Verdict: In this small saved set, 6 of 6 AI-generated inputs and 2 of 6 provenance-checked human excerpts received displayed AI scores of at least 90%. Use the result as a screening signal, not proof of authorship. We cannot infer a general accuracy or false-positive rate from this sample.
What We Actually Tested
We submitted the same independently labeled set of editorial, marketing, and technical excerpts to three public detector interfaces. This review focuses on Sapling's recorded outputs. The exact inputs, labels, dates, full outputs, and screenshots are retained in our experiment archive; the public research dataset shows the recorded results and methodology, not every archived input or screenshot. The older three-example check from Sapling's own interface remains in that dataset but is not mixed into the independent table below.
| Saved sample and provenance | Known label | Genre | Sapling displayed AI score |
|---|---|---|---|
ai-editorial-01 | ai | editorial | 100.0% |
ai-editorial-02 | ai | editorial | 100.0% |
ai-marketing-01 | ai | marketing | 100.0% |
ai-marketing-02 | ai | marketing | 97.2% |
ai-technical-01 | ai | technical | 99.8% |
ai-technical-02 | ai | technical | 100.0% |
human-editorial-01Source text | human | editorial | 1.1% |
human-editorial-02Source text | human | editorial | 100.0% |
human-marketing-01Source text | human | marketing | 0.3% |
human-marketing-02Source text | human | marketing | 0.5% |
human-technical-01Source text | human | technical | 95.6% |
human-technical-02Source text | human | technical | 0.5% |
Where Human Texts Received High AI Scores
In this saved set, 2 of 6 provenance-checked human excerpts received displayed AI scores of at least 90%. This threshold identifies examples worth investigating; it was not preregistered and is not a population false-positive rate.
human-editorial-02: 100.0% displayed AI probability; A Landholder, I., in Essays on the Constitution of the United States.human-technical-01: 95.6% displayed AI probability; The History and Practice of the Art of Photography.
These highlighted sources are historical public-domain English texts. Their style may not represent present-day student, hiring, or editorial submissions. A high detector score cannot overturn the documented authorship of a source text.
How to Interpret This Result
The displayed percentage is an interface output, not a calibrated probability that a person used AI and not a verdict about misconduct. This limited set was assembled for reproducible comparison, not sampled to represent all modern writing. Interface behavior and model versions can change after the test date. Different detectors' percentages also need not mean the same thing, so a cross-product score comparison is descriptive rather than a universal ranking.
The most useful purchase decision here is whether your team needs a recorded screening step with human follow-up. If a workflow would automatically reject a student, applicant, or author based on one score, this evidence argues against using Sapling—or any detector—that way. Preserve the original document and revision history; check citations and context; let the author respond before a consequential decision.
What This Review Does Not Claim
- It does not claim Sapling is the most or least accurate detector.
- It does not estimate accuracy, precision, or false-positive rates for the wider population.
- It does not compare current prices or plan limits; verify those on the official site.
- It does not substitute a detector score for documented provenance or editorial judgment.
For the sample construction, all raw scores, and known exclusions, see the methodology and dataset. Product source: Sapling AI Detector. This page describes recorded product-interface output, not vendor-reported accuracy. AI assisted with organizing the saved evidence into this review.