Benchmarking 4 AI Detectors on 1,000 Texts: Why False Positives Matter More Than Accuracy

Matthew Chen — Fri, 22 May 2026 19:05:15 +0000

I’ve been testing AI detectors recently because they are increasingly used in schools, publishing workflows, hiring, and content moderation.

Most AI detector comparisons focus on one headline number: accuracy.

But after running a benchmark across 1,000 English texts, I think that may be the wrong metric to obsess over.

Dataset

The benchmark included:

500 human-written texts
500 AI-generated texts
AI samples generated from 13 different models
Mixed text lengths: short, medium, and long passages

I tested four AI detectors:

GPTHumanizer
GPTZero
ZeroGPT
Sapling

The surprising part

The most important question was not:

Which detector catches the most AI?

It was:

Which detector is least likely to falsely accuse a human writer?

That distinction matters a lot.

In a real school, workplace, or publishing environment, a false positive is not just a bad prediction. It can become an accusation.

Results

Detector	Overall Accuracy	Human False Positive Rate
GPTHumanizer	98.0%	0.0%
GPTZero	98.7%	2.2%
ZeroGPT	88.2%	18.4%
Sapling	88.6%	19.4%

GPTZero achieved the highest overall accuracy in this benchmark.

But GPTHumanizer had the lowest human false positive rate.

ZeroGPT and Sapling were much more aggressive, but that also meant they mislabeled more human-written text as AI.

Why this matters

If you are using an AI detector for low-stakes filtering, raw accuracy might be useful.

But if you are using it to judge students, writers, job applicants, or employees, false positives should probably be treated as the most important metric.

A detector that catches slightly more AI but wrongly flags real human writing may be more harmful than a conservative detector that avoids false accusations.

My takeaway

AI detectors should not be used as final proof.

At best, they should be used as weak signals alongside human review, writing history, source drafts, editing patterns, and context.

Full benchmark and methodology:
https://www.gpthumanizer.ai/blog/2026-ai-detector-benchmark

Curious how others evaluate AI detectors: would you prioritize raw accuracy, AI recall, or human false positive rate?

Any good advice using openclaw?

Matthew Chen — Thu, 26 Mar 2026 06:12:57 +0000

I’ve been using OpenClaw for about 30 days now. Not every single day, but fairly frequently. I’ve also spent some time building my own custom skills tailored to my workflow.

That said, I still feel like the overall efficiency isn’t where I expected it to be.

Right now, it feels like OpenClaw can get things about 60% done, especially for programming tasks. It’s usable, but not quite good enough to rely on fully. As a developer, I often find that tools like Cursor or other coding assistants can produce better results with less friction.

On the other hand, for non-technical users, I’m not even sure what the “killer use case” is. It doesn’t feel obvious what they should use it for, and the value isn’t immediately clear.

At the moment, I mainly use it to:

Check my website sales data

Track daily user behavior / analytics

But that feels pretty limited compared to what I hoped it could do.

So I’m curious, for those of you who use OpenClaw regularly:

What are your main use cases?

Have you found any workflows that really make it shine?

Any tips on building better skills or improving output quality?

Would love to hear how others are actually getting value out of it.