has deliberately informal grammar. It's short and direct. No AI would write "Quick heads up" followed by a paragraph of formal explanation. The human writer knows exactly how to communicate in their context.

The Problem with Automated AI-Detection Tools

You might be wondering: why not just use an AI detector? There are tools online that claim to identify AI text. Here's the thing. Those tools are unreliable. Actually, that's an understatement. They're dangerously unreliable.

Here's what happened. A Stanford study found that automated detectors wrongly flagged most essays written by non-native English speakers as AI-generated. The reason is instructive. Non-native speakers who learned English from textbooks often produce grammatically correct, formally structured writing that shares stylistic features with AI output. They don't have the messy, idiomatic, conversational quality of native speakers. So the detector sees "too perfect" and decides it's AI.

The consequences of false accusations are severe. Students get accused of academic dishonesty. Job applicants get rejected based on an algorithmic guess. Professionals face reputational damage. And the tools cannot account for individual variation in writing style. Some humans just write very well. Some humans write formally. Some humans have learned English from textbooks and write exactly like AI defaults.

There's also a cat-and-mouse problem. Modern AI models can be prompted to sound substantially more human. If you tell the model to vary sentence lengths, avoid the rule of three, express opinions directly, and use phrasal verbs, the output is much harder to detect. As AI models improve, the gap between what detectors can identify and what AI can produce will continue to widen.

So what's the bottom line? Automated detectors should not be trusted as sole evidence of AI authorship. They're one signal among many. And their limitations should be communicated clearly to anyone using them.

How AI Can Be Instructed to Sound More Human

Understanding these tells also gives you the ability to do the reverse. If you're using AI as a drafting tool, you can instruct it to avoid its own tells. This is a legitimate practice for professionals who want to maintain a personal voice while using AI for efficiency.

Here are some prompts that work:
"Do not use these common AI words: delve, tapestry, testament, pivotal, realm, seamless, robust."
"Vary your sentence lengths significantly. Use short sentences, then long ones."
"Do not use lists of three. Use two or four items, or whatever is natural."
"Stop being diplomatic. Give a direct opinion without hedging."
"Avoid the formula 'not just X, it's Y.'"
"Do not repeat yourself or summarize what you already said."
"Use contractions, phrasal verbs, and informal register where appropriate."

You can also train AI to replicate your personal writing style. Provide samples of your writing, ask the model to identify patterns in your sentence length, vocabulary, and punctuation, and then match those patterns in future output. This is how you keep your voice while using AI.

But here's the deeper insight. The more people use AI and then edit it to sound human, the more human the AI output becomes. And the more human the AI output becomes, the harder it is to detect. This is an arms race. The tells I've described are accurate now, but they will shift as AI models evolve. You need to stay observant and keep updating your mental models.

Key Insights and Takeaways

Let me boil this down to the most important things to remember.

AI has a detectable accent of smoothness. The uniform quality of AI output is itself the primary identifying feature. When every sentence is perfectly constructed, every statement is balanced, and every point is explained with equal thoroughness, that smoothness is a red flag.

Multiple indicators, not single words, provide reliable evidence. A text showing five or more of the seven tells has a very high probability of being AI-generated. One clue is just one clue.

Specificity is the hallmark of human writing. Humans reference real dates, count things ("40%, I counted"), name dishes, mention colleagues, and share history. AI generates generic, representative examples. The more specific a text is, the more likely it's human.

Phrasal verbs are a human marker in English. In everyday speech, English speakers prefer phrasal verbs like "push back on," "show up," and "crack it" over single-word formal verbs like "disagree," "appear," and "solve." AI defaults to the formal register unless instructed otherwise. When a text is full of formal, single-word verbs and lacks phrasal verbs, that's a signal.

Imperfection is now a sign of authenticity. The deliberate inclusion of typos, slang, and grammatical informality has become a strategy for proving human authorship. In a world where machines produce flawless prose on demand, the small errors of human expression have become markers of being real.

