# MeteGPT

> AI text humanizer paired with an AI detector, built around one idea: humanize, then prove it. Paste AI-generated text from ChatGPT, Claude, or Gemini, get a corpus-trained rewrite, and check the result against a real detector before you submit. The engine is trained on a 2,590-essay / 5M+ word corpus. Free tier: 4 humanizations per day at 125 words (no signup).

Founded by Fırat Mıhcı (applied-linguistics researcher, ResearchGate: https://www.researchgate.net/profile/Firat-Mihci). Honest-ceiling policy: the one first-party performance figure MeteGPT publishes is a measured, dated detector eval of its own humanized output (GPTZero 4%, QuillBot 30/30 clean, under 20% on Turnitin as the hardest checker; ~30 academic passages, May 2026), always carried with its caveats (out-of-distribution text can spike the strict checkers to 30-60%; small sample; a measurement, not a guarantee) and its full method on the methodology page. No fabricated numbers. Every claim about a competitor tool or detector is a dated, linked, third-party citation, never an invented number.

## Core Tools

- [AI Humanizer](https://metegpt.com/): Paste AI text, get a corpus-trained rewrite, with a detector check run on the result before you submit. Free tier, no signup. (The homepage is the humanizer tool.)
- [AI Detector](https://metegpt.com/detect): Free AI-detection check, no account. Honest about its own limits: a score is one dated signal, not a verdict.
- [Pricing](https://metegpt.com/pricing): Free tier limits and paid plans.

## Detector Evaluations (dated evidence records)

Each page sets a detector's own marketing claims against the independently published record, with a date and a link on every figure, under the MeteGPT Evidence Protocol (MEP) v1.0. The only first-party figure is MeteGPT's own measured, dated detector eval (see the methodology page); competitor figures are always third-party citations.

- [Turnitin AI Checker](https://metegpt.com/turnitin-ai-checker): How accurate Turnitin's AI checker is, its documented false-positive record, and the August 2025 update.
- [Scribbr AI Detector](https://metegpt.com/scribbr-ai-detector): Accuracy, word limits, and the open question of who built it.
- [Quetext AI Detector](https://metegpt.com/quetext-ai-detector): Whether Quetext's AI detector is reliable, and what the 98% claim rests on.
- [Winston AI Detector](https://metegpt.com/winston-ai-detector): Winston's 99.98% accuracy claim set against the independent record (one peer-reviewed test found a 45.8% false-positive rate).
- [GPTZero AI Detector](https://metegpt.com/gptzero-ai-detector): GPTZero's self-reported 99.76% accuracy / 0.08% false-positive rate set against two peer-reviewed studies that measured 10% and 18% false-positive rates instead.

## Humanizer Comparisons

Dated, evidence-backed reviews of competing humanizer tools. Competitor claims are cited from third-party sources, never adopted as our own.

- [Best AI Humanizer](https://metegpt.com/best-ai-humanizer): What documented user reports and third-party tests say about the top humanizer tools, fully linked.
- [Duey.ai vs MeteGPT](https://metegpt.com/vs/duey-ai): What Duey.ai actually is (an auto-typer first, a rewriter second) and where it fits.
- [GPTHuman.ai vs MeteGPT](https://metegpt.com/vs/gpthuman-ai): GPTHuman's marketing claims set against its own terms of service.
- [QuillBot Humanizer vs MeteGPT](https://metegpt.com/vs/quillbot-humanizer): Whether QuillBot's humanizer works, from the dated evidence.
- [Grammarly Humanizer vs MeteGPT](https://metegpt.com/vs/grammarly-humanizer): Whether Grammarly's humanizer is worth paying for, evidence not hype.

## Methodology + Evidence

- [Methodology](https://metegpt.com/methodology): The MeteGPT Evidence Protocol (MEP) v1.0. How every claim gets a dated, primary source; how sources are screened, labeled, and dual-coded; what we refuse to publish.
- [Public Evidence Log (GitHub)](https://github.com/metegpt/metegpt-evidence): The machine-readable dataset behind every competitor and detector claim on the site. Each article citation like (EV-gptzero-ai-detector-01) resolves to a dated entry here.

## Company

- [About](https://metegpt.com/about): Founder's note, mission, the two-tool story (MeteGPT ships both a humanizer and a detector, disclosed on every page).
- [Contact](https://metegpt.com/contact): hello@metegpt.com
- [Privacy](https://metegpt.com/privacy)
- [Terms](https://metegpt.com/terms)
- [Cookies](https://metegpt.com/cookies)

## Editorial Policy

MeteGPT reviews AI humanizers and AI detectors without asking readers to take any number on faith. Every competitor-tool or detector claim traces to a dated, primary, linkable source recorded in the public evidence log (github.com/metegpt/metegpt-evidence), collected under the MeteGPT Evidence Protocol v1.0 (documented census sweep, screening with reason codes, live re-verification, dual-independent coding). Community reports stay reports, never converted into invented percentages. The one performance figure MeteGPT publishes about its own tool is a measured, dated eval shown in full with its caveats on the methodology page; it never borrows or fabricates a number. Author byline: Fırat Mıhcı (ResearchGate: https://www.researchgate.net/profile/Firat-Mihci). Founder discloses building two tools in the category (MeteGPT and a sibling site); that conflict is stated on every page, which is why every claim carries a source and a date.