This page publishes the raw outputs behind AscendProse's AI detector benchmarks. It is a transparent dataset, not a universal accuracy leaderboard. Detector outputs are probabilistic and should not be used alone for academic, hiring, or disciplinary decisions.
Download the raw CSV dataset · Download JSON
Raw Detector Outputs
Showing all 45 recorded observations.
| Detector | Sample | Displayed AI probability | Directly comparable? | Recorded output |
|---|---|---|---|---|
| Grammarly AI Detector | AI mixed sample | 0.0% | No | 0% appears AI-generated; 100% no AI text patterns found |
| Grammarly AI Detector | AI résumé sample | 0.0% | No | 0% appears AI-generated; 100% no AI text patterns found |
| Grammarly AI Detector | Human-written excerpt | 0.0% | No | 0% appears AI-generated; 100% no AI text patterns found |
| QuillBot AI Detector | AI mixed sample | 100.0% | Yes | Model v7.1.0: 100% likely AI |
| QuillBot AI Detector | AI résumé sample | 100.0% | Yes | Model v7.1.0: 100% likely AI |
| QuillBot AI Detector | Human-written excerpt | 0.0% | Yes | Model v7.1.0: 0% likely AI, 100% human-written |
| Sapling AI Detector | AI-generated editorial sample (ai-editorial-01) | 100.0% | Yes | Fake probability 100.0%; 374 tokens |
| Sapling AI Detector | AI-generated editorial sample (ai-editorial-02) | 100.0% | Yes | Fake probability 100.0%; 388 tokens |
| Sapling AI Detector | AI-generated marketing sample (ai-marketing-01) | 100.0% | Yes | Fake probability 100.0%; 311 tokens |
| Sapling AI Detector | AI-generated marketing sample (ai-marketing-02) | 97.2% | Yes | Fake probability 97.2%; 348 tokens |
| Sapling AI Detector | AI-generated technical sample (ai-technical-01) | 99.8% | Yes | Fake probability 99.8%; 354 tokens |
| Sapling AI Detector | AI-generated technical sample (ai-technical-02) | 100.0% | Yes | Fake probability 100.0%; 388 tokens |
| Sapling AI Detector | AI mixed sample | 73.6% | Yes | Fake probability 73.6% |
| Sapling AI Detector | AI résumé sample | 100.0% | Yes | Fake probability 100.0% |
| Sapling AI Detector | Human-written excerpt | 0.9% | Yes | Fake probability 0.9% |
| Sapling AI Detector | Human-written editorial sample (human-editorial-01) | 1.1% | Yes | Fake probability 1.1%; 318 tokens |
| Sapling AI Detector | Human-written editorial sample (human-editorial-02) | 100.0% | Yes | Fake probability 100.0%; 312 tokens |
| Sapling AI Detector | Human-written marketing sample (human-marketing-01) | 0.3% | Yes | Fake probability 0.3%; 300 tokens |
| Sapling AI Detector | Human-written marketing sample (human-marketing-02) | 0.5% | Yes | Fake probability 0.5%; 305 tokens |
| Sapling AI Detector | Human-written technical sample (human-technical-01) | 95.6% | Yes | Fake probability 95.6%; 291 tokens |
| Sapling AI Detector | Human-written technical sample (human-technical-02) | 0.5% | Yes | Fake probability 0.5%; 297 tokens |
| Undetectable AI Detector | AI-generated editorial sample (ai-editorial-01) | 99.0% | Yes | Your text is AI generated; 99% AI / GPT |
| Undetectable AI Detector | AI-generated editorial sample (ai-editorial-02) | 99.0% | Yes | Your text is AI generated; 99% AI / GPT |
| Undetectable AI Detector | AI-generated marketing sample (ai-marketing-01) | 99.0% | Yes | Your text is AI generated; 99% AI / GPT |
| Undetectable AI Detector | AI-generated marketing sample (ai-marketing-02) | 99.0% | Yes | Your text is AI generated; 99% AI / GPT |
| Undetectable AI Detector | AI-generated technical sample (ai-technical-01) | 99.0% | Yes | Your text is AI generated; 99% AI / GPT |
| Undetectable AI Detector | AI-generated technical sample (ai-technical-02) | 99.0% | Yes | Your text is AI generated; 99% AI / GPT |
| Undetectable AI Detector | Human-written editorial sample (human-editorial-01) | 1.0% | Yes | Your text is human written; 1% AI / GPT |
| Undetectable AI Detector | Human-written editorial sample (human-editorial-02) | 1.0% | Yes | Your text is human written; 1% AI / GPT |
| Undetectable AI Detector | Human-written marketing sample (human-marketing-01) | 1.0% | Yes | Your text is human written; 1% AI / GPT |
| Undetectable AI Detector | Human-written marketing sample (human-marketing-02) | 1.0% | Yes | Your text is human written; 1% AI / GPT |
| Undetectable AI Detector | Human-written technical sample (human-technical-01) | 1.0% | Yes | Your text is human written; 1% AI / GPT |
| Undetectable AI Detector | Human-written technical sample (human-technical-02) | 1.0% | Yes | Your text is human written; 1% AI / GPT |
| ZeroGPT AI Detector | AI-generated editorial sample (ai-editorial-01) | 78.4% | Yes | Your Text contains mixed signals, with some parts generated by AI/GPT; 78.4% |