Automated AI detectors are unreliable. They disproportionately penalize non-native English speakers. Their results should be treated with extreme caution.

Implications for Different Roles

What does this mean for you specifically? Let's break it down by role.

For educators and academic institutions: You need assessment methods that don't rely solely on detectors. Focus on process-based assessment. Drafts, oral presentations, in-class writing, and other authentic demonstrations of skill. Educate students about both the capabilities and the detectable patterns of AI writing. And recognize that non-native English speakers are particularly vulnerable to false accusations.

For professionals and employers: When reviewing writing samples or job applications, compare the text to the author's known style. If someone's writing suddenly changes dramatically, ask questions. Establish clear norms for AI use in workplace communication. There's nothing wrong with using AI for some tasks, but transparency matters. If a leader's public communications suddenly shift register, people will notice.

For language learners and educators: The stylistic features that make non-native speakers sound "AI-like" can be addressed through authentic language learning. Expose yourself to real-world English: podcasts, YouTube debates, sports commentary, casual conversations. Not just textbooks. Read aloud. Reading aloud improves language acquisition through the connection between oral articulation and language processing. And when you write, start from your natural speaking voice. Then tidy it up. Preserve your personality rather than polishing it into generic perfection.

For communication professionals: Authenticity is a valuable currency. The trend toward deliberate imperfection suggests that audiences value human unpredictability over mechanical polish. If you're drafting communications on behalf of a leader, be aware that sudden style shifts are noticeable. Even if people can't articulate what changed, they'll sense it.

Action Items: What to Do Now

Here's your plan of action. Seven things you can start doing today.

One: familiarize yourself with the seven tells. Review them. Practice identifying them in emails, social media posts, reviews, and news articles. The skill improves with practice.

Two: maintain a personal writing fingerprint. Whether you use AI or not, develop a distinctive writing style. Varied sentence lengths. Specific details. Natural idioms. This protects you from being mistaken for AI and strengthens your authentic voice.

Three: if you use AI for writing assistance, give it explicit stylistic instructions. Prompt it to avoid common AI words, vary sentence rhythm, minimize triadic lists, state opinions directly, and avoid over-explaining. Then review and edit all AI output before publishing.

Four: don't rely on AI-detection tools as definitive evidence. If you're in a position to evaluate authorship, treat detectors as one signal among many. Consider the possibility of false positives, especially for non-native speakers.

Five: when evaluating a text, look for clusters. A single AI tell proves nothing. Five or more tells in a short text is strong evidence. Always compare the text to the known writing style of the purported author.

Six: if you're a language educator, incorporate "real English" into your curriculum. Use authentic materials: conversations, debates, commentary, casual correspondence. Teach phrasal verbs explicitly. Encourage learners to read aloud and write from their speaking voice.

Seven: if you're in a position of authority, revise your academic integrity policies. Address AI use transparently, focusing on disclosure and ethical use rather than purely on detection and punishment.

Conclusion: The Messy, Specific, Personal Future

Here's the thing about this skill. It's not just about catching people who cheat on essays. It's about understanding what's happening to written communication in real time. AI is getting better at writing. The tells I've described will shift as models evolve. Some of them will disappear. New ones will emerge. The word preferences and patterns you've learned in this course are accurate right now, but they require periodic reassessment.

The deeper lesson is about what makes writing human. The rise of AI has reframed what "writing like a person" means. The growing trend toward deliberate imperfection, typos, slang, phrasal verbs, and unfiltered opinions, represents a cultural shift. In an era when machines produce flawless prose on demand, the small errors and idiosyncrasies of human expression have become markers of authenticity.

The messy. The specific. The personal. These are the qualities that are increasingly recognized as distinctively human. And they're worth preserving.

