Which AI Models Our AI Detector Recognises
The detector tells whether a text was written by a person or generated by AI, and shows which fragments look generated. It recognises texts from today's models:
- ChatGPT: GPT-6 Sol, GPT-6 Luna, GPT-5.6 and earlier versions;
- Claude: Opus 5, Sonnet 5.5;
- Gemini: 3.8 Flash, 3.5 Flash Lite, 2.0;
- DeepSeek: V4 Pro, V4.1 Flash;
- GigaChat, Grok, Qwen, Kimi, GLM.
How Accurate Is It?
We tested the detector on thousands of texts it had never seen during training: texts written by people before ChatGPT appeared, and texts generated by AI models.
| AI model | Detected as AI |
|---|---|
| ChatGPT (GPT-6 Sol, GPT-6 Luna, GPT-5.6 and earlier versions) | 97.7–100% |
| Claude (Opus 5, Sonnet 5.5) | 100% |
| Gemini (3.8 Flash, 3.5 Flash Lite, 2.0) | 99.5–100% |
| DeepSeek (V4 Pro, V4.1 Flash) | 100% |
| GigaChat | 98.3% |
| Grok 4.7, Qwen 3.8, Kimi K3, GLM 5.3 | 100% |
| All AI models | 98.8% |
| Texts by people | Recognised as human |
|---|---|
| English | 98.8% |
| Spanish | 98.4% |
| Portuguese | 97.7% |
| German | 98.5% |
| French | 98.5% |
| Italian | 98.0% |
| Russian | 100% |
| All texts by people | 99% |
Presumption of innocence. A false accusation of AI use can harm a writer, so the detector is built above all not to misjudge people: it recognises 99% of human texts as human. Essays, articles, news, reviews, classic literature, the Bible and constitutions are not mistaken for AI. New AI models appear all the time: texts from models the detector has not met yet are recognised in about 95% of cases, and we add new models regularly. The result shows a probability, not a verdict.
Languages
The detector checks texts in English, Spanish, Portuguese, German, French, Italian, Russian, Ukrainian, Polish, Turkish, Bulgarian, Chinese, Arabic, Urdu, Indonesian and other languages.
Interactive Analysis: Understanding the AI Score
The tool does more than give a "Human or AI" verdict: it marks the result right in your text. Click any highlighted fragment to see how confident the detector is about it.
The final score combines two figures: the share of clear AI generation and the share of suspicious text. Colours help you assess the risk at a glance:
- Red: high probability of AI; these parts affect the score most.
- Yellow: suspicious text; these parts affect the score moderately.
Mixed Texts
Many authors write the introduction and conclusion themselves and leave the main part to AI. The detector assesses the text part by part and highlights exactly the generated passages.
Tip: to rule out both AI generation and copying from the web, use the AI detector together with our plagiarism checker.
The detector flags your text, but you wrote it yourself
The detector estimates a probability; it does not deliver a verdict. An even, textbook style without personal details sometimes looks like AI writing. If the result raises questions from a teacher or a client, the best proof of authorship is how the text was written:
- Version history. In Google Docs open “File → Version history”, in Word for files on OneDrive “File → Info → Version History”. It shows how the text appeared and changed day by day.
- Drafts and notes: the outline, notes from sources, early versions of chapters.
- Sources: the reference list and the pages the facts and figures come from.
- Talking about the work. The author can explain why the topic was chosen and how the results were obtained, and retell any part in their own words. This convinces more than any percentage.
If you edited some passages with the help of AI, say so in advance: many universities and clients allow such help when it is disclosed.
To have the writing history ready in advance, keep a writing diary: the service saves versions with the exact time and shows a reviewer how the text grew day by day.
Limits and Plans
| Plan | Characters per check | Features |
|---|---|---|
| Guest | Up to 10,000 | Standard queue; daily check limits apply. |
| Registered user | Up to 15,000 | More checks per day. |
| PRO | Up to 1,000,000 | Priority processing without queues or frequency limits. Checks use the characters of your plagiarism pack. |
Frequently Asked Questions (FAQ)
Do I still need to check for uniqueness?
Yes. The AI detector identifies only the origin of a text, human or machine, but does not look for matches with content already published on the web. A text can be fully human-written and still be copied. For complete content security, use both the AI detector and the plagiarism checker.