AI-generated text is getting better, but it still has tells. If you know what to look for, you can often spot it without a tool. Here are the specific patterns that AI detectors flag, with examples.

1. The hedging opener

AI loves to start essays with qualifications that add nothing:

"In today's rapidly evolving digital landscape, it is important to note that artificial intelligence has fundamentally transformed the way we approach education."

Count the filler: "rapidly evolving digital landscape" (cliche), "it is important to note that" (unnecessary), "fundamentally transformed" (vague). This sentence could be replaced with "AI is changing education" and lose nothing.

The human version: A human might start with a specific observation, a question, or even a complaint. "I spent three hours trying to grade essays last week and couldn't tell which ones were human."

2. The diplomatic balance

AI text never takes a strong position. It always acknowledges both sides:

"While AI offers tremendous potential to enhance learning outcomes, it is crucial that educational institutions develop robust frameworks to navigate these challenges."

"While X, also Y" is AI's favourite sentence structure. It sounds thoughtful but says nothing. Real writers have opinions.

The human version: "Universities need AI policies. Most don't have one. The ones that do are usually unworkable."

3. Mechanical transitions

AI uses transition words as structural scaffolding, not as actual transitions:

"Furthermore, the integration of AI tools into classroom environments has created both opportunities and challenges. Moreover, the implications extend beyond the classroom. Additionally, stakeholders must consider..."

"Furthermore", "Moreover", "Additionally" — three paragraphs, three mechanical transitions. Each one could be deleted with no loss of meaning.

The human version: Humans often don't use transition words at all. We rely on logical flow between ideas, or we use conversational links: "But here's the thing...", "That's not even the worst part...", "Which brings me to..."

4. The vocabulary cluster

Certain words appear 10-50x more often in AI text than in human writing:

Delve — humans almost never use this word in normal writing

Multifaceted — a favourite AI adjective for anything complex

Leverage — used as a verb ("leverage technology")

Tapestry — metaphorical use ("the tapestry of human experience")

Holistic — "a holistic approach"

Paradigm shift — AI's go-to phrase for any change

Seamlessly — "seamlessly integrate"

Robust — "robust frameworks"

Navigate — "navigate challenges"

Foster — "foster collaboration"

Any one of these is fine. Three or more in a single essay is a strong signal.

5. Uniform sentence length

This is the subtlest but most reliable pattern. Read this:

"Artificial intelligence has transformed educational practices significantly. Teachers must adapt their methods to accommodate new technology. Students benefit from personalised learning experiences. Institutions face challenges in implementing these systems effectively."

Four sentences. Word counts: 7, 9, 7, 9. Almost identical length. Human writing varies much more — a 3-word sentence followed by a 30-word one.

The human version: "AI's changed everything. It's in the classroom now, whether teachers like it or not, and honestly most of them are still figuring out what to do about it. Some love it. Most are terrified."

6. Paragraph uniformity

AI doesn't just write uniform sentences — it writes uniform paragraphs. Look at the length of each paragraph in a suspected AI essay. If they're all roughly the same length (3-5 sentences, 60-100 words), that's a pattern.

Human writing has long paragraphs, short paragraphs, one-sentence paragraphs, and paragraphs that go on too long before the writer realises and moves on.

7. The generic conclusion

AI conclusions summarise without adding anything:

"In conclusion, it is evident that artificial intelligence presents both opportunities and challenges for the educational sector. By developing comprehensive strategies and fostering collaboration, institutions can navigate this evolving landscape effectively."

"In conclusion" + summary of previous points + vague recommendation = AI conclusion.

The human version: A human conclusion adds a final thought, makes a prediction, asks a question, or takes a position. "Until universities actually test their AI policies on real students, these frameworks are just theory."

8. No personal voice

This is the hardest pattern to quantify but the easiest to feel. AI text has no personality. No frustration, no excitement, no humour, no tangents. It's competent but lifeless.

Read any flagged passage and ask: "Could I guess who wrote this from the writing alone?" If the answer is no — if the text could have been written by anyone — that's a strong indicator.

What to do with this knowledge

For teachers: These patterns give you a vocabulary for the conversation. Instead of "the detector flagged your essay", you can say "I noticed four paragraphs in a row all the same length, and three uses of 'furthermore' — can you tell me about your writing process?"

For students: If your own writing accidentally includes these patterns, you know what to revise. Add some variety to your sentence lengths. Cut the filler phrases. Add a personal observation or two.

Is It AI? flags these patterns automatically and explains each one — paste any text and see what it finds.

What AI Writing Looks Like: The Patterns Detectors Find