As we all know, Deepseek is an AI writer that many students out there are using nowadays. However, can AI detectors like Originality.ai, Turnitin, GPTZero, or ZeroGPT actually detect it?
The short answer is YES. The longer answer is the devil lies in the details. Keep reading to know more about it.
Why Deepseek Gets Detected by AI Detectors?
The simple answer is: Deepseek is not specifically built to bypass AI detectors. If you look at their website you will not see any mention of bypassing Originality.ai, Turnitin, GPTZero, etc. as a feature. Hence, if something is not made to do that particular task, it simply won't be consistent in doing it.
Deepseek also has two variants – deepseek-chat and deepseek-reasoner. Both produce AI-generated texts that look quite human-like. AI detectors are made to spot these patterns through machine learning, so they often catch Deepseek content.
Detection By Deepseek Model
| Detector | Deepseek Model | Detection Rate (%) | Samples |
|---|---|---|---|
| GPTZero | deepseek-chat | 89.2 | 102 |
| GPTZero | deepseek-reasoner | 92.9 | 98 |
| Originality.ai | deepseek-chat | 100.0 | 117 |
| Originality.ai | deepseek-reasoner | 94.0 | 83 |
| Turnitin | deepseek-chat | 92.2 | 102 |
| Turnitin | deepseek-reasoner | 98.0 | 98 |
| ZeroGPT | deepseek-chat | 77.5 | 102 |
| ZeroGPT | deepseek-reasoner | 73.5 | 98 |
As you can see, Originality.ai and Turnitin do a great job at detecting Deepseek. GPTZero also delivers decent performance. ZeroGPT is still weaker in comparison.
Overall Detection Rates of Each Detector
| Detector | Overall detection rate* | Notes on scoring scale |
|---|---|---|
| Originality.ai | 97.5% | 0% = AI, 100% = human → judged “AI” at ≤ 10% |
| Turnitin | 95.0% | 0% = human, 100% = AI → judged “AI” at ≥ 20% |
| GPTZero | 91.0% | 100% = human → judged “AI” at ≤ 10% human-score |
| ZeroGPT | 75.5% | 0% = human, 100% = AI → judged “AI” at ≥ 50% |
*These percentages reflect how many times each detector flagged Deepseek as AI.
Overall ranking: Originality.ai > Turnitin > GPTZero > ZeroGPT.
Interesting Patterns
- Deepseek-chat is actually caught slightly more by Originality.ai (100% detection vs. 94% for deepseek-reasoner).
- Turnitin flags deepseek-reasoner a lot, with about 98%.
- ZeroGPT struggles for both variants but is especially weaker on reasoner.
So, Can AI Detectors Detect Deepseek?
Yes, they absolutely can! Most modern AI detectors are easily able to catch Deepseek, whether it is the chat variant or the reasoner one, and that too with high confidence.
If you are a teacher or a professor and you need near-certain identification, pair up Originality.ai or Turnitin with GPTZero. Usually, these two will never disagree on more than a handful of texts.
On the other hand, ZeroGPT alone doesn’t cut it. They let a quarter of the text slip through. While they could be useful, you should combine them with another AI detector if you want a more thorough approach.
Caveats & Next Steps
- Thresholds: We used standard cut-offs provided by each vendor. If you tweak them, the rates would shift a bit but you would still see the same ranking.
- Sample size: We used 200 total passages (≈100 each model). Different prompts or writing styles might yield different results.
- Evasion: Deepseek alone doesn’t do any specialized trick to fool detectors. If you want to bypass detection, use an AI humanizer like Deceptioner, which is built for that very purpose.
The Bottom Line
Deepseek can be detected by AI detectors. If you’re an instructor who relies on Originality.ai or Turnitin, you’re already in a good position to catch Deepseek’s outputs. GPTZero is also quite formidable. However, ZeroGPT is not so reliable.
If you absolutely want near-perfect detection, combine Originality.ai/Turnitin with GPTZero & you will cover almost everything. If you really want to bypass detection when using Deepseek, explore specialized rewriters like Deceptioner.
