Our own data
We tested an AI detector on 18 texts in three languages
Every detector round-up says the same thing: these tools are unreliable, treat the score as a signal not a verdict. Almost none of them show what that unreliability actually looks like. So we measured it — on a corpus where we know, for every single text, whether a machine or a person wrote it.
The result
Counting any score at or above 50% as a verdict of "AI", the detector was right on 6 of 18 texts. On a two-way question, that is worse than a coin toss.
The part that matters: language
For Polish and German the detector returned exactly 0.0% on every machine-written text — not "uncertain", but a confident verdict of "human written". At the same time it rated genuine human Polish encyclopaedia entries at an average of 74% AI. For Polish it is not unreliable so much as inverted.
| Language | AI caught | False accusations | Avg score on AI texts | Avg score on human texts |
|---|---|---|---|---|
| English | 2/3 | 2/3 | 45.4% | 69.0% |
| Polish | 0/3 | 2/3 | 0.0% | 73.8% |
| German | 0/3 | 1/3 | 0.0% | 33.3% |
Every text, every score
The full corpus, so you can judge the test rather than take our word for it. "Expected" is what we know to be true about each text; "score" is the raw percentage the detector returned.
| Text | Origin | Expected | Detector score | Verdict |
|---|---|---|---|---|
| English · 170 words | our article · en-are-we-in-an-ai-bubble.md | AI | 58.6% | correct |
| English · 170 words | our article · en-best-ai-detectors-2026.md | AI | 20.8% | wrong |
| English · 170 words | our article · en-biggest-ai-companies-2026.md | AI | 56.8% | correct |
| Polish · 170 words | our article · pl-czy-mamy-banke-ai.md | AI | 0.0% | wrong |
| Polish · 170 words | our article · pl-najlepsze-detektory-ai-2026.md | AI | 0.0% | wrong |
| Polish · 164 words | our article · pl-najwieksze-spolki-ai-2026.md | AI | 0.0% | wrong |
| German · 170 words | our article · de-ki-blase-2026.md | AI | 0.0% | wrong |
| German · 170 words | our article · de-beste-ki-detektoren-2026.md | AI | 0.0% | wrong |
| German · 170 words | our article · de-groesste-ki-unternehmen-2026.md | AI | 0.0% | wrong |
| English · 170 words | Wikipedia, revision 2020-05-31 · Baltic Sea | human | 100.0% | wrong |
| English · 170 words | Wikipedia, revision 2020-05-30 · Wind power | human | 40.9% | correct |
| English · 170 words | Wikipedia, revision 2020-05-25 · Lighthouse | human | 66.1% | wrong |
| Polish · 170 words | Wikipedia, revision 2020-04-08 · Morze Bałtyckie | human | 88.2% | wrong |
| Polish · 170 words | Wikipedia, revision 2020-04-05 · Latarnia morska | human | 100.0% | wrong |
| German · 170 words | Wikipedia, revision 2020-05-30 · Ostsee | human | 0.0% | correct |
| German · 170 words | Wikipedia, revision 2020-05-18 · Windenergie | human | 100.0% | wrong |
| German · 170 words | Wikipedia, revision 2020-04-11 · Leuchtturm | human | 0.0% | correct |
| Polish · 170 words | Wikipedia, revision 2020-05-16 · Wisła | human | 33.2% | correct |
How the corpus was built
The corpus is 18 texts of about 170 words each, six per language. Nine were written by a language model: they are excerpts from articles published on this site, so their origin is not a guess. The other nine are human-written, taken from Wikipedia revisions saved before June 2020 — well before the current generation of models existed, which is what makes them usable as a control. Each text was submitted once, and we recorded the raw percentage returned.
What this test does not show
- One detector, on its free tier, on a small sample. This is not a market ranking — it is a check on one claim.
- The paid tools (Sapling, GPTZero, Copyleaks) require creating an account, so they were not tested and we publish no numbers for them.
- Encyclopaedic prose reads differently from blog writing, and detectors may respond to that register differently.
- Wikipedia is heavily represented in model training data, which may itself affect how a detector scores it.
Using these results
Cite or reproduce this table with a link back to this page. The human half of the corpus comes from Wikipedia and is available under CC BY-SA; the machine half is our own published writing.