How to Check If Something Was Written by AI

You cannot reliably tell whether a text was written by AI from one clue. The most dependable approach combines four things: a careful reading of the text, a check of its sources and facts, an AI detector result, and evidence about how the writer produced the work.

Each of these is imperfect on its own. Together they give a fair picture — and they protect honest writers whose work happens to look formal or polished.

These patterns are common in generated text, but every one of them also appears in human writing, so treat them as reasons to look closer.

• Smooth, evenly balanced paragraphs that say little that is specific to the topic, course or situation.

• Confident general statements with few concrete examples, numbers, names or personal observations.

• Repetitive transitions and summary phrases at the start or end of paragraphs.

• References that do not exist, point to the wrong work, or cannot be found at all.

• A noticeable change from the writer's earlier work in vocabulary, structure or level of detail.

• Answers that respond to a general version of the question rather than the specific prompt that was set.

Source checking is the most objective step. Look up a sample of the references: do they exist, are the authors and years right, and does the cited work actually support the sentence it is attached to? Generative tools can produce realistic-looking citations that do not exist, and a reference list that does not resolve is a factual finding rather than a judgement about style.

Do the same for specific claims, quotations and statistics. Accurate, traceable detail is a good sign; detail that cannot be traced deserves a question.

An AI detector estimates how closely the text matches patterns associated with generated writing. Check the full document rather than a short excerpt, read the passage-level highlights instead of only the overall percentage, and remember that editing, short length, formal style and second-language writing all affect the result.

Treat the output as a probability that tells you where to look. Detectors can flag human writing and miss edited AI text, so a result is the start of a review, not its conclusion.

The best evidence of authorship sits outside the final text. Version history in Google Docs or Microsoft Word shows whether the document grew gradually or appeared in large pasted blocks. Outlines, notes, annotated sources and earlier drafts show how the argument developed.

A short conversation is often the fastest test of all, and it is one of the main things teachers rely on when they suspect ChatGPT use. Someone who wrote the work can usually explain why they made their main argument, which source they found most useful, and what they would change.

Several common shortcuts lead to unfair or unreliable conclusions.

• Do not paste text into a chatbot and ask whether it wrote it. OpenAI notes that ChatGPT has no way of knowing whether content was AI-generated and may simply make up an answer.

• Do not rely on a single detector score, especially for short passages.

• Do not treat formal, polished or second-language writing as suspicious by itself.

• Do not reach a conclusion without giving the writer a chance to show their process.