ChatGPT detector
Paste the text. See which sentences carry ChatGPT’s habits – and which carry a human’s.
The scan runs in your browser. Nothing is uploaded unless you ask for the AI second opinion.
Your result appears here: an overall score, the rhythm trace of your sentences and every sentence lit by how many AI-typical features it carries.
ChatGPT’s habits, and why they happen
ChatGPT is trained to be helpful, balanced and inoffensive, then tuned by human raters. That training leaves a recognisable house style:
- The framing sandwich. An opener that restates the question (“In today’s fast-paced world…”), balanced middle paragraphs, and a closer beginning “In conclusion” or “Ultimately”.
- Favourite words. “Delve”, “tapestry”, “pivotal”, “underscore”, “navigate the complexities”, “a testament to”. A 2024 study of millions of PubMed abstracts found “delves” appearing many times more often after ChatGPT’s release than before.
- Lists of three. “clear, concise, and engaging”. Rhetorical triplets everywhere.
- Hedged balance. “While X has benefits, it also presents challenges.” Rarely a strong, specific opinion.
- No lived detail. No names, dates, places or mishaps unless the prompt supplied them.
None of these is proof on its own; plenty of people write in exactly this register, especially in school essays that teach “firstly, furthermore, in conclusion”. That’s why the detector shows you where and why, rather than just a number.
For the statistics behind it – perplexity and burstiness – see how AI detectors work. For a human checklist, read how to tell if something was written by AI.
Questions people ask
Can you detect ChatGPT-written text?
You can detect patterns that ChatGPT tends to produce – even rhythm, stock vocabulary, formulaic transitions – and this tool highlights them sentence by sentence. You cannot prove ChatGPT wrote something from the text alone, and lightly edited output can pass any detector.
Does it work for GPT-5, Claude and Gemini too?
The signals are shared by most large language models because they all generate likely next words, so the detector responds to all of them. Newer models vary their style more, which makes every detector less reliable over time.
What words does ChatGPT overuse?
Research on scientific abstracts after 2022 (Kobak et al., 2024) found sharp rises in words like “delve”, “showcasing”, “underscores”, “intricate” and “pivotal”. The detector counts these and similar phrases such as “it’s important to note”.
Is ChatGPT detection free here?
Yes, with no sign-up. The core check runs in your browser; the optional AI second opinion has a fair-use limit.
The rest of the kit
Your text follows you between tools, so you can paste once and check everything.