Home / Guides

Is this AI written? Signs of AI-generated text

People who read a lot of AI text start to notice a certain voice. Wikipedia's volunteer editors, who deal with a flood of AI-written drafts, have written down the patterns they see most. This guide walks through the most useful ones in plain English, and explains why none of them proves anything on its own.

Updated October 11, 2026 · By the IsThisAI? team

A word of caution first

Wikipedia's editors are blunt about the limits of their own list. They describe it as observations, not rules, and note that not all text with these signs is AI-generated, since language models learned from human writing in the first place. They also warn against relying on your own judgment too much, citing research that people are often no better than chance at telling AI text from human text (Wikipedia).

So treat everything below as a reason to look closer. Formal writers, people writing in a second language, and anyone following a template can show some of these patterns naturally. Research has shown detectors themselves flag non-native English writers far more often (Liang et al., 2023), and human readers can make the same mistake.

Telltale phrases

Puffing up importance

One of the clearest patterns is text that keeps telling you how significant something is. Wikipedia lists phrases like "stands as a testament," "plays a pivotal role," "underscores its importance," "marking a key turning point," and "the evolving landscape" (Wikipedia). The editors explain why: models drift toward generic, important-sounding statements and away from the specific, unusual details a human expert would include. A small town becomes "a vibrant hub"; a minor inventor becomes "a titan of industry."

Overused "AI vocabulary"

Certain words show up far more often in text written after chatbots became popular. Wikipedia's list includes "additionally" at the start of a sentence, "crucial," "delve," "enhance," "fostering," "intricate," "showcase," "tapestry," "testament," "underscore," and "vibrant." The guide notes that the favorite words change over time; "delve" was a famous ChatGPT tell in 2023 and early 2024 but became much less common later (Wikipedia). One of these words means nothing. A paragraph stuffed with several of them is more suspicious.

Vague attributions

Watch for claims pinned on nobody in particular: "experts argue," "observers have noted," "industry reports suggest," especially when no source is named or only one is cited (Wikipedia).

Chatbot leftovers

Sometimes the giveaway is text that was meant for the person chatting, not the reader: "I hope this helps," "Certainly!", "Let me know if you'd like more detail," or unfilled placeholders like "[Your Name]." Wikipedia also documents leftover citation codes from specific chatbots that appear when someone copies an answer straight in (Wikipedia). These are about as close to proof as this list gets.

Sentence patterns

• Negative parallelisms. Constructions like "It's not just about X, it's about Y" or "not a mirror but a portal." Wikipedia calls these a stereotypical AI sign, while noting humans use them too (Wikipedia).
• The rule of three. Lists of exactly three adjectives or phrases, used over and over to make shallow points sound thorough.
• Avoiding "is." Instead of "The gallery is the museum's exhibition space," AI text tends to say "serves as" or "functions as," and prefers "features" or "boasts" over plain "has."
• "-ing" add-ons. Sentences that end with a tacked-on phrase about significance, like "...highlighting its role in the community" or "...contributing to the region's development."
• Formulaic endings. A closing section that starts "Despite its challenges..." and ends on vague optimism about the future.

Uniform rhythm and missing specifics

Turnitin explains that language models tend to pick the next word in a "consistent and highly probable fashion," while human writing "tends to be inconsistent and idiosyncratic" (Turnitin FAQ). In practice that shows up as paragraphs where every sentence is a similar length and shape, and where nothing surprising happens.

The bigger tell is often what's missing. Human writing about real experience tends to include specific names, numbers, dates, odd details, opinions, and the occasional tangent. AI text on the same subject is often smooth, positive, and general enough that it could describe almost anything in the category.

Formatting quirks

Wikipedia's editors also note formatting habits common in chatbot output (Wikipedia):

• Heavy use of bold text for emphasis, often in a "key takeaways" style
• Bulleted lists where each item starts with a bold mini-heading and a colon
• Headings in Title Case Where Every Word Is Capitalized
• Emoji in front of headings or bullet points
• Markdown symbols (like ** or ##) showing up in places that don't use Markdown
• Curly quotation marks, though the guide notes Word and phones add these automatically, so they prove little

Check it instantly

Spotting these by eye takes practice and is easy to get wrong. Our free detector combines two open-source AI-detection models with a set of style checks adapted from Wikipedia's guide, and highlights the sentences that read most like AI. It's a second opinion, not a verdict.

If you're checking a student's work, read our guide for teachers first. If your own writing got flagged, see why human writing gets flagged.

More guides

Sources

1. Wikipedia contributors (WikiProject AI Cleanup), "Wikipedia:Signs of AI writing" (accessed Oct 11, 2026), CC BY-SA 4.0. https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
2. Creative Commons, Attribution-ShareAlike 4.0 International license. https://creativecommons.org/licenses/by-sa/4.0/
3. Turnitin, "Turnitin's AI writing detection capabilities FAQs" (Turnitin Guides, updated Oct 7, 2026). https://www.turnitin.com/products/features/ai-writing-detection/faq
4. Liang, Yuksekgonul, Mao, Wu & Zou, "GPT detectors are biased against non-native English writers," Patterns 4(7), 2023. https://www.cell.com/patterns/fulltext/S2666-3899(23)00130-7

This guide is general information, not legal or academic-policy advice. AI detection results, including ours, can be wrong.