# Fakespot vs MAGE, which AI detector is more accurate?

Source: https://toolbenchmarks.com/compare/fakespot-vs-mage/  
Tested: 30 October 2026. Publisher: Tool Benchmarks. Data licence: CC BY 4.0.

> Fakespot is the more accurate of the two, with an AUROC of 0.741 against 0.666 and a higher catch rate on AI text, 66.3% against 46.4%. MAGE flags less human writing, 19.0% against 34.9%, so it is the pick only if a false accusation worries you more than missed AI text. Both flag too much human writing to use as proof.

| Measure | Fakespot AI text detector | MAGE |
|---|---|---|
| Balanced accuracy | 65.7% | 63.7% |
| Catches AI text | 66.3% | 46.4% |
| Flags human text | 34.9% | 19.0% |
| Catches humanized AI text | 47.6% | 23.8% |
| AUROC | 0.741 | 0.666 |
| Flags human text: Academic abstracts | 12.0% | 12.0% |
| Flags human text: Q&A answers | 39.3% | 7.1% |
| Flags human text: Encyclopedia articles | 40.0% | 40.0% |
| Flags human text: Fiction | 12.5% | 4.2% |
| Flags human text: News reports | 70.8% | 33.3% |

## Where they make the same mistakes

Fakespot and MAGE are both trained classifiers, and both remove line breaks before they score a text, so formatting did not change either one's verdicts. On two kinds of human writing their false positives matched exactly. Each flagged 3 of 25 academic abstracts (12.0%) and 10 of 25 encyclopedia articles (40.0%). Wikipedia-style prose seems to look machine-made to both.

## Where Fakespot says AI and MAGE does not

The difference is how readily each one says "AI". [Fakespot](/ai-detectors/fakespot/) flagged 70.8% of human news reports, against 33.3% for MAGE, and 39.3% of Q&A answers, against 7.1%. Fiction showed the same pattern at a smaller scale, 12.5% against 4.2%.

That eagerness also catches more AI text. Fakespot had the higher catch rate in every genre and on every Claude model. On the hardest model to detect, Claude Fable 5.1, Fakespot caught 50.0% and MAGE 28.6%. On Claude Opus 5.5 and Sonnet 5.5, [MAGE](/ai-detectors/mage/) caught 45.2% each, under half.

The famous texts follow the same split. Fakespot flagged 13 of 30 public-domain classics and MAGE 6. MAGE left every novel opening alone, while Fakespot flagged several, including Moby-Dick and Jane Eyre. MAGE's flags fell on scripture, founding documents and political writing, among them the Declaration of Independence, which Fakespot passed.

## Short and edited texts

MAGE needs length to work. It caught 20.6% of AI texts under 200 words and 57.6% of those with 300 words or more. Fakespot also improved with length, from 43.3% to 76.6%, but it started from a higher base. After a humanizer rewrite of the Opus texts, Fakespot caught 47.6% and MAGE 23.8%.

Fakespot is also the faster of the two, at 0.07 seconds per 1,000 words against 0.39 for MAGE's Longformer model.

## Choosing between them

Fakespot is ahead on balanced accuracy (65.7% against 63.7%) and AUROC, mainly because it catches more. MAGE suits a reader who would rather miss AI text than accuse a person, but it still flagged 19.0% of human texts, so that caution has limits. If false positives are your main concern, neither is a good choice. [Fakespot vs Fast-DetectGPT](/compare/fakespot-vs-fast-detectgpt/) shows a detector that catches as much as Fakespot while flagging less human writing.

## Questions people ask

### Does MAGE have fewer false positives than Fakespot?

Yes. MAGE flagged 24 of 126 human texts and Fakespot 44. On academic abstracts and encyclopedia articles they flagged exactly the same share, 12.0% and 40.0%.

### Which is better on short texts?

Fakespot. Under 200 words it caught 43.3% of AI texts, while MAGE caught 20.6%.

### Do Fakespot and MAGE care about paragraph breaks?

No. Both clean their input in a way that removes line breaks, and both gave the same verdicts when we scored the texts as single paragraphs.