Can Turnitin Detect Gemini? What It Checks, and Why SynthID Isn't Part of It

Yes, for English submissions. Turnitin's own FAQ names Google Gemini models, from Gemini 1.0 Pro to Gemini 3.1 Pro preview, among those its English AI-writing detector can detect. The check doesn't depend on knowing which chatbot you used, because Turnitin scores patterns in the text itself.

What Turnitin does not do is read Google's hidden SynthID watermark. Its FAQ never mentions watermarks, and as of September 2026 Google offers no public way for anyone to check text for SynthID. Those are two separate systems, and most confusion about "Gemini detection" comes from mixing them up.

How Turnitin describes its detector

Turnitin's AI writing FAQ is fairly specific about the mechanics. Sentences are split into overlapping segments. Each segment gets a score between 0 and 1 for how likely it is to be AI-generated, and those scores are pooled into sentence scores and then a document percentage.

On what the model looks for, Turnitin says it "is not explicitly programmed to evaluate specific signals such as 'burstiness,' 'perplexity,' or other individual metrics." Instead, "it learns statistical patterns from our training data." Elsewhere the same FAQ says its classifiers are trained to detect differences in word probability between human and machine writing.

Two consequences follow:

The report doesn't name a model. The indicator shows the share of qualifying text that "AI writing tools or large language models may have generated or modified." Your instructor sees a percentage and highlighted passages, not "this was Gemini."
The question is about patterns, not brands. Turnitin doesn't need to know you used Gemini. It asks whether your prose looks like the machine writing it was trained on.

The usual limits apply to Gemini text as to any other. Turnitin needs at least 300 words of long-form prose, it ignores bullet points and lists, and it shows scores from 1% to 19% only as an asterisk. It also says the percentage "should not be used as the sole basis for action or a definitive grading measure by instructors."

Which Gemini models Turnitin names

As of September 2026, the FAQ says its English detector can detect content from seven Gemini models, named as Turnitin writes them:

Gemini-1.0-pro and Gemini-1.5-pro
Gemini-2.5-pro and Gemini 2.5 Flash
Gemini-3-pro-preview and Gemini-3-flash-preview
Gemini-3.1-pro-preview, which Turnitin dates to February 2026 and which is the newest Gemini entry

The same list covers GPT, o1, Claude, LLaMA, Mistral, DeepSeek, Nova and Grok models, and ends with "tools based on these LLMs as well." Three details are easy to miss.

It's an English list. The sentence begins "Turnitin's AI writing detection model for English submissions." When Turnitin launched its Spanish model in September 2024 and its Japanese model in April 2025, its release notes described both as trained to detect GPT-family output. No Gemini model is named for those languages.

Google has shipped newer models since. Google's Gemini API model list, checked on 27 September 2026, shows Gemini 3.5 Flash, 3.6 Flash, 3.7 Flash and Gemini 3.8 Flash, which the page labels a new stable model. None of these appears on Turnitin's public list. Turnitin says it is "continuously updating our model to detect these additional LLMs." Its release notes record a February 2026 update "to improve recall." The FAQ says a July 2026 change consolidated its "multi-model ensemble into a single model" while "maintaining a less than 1% false positive rate." Neither says which newer Gemini versions were in the training data.

Scores don't update by themselves. A submitted paper's AI percentage changes "only if it's re-submitted."

Because detection is pattern-based, a newer Gemini version isn't automatically invisible to Turnitin. It isn't automatically caught either. Turnitin simply doesn't publish per-model catch rates.

What SynthID is

SynthID is Google DeepMind's watermarking system. The DeepMind SynthID page explains the text version simply. A language model picks each next word from candidates with probability scores, and "SynthID adjusts these probability scores to generate a watermark." Readers can't see it, and Google says it doesn't affect quality.

The method was published in Nature on 23 October 2024 (Dathathri et al.). The authors ran a live experiment on roughly 20 million watermarked and unwatermarked Gemini responses. They found the difference in user ratings "negligible." They wrote that the watermark "has been productionized and is currently watermarking responses in Gemini and Gemini Advanced."

The key difference from Turnitin: a watermark check isn't guessing from style. Whoever holds the watermarking keys can score how closely a text matches the pattern those keys produce. Turnitin doesn't have Google's keys, and its FAQ describes no watermark check.

Anthropic says Claude's new text watermark "is a version of the SynthID-Text approach" (Anthropic). We covered who can check that one in does Claude watermark text?.

Does it apply to text you wrote with Gemini?

It depends on where the text came from, and Google is only explicit about one place.

The Gemini app and website: yes. DeepMind's page says, as of September 2026: "We've expanded SynthID to watermarking and identifying text generated by the Gemini app and web experience."
The Gemini API and AI Studio: not stated. The Google pages we checked don't say whether API text output carries SynthID.
Gemini in Google Docs: not stated. Google's help page for writing with Gemini in Docs doesn't mention watermarking.

