AI detector checklist: does this text look AI-written?

A plain checklist for teachers, tutors and students who want to think clearly about a piece of writing. Tick what you notice and you’ll get a fair set of next steps.

Read this first: no tool or checklist can prove whether AI wrote something. AI detectors are unreliable, and they can wrongly flag honest work, especially from people writing in a second language. Use this to guide a conversation, never as evidence on its own.

Style

It doesn’t sound like the writer’s other work
A sudden change in vocabulary, sentence length or confidence compared with earlier pieces. This is often the most useful sign, but people do improve, and they get help from tutors, parents and editing tools.
It’s fluent but says little
Smooth, tidy paragraphs that circle the topic without a clear argument, specific example or opinion of the writer’s own.
Stock phrases and tidy patterns
Repeated framings like “In today’s fast-paced world”, “It’s important to note”, “delve”, or lists that always come in threes. Plenty of human writers do this too, especially when following a template.
Every paragraph has the same shape
Similar length, a topic sentence, three points and a neat summary line, over and over.

Content

No personal or local detail
Nothing that could only come from this person: a class discussion, a lab result, their town, something that happened to them.
It misses the actual task
Answers a more general version of the question, ignores a required source, or skips something discussed in class.
Confident errors or odd facts
Wrong dates, invented quotes, statistics with no source. AI tools can state errors with total confidence.

Sources

References that don’t check out
Citations that don’t exist, page numbers that don’t match, or real papers that don’t say what’s claimed. This is one of the few signs you can verify for certain.
Nothing recent, or oddly out of date
Missing a well-known recent development the writer would likely know about.

Process

No drafts or notes to show
The writer can’t show earlier versions, notes or version history. Not everyone keeps drafts, so treat this gently.
They can’t talk about it
In a relaxed conversation, the writer struggles to explain their main point or why they chose a source. Nerves can look like this too.
Formatting leftovers
Odd markdown symbols (** or ##), different fonts pasted in, or a phrase like “As an AI language model”.

What you’ve noticed

Tick any signs you’ve noticed. You’ll get a summary and some fair next steps here. There’s no score, on purpose.

Signs that point the other way

It’s worth looking for these too. They don’t prove a human wrote it either, but they should weigh on your thinking:

Specific personal, local or classroom detail that a chatbot wouldn’t know.
A clear voice that matches the writer’s other work, including their usual mistakes.
Notes, outlines or a version history that show the piece developing over time.
References that are real and actually support the point being made.

Why detector scores aren’t enough

AI detectors look for statistical patterns, such as how predictable each word is. Plain, careful writing can look “predictable” too, which is why formal essays, simple English and heavily edited work can be flagged. OpenAI withdrew its own classifier in 2023 (opens in new tab) because of its low accuracy, and a Stanford-led study found popular detectors misclassified many essays by non-native English writers (opens in new tab) as AI-generated. Running text through a paraphraser can also fool them easily. A score might start a conversation; it shouldn’t end one.

For students: protect yourself

Write in Google Docs or Word with version history on, and keep your notes.
Check your course rules. Many allow AI for brainstorming or proofreading but ask you to declare it.
If you use AI, say how. A short note is usually enough.

More on this in Best AI essay writing tools (with a section on academic integrity) and Best AI tools for students.

FAQ

Can an AI detector tell me for sure if text is AI-written?

No. Detection tools give a probability, not proof, and they get it wrong in both directions. OpenAI withdrew its own AI text classifier in 2023, citing its low rate of accuracy, and researchers have found that detectors flag writing by non-native English speakers as AI-generated more often than writing by native speakers.

Why doesn’t this checklist give a percentage?

Because any number we gave you would be invented. A checklist can help you notice things and decide what to look at next. It can’t measure authorship, and pretending otherwise would be unfair to the writer.

What should a teacher do if they suspect AI use?

Follow your school’s policy. In most cases that means looking at the writing process (drafts, notes, version history), comparing with in-class work and talking with the student before reaching any conclusion.

I’m a student and my work was flagged. What can I do?

Stay calm and gather evidence of your process: notes, outlines, browser history of your research and the document’s version history. Ask what the concern is based on, and offer to talk through your work. Keeping drafts from the start is the best protection.

Does this page store or send the text I’m checking?

You don’t paste any text here: it’s a checklist you tick yourself. Nothing is sent or saved.

Link to this tool

If it helped, you’re welcome to link to it from your blog, class page or newsletter. Here’s a snippet:

<a href="https://whataitool.ai/ai-detector-checklist/">AI detector checklist</a> by What AI Tool