So here's what I want you to do. Start paying attention. Read the next email you get with fresh eyes. Look at the next LinkedIn post and ask yourself if a machine wrote it. Notice the rhythm, the word choices, the level of specificity. You'll be surprised at how quickly you start seeing the patterns.

Because that's the thing about AI text. It's not invisible. It's just smooth. And once you know what smoothness means, you'll never read the same way again.

Frequently Asked Questions

Introduction

This FAQ answers the real questions people ask when they're trying to tell machine-written text from human writing. The answers reflect what actually works in practice, not theory. Each response gives you something you can use immediately, whether you're reviewing a report, evaluating an essay, or scrolling through social media wondering if a post was written by a person or a program. Questions move from basic pattern recognition to advanced countermeasures, so you can jump to whatever level you need.

Fundamental Detection Signs

What is the most reliable indicator that a text was written by AI?

The most reliable indicator is the accumulation of multiple recognizable patterns rather than any single clue. Individual word choices or stylistic quirks can occasionally appear in human writing, but when several AI tells appear together,such as overused vocabulary, uniform sentence lengths, lists of three, diplomatic hedging, and formulaic transitions,the probability of AI authorship becomes extremely high. Think of these as clues that build a case rather than proof that seals one. When five or more of the patterns described in this document appear in a single text, the text was almost certainly generated by AI.
For a quick real-world example: if a LinkedIn post uses "delve," contains a rule-of-three list, and ends with "in summary," you're looking at a machine-written piece.

Which words does AI tend to overuse?

AI language models show strong statistical preferences for certain words that appear disproportionately frequently in their training data,primarily academic papers, scientific literature, marketing copy, and professional writing. Common AI-favored words include:

Delve
Meaning to explore deeply; the word appeared roughly ten times more frequently in scientific papers after ChatGPT launched. Tapestry
Used as a metaphor for a diverse mix or rich combination ("a tapestry of cultures"). Boast
As in "the hotel boasts three pools," frequently found in AI-generated hospitality or marketing content. Testament
As in "a testament to years of hard work." Pivotal
Used to describe something as critically important. Realm
As in "in the realm of politics." Robust
Describing something as strong or comprehensive. Seamless
Describing smooth, effortless processes or experiences.

The phrase "in today's fast-paced world" is an especially common AI opener that adds no genuine information. "It's important to note" is another filler phrase AI uses far more frequently than humans do.

Why does AI gravitate toward these particular words?

Large language models are trained on enormous corpora of text, with academic papers and professional documents heavily represented in the training data. Words like delve and testament appear frequently in academic writing, while words like boast and tapestry show up often in travel writing and marketing materials. The models learn statistical associations between context and word choice, and they tend to converge on these "medium-rare" vocabulary items,words that sound formal enough to be impressive but are common enough to fit many contexts. Additionally, AI is fundamentally trained to produce text that is clear, polite, and safe, which encourages the use of measured, non-confrontational vocabulary.
This is why you'll rarely see AI use strong, emotionally charged language or informal contractions in its default output.

What is the "sentence rhythm" or "metronome" test?

This refers to examining the length and variation of sentences in a text. Humans naturally write with varied sentence lengths,short, punchy sentences followed by longer, more complex ones, then a brief fragment. This irregularity creates a distinctive heartbeat or rhythm. AI, by contrast, tends to produce sentences of remarkably similar length because it optimizes for balance and completeness. A text where every sentence runs approximately 12-15 words with the same internal structure is far more likely to be AI-generated.
To test this, read the text aloud: if it sounds monotonously even, that uniformity itself is a tell. Compare "The weather was lovely that morning. The birds were singing in the trees. The children played happily in the park." with "Beautiful morning, the park was already full by 9:00. Kids everywhere, the birds going off their heads like they knew something we didn't." The second version sounds human because it breathes.

What is the "rule of three" and why is it a sign of AI writing?

