Gemini text detector

Gemini’s house style

Gemini answers are built to be skimmed. A typical response opens with a one-sentence summary, often in bold, then breaks the topic into headed sections with bulleted points, each starting with a bold label and a colon. Comparisons go into tables. It closes with a short wrap-up and, in the app, an offer to go deeper. When that structure is pasted into an email, a forum post or an essay, the scaffolding comes along: stray asterisks, pound-sign headings, and title-case section names on their own lines.

GPTTrace looks for exactly these artifacts. It highlights Markdown that didn’t render, title-case headings, “**Label:**” bullet patterns and the emoji bullets that Gemini and other assistants use in casual answers. These are formatting habits, not proof — some people write in Markdown — but in ordinary prose they are a strong indicator of copied chat output.

Beyond formatting

Strip the formatting and Gemini’s prose shares the general signs of AI writing: explanations that move in neat three-part steps, vague attributions like “experts note”, emphatic claims of significance and smooth transitions such as “Additionally” and “Furthermore”. Its encyclopedic tone is a particular challenge, because human-written reference text uses the same register — which is why GPTTrace weights the signs using measured data instead of treating each as damning.

Because Gemini is embedded in Google Workspace, a lot of its text arrives as lightly edited business writing: meeting summaries, project updates, customer emails. These are often short. Below about 150 words any detector, including this one, should be treated as a rough indication only.

Gemini images and audio

If the question is about a picture rather than a paragraph, Google labels its generated images with metadata that GPTTrace can read — see the Gemini image detector. For songs or voices from Google’s audio models, use the AI audio detector.

Checking AI text in work email

Many organisations now allow Gemini or similar assistants for drafting, so AI involvement in an email is not in itself a problem. Detection matters when authorship matters: a personal statement, a reference letter, a complaint presented as a customer’s own words, or a phishing email impersonating a colleague. For phishing in particular, the content of the request — urgency, payment, credentials — is a better warning sign than the writing style, and modern scams are often fluent precisely because they are written by AI.

For long internal documents, check a few sections separately. Mixed authorship is normal in collaborative work: one section may be drafted by Gemini and others written by hand, and checking the whole document at once averages them out.