How to tell if something was written by AI
By Sourabh Singh11 October 20266 min read
Look for stock phrases, flat rhythm and vague detail, then check the text with a detector that shows which sentences it flags.
You can often tell if text was written by AI from a handful of patterns that show up again and again in ChatGPT, Claude and Gemini output. An AI detector can confirm what you see, but the patterns themselves are the most useful thing to learn.
8 signs text was written by AI
- Stock openers and transitions: 'In today's fast-paced world', 'Moreover', 'Furthermore', 'It is important to note'.
- Inflated vocabulary: 'pivotal', 'multifaceted', 'testament to', 'navigate the complexities'.
- Even rhythm: most sentences and paragraphs are almost the same length.
- Vague examples where a person would name something specific.
- Balanced hedging on every point, with no clear opinion.
- Tidy summary endings: 'In conclusion' or 'Ultimately', restating what was just said.
- Lots of em dashes or perfectly parallel lists of three.
- Facts or citations that look plausible but can't be found.
Want the full list? Every AI phrase and sentence signal we flag, with what to write instead.
See all signsHow AI detectors check text
Detectors estimate how predictable each word is given the ones before it, and how much sentence length and structure vary. AI text tends to be highly predictable and evenly paced; human writing is lumpier. A good detector shows which sentences drove the score, not just a percentage.
Why one score isn't enough
Detectors disagree with each other and produce false positives, especially on formal writing and on non-native English. Treat a high score as a prompt to read the flagged sentences closely, and look for other evidence such as drafts or version history before drawing conclusions about a person's work.
Paste any text to see which sentences read as AI, and why.
Check if it's AICommon questions
Not from the text alone. Detectors and patterns give strong hints, but only process evidence, like drafts and version history, comes close to certainty.
Common ones include 'delve', 'pivotal', 'multifaceted', 'tapestry', 'testament', 'moreover' and 'it is important to note'.
Usually, yes. They look for patterns shared across language models rather than one model's fingerprint.
Not on its own. Em dashes are common in AI text, but plenty of people use them. Look for several signs together.