The rule of three refers to AI's strong tendency to present information in groups of three,three adjectives, three examples, three benefits, three reasons. Humans do use triads naturally (think "blood, sweat, and tears"), and the pattern is rhetorically pleasing. However, AI models observed that humans respond well to threes and applied this pattern so consistently that it became a default behavior. AI will produce lists of three even when the honest number is two, four, or five.
When you see multiple instances of three-part structures in a single text ("comprehensive, engaging, and accessible"; "flexibility, structure, and support"),especially when a more natural count would be different,the text is likely AI-generated. A human writer would vary the pattern, sometimes listing two items, sometimes four, without even thinking about it.

What is a "linguistic fingerprint" and how does it help with detection?

A linguistic fingerprint is the unique set of writing habits that identifies an individual: their characteristic word choices, sentence length preferences, punctuation habits, common typos, and level of formality. Everyone has one, and it develops over years of writing. The most powerful detection method is the "sudden change" test: when a person's writing fingerprint suddenly transforms, something else is doing the writing.
For example, a colleague who always writes brief, casual emails with lowercase letters and missing commas suddenly sends a formally structured, perfectly spelled, multi-paragraph message. That change itself is the warning sign. People rarely improve their writing style overnight. The fingerprint concept also explains why AI text feels generic,it has no individual fingerprint because it's an average of millions of writers.

What is "register" and why does AI often get it wrong?

Register is the level of formality and type of language used in a given context for a given audience. A text message to a friend, a formal proposal, and an academic essay all use different registers. AI frequently gets the register wrong, being overly formal or overly polite for the situation. This register mismatch is one of the quickest ways to spot AI text.
If a friend replies to a request for Thai food with "That sounds wonderful. I would be delighted to join you this evening. Please let me know what time works best for your schedule," they're likely a robot. A real friend would say "Yeah, let's do it. What time?" The AI default is polite, formal, and complete,which is exactly what a human isn't in casual contexts.

Stylistic and Structural Tells

Why does AI avoid giving direct opinions, and what does "diplomatic hedging" look like?

AI models are trained to avoid offending users and to maintain neutrality, which results in an excessive use of balanced, non-committal language. The phrase "it depends" and its variations are hallmarks of this behavior. AI will frequently state that "there are both advantages and disadvantages" even when the situation clearly favors one side, and it will conclude with "ultimately, the right choice depends on your individual circumstances and goals" rather than delivering a definitive answer. Humans, in contrast, are generally willing to share their views directly,sometimes before the question has even been fully asked.
A text that never commits to a position, always seeking to balance every observation with its opposite, should raise suspicions. This is most noticeable in contexts like product reviews, opinion posts, or recommendation requests where a human would naturally express a preference.

What is the "not just X, it's Y" formula?

This is one of the most distinctive sentence shapes in AI-generated writing: a formulaic construction where the author claims that something isn't simply one thing but is actually a grander, more significant thing. Examples include:

"Remote work isn't just a cost-saving measure, it's a rethink of how teams operate."
"Growth isn't just about talent, it's about consistency."
"It's not about the money, it's about the mission."

This construction appears occasionally in human writing,particularly in journalism and opinion columns,but AI produces it with striking frequency, often multiple times within a single document. If you encounter a text using this formula more than once, AI authorship is highly likely.
The pattern is so common because AI training data contains many examples of this rhetorical flourish, and the model learned to overuse it.

Why does AI use em dashes so frequently?

Em dashes (the long dash used to set off parenthetical phrases,like this) appear in AI-generated text at rates far exceeding human usage. The reason may relate to training data that includes a great deal of punctuation-rich professional and editorial writing, but the result is unmistakable: AI text is peppered with em dashes in ways most human writers would never attempt. Interestingly, human writers have begun deliberately avoiding em dashes because using them now makes their writing appear AI-generated.
Even when an em dash would be grammatically appropriate, many professional writers now avoid it purely to sound more human. If you see three or more em dashes in a short piece of text, that's a signal worth noting.