Even where the mark exists, Google lists its limits. SynthID text watermarks survive "cropping pieces of text, modifying a few words, or mild paraphrasing." But they are "less effective on factual responses," and "detector confidence scores can be greatly reduced when an AI-generated text is thoroughly rewritten, or translated to another language" (Google's SynthID developer docs).

Can anyone check a SynthID text watermark?

Not publicly, as of September 2026. Here is what exists.

Gemini's built-in verification skips text. Google's verification help page says SynthID "can be used to verify whether images, videos, or audio files were created or edited by Google AI tools." Text isn't on that list.

The SynthID Detector portal isn't public. Google announced it on 20 May 2025 as a portal for uploading "an image, audio track, video or piece of text." It was rolled out to early testers, with a waitlist for journalists, media professionals and researchers. DeepMind's page now describes it as still being tested with journalists and media professionals.

Google's 2026 expansion is about media. On 19 May 2026 Google said SynthID checks for image, video and audio in the Gemini app had been used 50 million times. It said those checks were coming to Search and Chrome, and it launched a Cloud "AI Content Detection API" for businesses with a group of trusted partners (Google). None of it is described as a public text checker.

The open-source version can't read Gemini's mark. Google open-sourced SynthID Text in Hugging Face Transformers v4.46.0 in October 2024 so developers can watermark their own models. The launch post says each watermarking configuration "should be stored securely and privately," and "every watermarking configuration you use with your models needs to have a detector trained to recognize the mark" (Hugging Face). Without Google's configuration, that code can't check Gemini output.

"SynthID detector" sites can't verify text either. Some humanizer sites market SynthID "detectors" and "removers." One of them, Rephrasy, notes on its own page that "only Google can officially verify a SynthID watermark," and its upload box accepts JPG and PNG images (Rephrasy).

So neither your professor nor Turnitin can paste your essay into a public SynthID text checker. The practical risk remains the classifier your school already licenses.

Gemini in Google Docs: the context that matters more

For students, the more important change isn't the watermark. It's where Gemini now sits. Google's Docs help page describes Gemini drafting and refining text inside a document. Suggestions appear directly in the doc for you to accept one at a time or all at once, and an experimental option can mirror "the writing style of an existing file."

On 3 February 2026, Google made Gemini in Docs, including Help Me Write and the side panel, available to users aged 18 and over on Google Workspace for Education Plus or the Teaching and Learning add-on, at no extra cost (Workspace Updates). The features are on by default unless a school admin has disabled them.

So Gemini may be one click away in your school account even if your course bans AI. Once you accept a suggestion, it's ordinary document text, and Turnitin scores whatever you submit, wherever it came from. Our guide to humanizing Gemini text covers the list-heavy habits that make raw Gemini drafts easy to spot.

What to do if you used Gemini

Check the course rule, not the tool's availability. Gemini being switched on in your school's Docs doesn't mean your syllabus allows it.
If AI use is allowed, disclose it the way your course asks. Keep your prompts and drafts as a record.
Know how the Turnitin score works. It's instructor-only, it needs 300 words of prose, and anything from 1% to 19% shows as an asterisk. Our explainers on what AI percentage is acceptable and whether students can see the score go further.
Protect your version history. Google Docs lets anyone with edit access browse earlier versions. It also lets the file owner permanently delete that history (Google Docs help). Don't, if you may need to show how the paper came together.
Skip "SynthID remover" and text-watermark-checker products. Nobody outside Google can verify Gemini's text watermark, so neither claim can be checked.
If you want a rough pre-check, SupWriter's AI detector gives one model's score with sentence highlights. It isn't Turnitin, it doesn't read watermarks, and results vary by detector and text length.

If you're flagged for work you wrote yourself, process evidence matters more than any score. See how to prove your writing is your own.

FAQ

Can Turnitin detect Gemini 3.x text? Turnitin names Gemini-3-pro-preview, Gemini-3-flash-preview and Gemini-3.1-pro-preview for English submissions. Newer releases such as Gemini 3.8 Flash aren't on its public list as of September 2026. Its detector scores statistical patterns rather than identifying models, and Turnitin doesn't publish per-model detection rates.

Can Turnitin read Google's SynthID watermark? Nothing in Turnitin's AI-writing FAQ mentions watermarks. Turnitin describes a classifier trained on human and AI writing, not a watermark reader, and Google hasn't released a public way to check SynthID in text.

Can I check whether my Gemini text is watermarked? Not as of September 2026. Gemini's built-in verification covers images, video and audio. The SynthID Detector portal is limited to early testers, and the open-source SynthID Text code can't check Gemini output without Google's private configuration.

Does editing Gemini text remove the SynthID watermark? Google says the watermark survives mild paraphrasing and small word changes. Confidence "can be greatly reduced" when text is thoroughly rewritten or translated. That doesn't affect Turnitin, which doesn't use the watermark and checks for AI-paraphrased text in English.

Is text from Gemini in Google Docs treated differently? Not by Turnitin, which scores the submitted text regardless of source. Google doesn't say whether Docs output carries SynthID.

Check a Gemini-assisted draft with SupWriter's free AI detector before you submit: 300 words free, no credit card required.