| ZeroGPT AI Detector | AI-generated editorial sample (ai-editorial-02) | 100.0% | Yes | Your Text is AI/GPT Generated; 100% |
| ZeroGPT AI Detector | AI-generated marketing sample (ai-marketing-01) | 28.7% | Yes | Your Text is Most Likely Human written, may include parts generated by AI/GPT; 28.7% |
| ZeroGPT AI Detector | AI-generated marketing sample (ai-marketing-02) | 83.3% | Yes | Your Text is AI/GPT Generated; 83.3% |
| ZeroGPT AI Detector | AI-generated technical sample (ai-technical-01) | 100.0% | Yes | Your Text is AI/GPT Generated; 100% |
| ZeroGPT AI Detector | AI-generated technical sample (ai-technical-02) | 100.0% | Yes | Your Text is AI/GPT Generated; 100% |
| ZeroGPT AI Detector | Human-written editorial sample (human-editorial-01) | 8.0% | Yes | Your Text is Human written; 8% |
| ZeroGPT AI Detector | Human-written editorial sample (human-editorial-02) | 0.0% | Yes | Your Text is Human written; 0% |
| ZeroGPT AI Detector | Human-written marketing sample (human-marketing-01) | 0.0% | Yes | Your Text is Human written; 0% |
| ZeroGPT AI Detector | Human-written marketing sample (human-marketing-02) | 0.0% | Yes | Your Text is Human written; 0% |
| ZeroGPT AI Detector | Human-written technical sample (human-technical-01) | 0.0% | Yes | Your Text is Human written; 0% |
| ZeroGPT AI Detector | Human-written technical sample (human-technical-02) | 0.0% | Yes | Your Text is Human written; 0% |
What the Results Show
The independently labeled subset produced the following displayed-score averages. These are observations from this sample set, not estimates of universal accuracy.
- Sapling AI Detector: mean displayed AI probability 33.0% on 6 human samples and 99.5% on 6 AI samples.
- Undetectable AI Detector: mean displayed AI probability 1.0% on 6 human samples and 99.0% on 6 AI samples.
- ZeroGPT AI Detector: mean displayed AI probability 1.3% on 6 human samples and 81.7% on 6 AI samples.
Human-Written Texts With High AI Scores
Among provenance-checked human samples in the recorded independent runs, the examples below received at least one displayed AI score of 90% or more. This is a post-test way to locate notable disagreements, not a predeclared accuracy threshold or a population false-positive rate.
| Human source and sample ID | Sapling AI Detector | Undetectable AI Detector | ZeroGPT AI Detector |
|---|---|---|---|
human-editorial-02A Landholder, I., in Essays on the Constitution of the United States 1787–1788 essays; digitized by Project Gutenberg | 100.0% | 1.0% | 0.0% |
human-technical-01The History and Practice of the Art of Photography 1849 technical manual | 95.6% | 1.0% | 0.0% |
These highlighted sources are historical public-domain writing. Their style may differ substantially from modern submissions; do not generalize these scores to current student, hiring, or editorial documents. Different interfaces also need not calibrate their percentages identically. The practical lesson is to verify provenance before acting on a score.
Historical rows remain available and are marked when they are not directly comparable; newer experiments do not overwrite them.
Methodology
在同一日期测试 Sapling、QuillBot 与 Grammarly 的公开 AI 检测器。先从 Sapling 的三个内置样本提取一组人类文本、AI 混合文本和 AI 简历文本，并把相同文本输入 QuillBot。Grammarly 对短文本未返回有效分数，因此仅为满足其有效扫描长度而重复同一文本至 285 至 309 词；记录各界面显示的 AI 概率、样本长度、模型版本（可见时）和截图。比较指标为各检测器 AI 概率及同一样本的最大分歧百分点。 使用 6 份有来源证明的人类公共领域文本和 6 份保留模型、提示词及生成日期的 AI 文本，逐一输入Sapling AI Detector、ZeroGPT AI Detector、Undetectable AI Detector的公开检测界面，共 36 次。每次保存原始显示结果、测试时间、输入文本 SHA-256 和截图；不改写、重复或截断样本。
Limitations
样本来自 Sapling 厂商内置示例，不是独立盲测语料；Grammarly 样本经过机械重复，不能与另外两款形成严格同条件准确率排名，重复本身也可能改变检测结果。当前结果只证明这些公开界面在该日期、该输入上的输出差异，不能证明任何检测器的总体准确率，也不应单独用于学术、招聘或纪律决定。 公开网页可能随模型版本、地区、登录状态或产品更新而变化；12 个样本不足以代表所有语言、体裁与长度。结果只描述记录日期的界面输出，不应用于惩罚、招聘或学术裁决。
Excluded Public Interfaces
These targets were not assigned scores:
- Grammarly AI Detector: public_ui_free_quota_signin_gate
- QuillBot AI Detector: public_ui_cloudflare_verification
- Smodin AI Detector: public_score_visually_blurred_behind_paid_insights
Official Product Sources
Reuse and Citation
You may cite individual results with a link to this page and the test date. Do not describe the dataset as a complete accuracy ranking. The downloadable files preserve the exact product, sample description, displayed probability, output text, and comparability flag.
AI Assistance Disclosure
AI assisted with research organization and record formatting. The detector values were recorded from public product interfaces and were not invented or statistically extrapolated.