What is AI "over-explaining," and how do you spot it?

AI-generated text frequently exhibits a particular structure: it announces what it will say, says it, then summarizes what it just said. You'll see phrases like "in summary," "this means that," "as mentioned above," and "the key takeaway is" stacked in close succession. A human writer trusts the reader to follow along and rarely restates points explicitly. AI, however, seems to operate on the assumption that the reader may have stopped paying attention, so it recaps itself unnecessarily.
A passage that repeats its central idea multiple times,once as a preview, once as the substance, and once as a summary,within a short span is characteristic of AI writing. This over-explaining makes simple ideas feel tedious and is a reliable tell when combined with other patterns.

How does punctuation reveal AI authorship in subtle ways?

Beyond em dashes, AI text tends toward unusually consistent punctuation usage. There is a lack of the idiosyncratic punctuation habits individual humans develop,the person who overuses ellipses, the person who loves semicolons, the person who writes entirely in lowercase. AI adheres to standard punctuation rules perfectly, which paradoxically makes the text feel sterile and detached.
The absence of any punctuation personality is itself a signal when other factors also point toward AI. A human writer will have quirks: a tendency to use exclamation points liberally, a habit of starting sentences with "And" or "But," or a preference for short fragments. AI text has none of these fingerprints because it averages out all punctuation habits into a perfectly standard norm.

What is stylometry and how does it relate to AI detection?

Stylometry is the quantitative study of writing style,measuring things like average sentence length, vocabulary richness, function word frequency, and syntactic patterns. It's been used for decades to attribute authorship of historical documents, including Shakespeare's plays. AI detection applies the same principles: measuring whether a text's statistical patterns match human norms or machine output.
For example, humans show natural variation in function word usage (words like "the," "and," "but") that is extremely difficult to replicate artificially. AI text tends to have more uniform function word distribution. While you don't need to run statistical analyses by hand, knowing that these measurable patterns exist helps you understand why certain texts feel "off" even when you can't pinpoint exactly why.

Regional and Contextual Indicators

Why is American English spelling a giveaway in non-American writing?

Most of the dominant AI language models (including ChatGPT, Claude, and others) are trained primarily on American English and default to American spelling unless explicitly instructed otherwise. This creates a detectable mismatch when a writer who normally uses British English (spelling words like colour, organise, and apologise) suddenly produces text with American spellings (color, organize, apologize). If a British colleague who has spelled "colour" with a "u" for the past five years sends an email with "color," that email was likely produced by an AI model operating at its default setting.
The same principle applies to Australian English, South African English, and other regional varieties,an AI defaulting to American English is a significant tell.

What common "chat phrases" reveal AI authorship?

Certain conversational formulas appear constantly in AI-generated responses, particularly when AI is used to reply to comments, customer inquiries, or emails. The most common include:

"Great question!"
(often appended even when the question was mundane) "I hope this helps."
"Certainly!"
as an opener "That's a wonderful question."
"Let me know if you have any other questions!"

These phrases reflect the AI's training to be helpful and polite, but humans rarely speak this way in casual or even professional correspondence. When a response feels like a polite customer-service script regardless of the context,especially if it's from a colleague or friend rather than a service provider,AI involvement is likely.
Real humans say "no problem" or "anytime" or just don't acknowledge the question at all.

What does "suspicious perfection" mean?

No native English speaker writes with perfect grammar, punctuation, and spelling at all times. Even highly educated professionals make typos, run-ons, or unusual word choices. AI text, in contrast, is grammatically flawless by design,every comma is placed correctly, every sentence is complete, every word is spelled correctly. If a text arrives with absolutely immaculate grammar and a formal register, particularly from someone who normally writes casually with lowercase letters, missing commas, and the occasional typo, that text was probably not written by that person.
The "too perfect" feel is a strong signal because imperfection is inherent to human writing. This is why some professionals now deliberately include typos in their messages,to prove they're human.