Field notes
Tested guides on Turnitin, GPTZero, and the math AI detectors actually run on your essay. Written for students who want the practical answer before they click submit.
Grammarly's AI Detector claims 99% accuracy. Independent testing found a lot less. Here's the official claim, the RAID benchmark it cites, and what press testing actually found.
Read articleGrammarly bundles its Humanizer into a $30/mo platform ($12/mo if you commit annually). Refrazr is a dedicated tool at $6.99/mo or $1.99 packs. Here's the honest comparison.
Read articleTurnitin's AI score and Similarity score are two separate reports, and most students only ever see one of them. Here's what actually shows up, to whom, and a repository-matching angle most guides skip.
Read articleGoHumanize.ai launched in April 2026 with sensible pricing and seven-language support. We tested it, read every page, and wrote what is genuinely good and what is missing.
Read articleQuillBot swaps words. AI detectors do not measure words — they measure perplexity and burstiness, which survive every synonym swap. That is why paraphrased essays still flag on Turnitin in 2026.
Read articleUndetectable.ai is the established option at $9.99/mo. Refrazr is the leaner option at $1.99 packs and $6.99/mo. Here is the honest comparison — verified pricing, real tradeoffs, and who should pick which.
Read articleThe ethics question is not about the tool. It is about whether your work represents your thinking. Here is the honest breakdown — including the cases where the answer is yes.
Read articleWe tested seven AI humanizers on 30 essays in March 2026. Structural rewriters change the writing; lexical tools just reword it. Here is the data and the recommendation.
Read articleFive patterns do most of the work in flagging text as AI: uniform sentence length, vocabulary clusters, transition density, em dashes, parallel structure. Fix all five and the writing stops reading as machine-made.
Read articleTurnitin catches raw ChatGPT 88–98% of the time. Hybrid edits trip it half the time. Deep rewrites read differently. Here is the data and the practical fix.
Read articleGPTZero is reasonably accurate on raw AI output and unreliable on anything edited, hybrid, or written by an ESL student. We pasted 100 essays and recorded what happened.
Read articleTurnitin's AI detector measures per-token probability. Polished writing scores highest — even when a human wrote it. Here's what actually changes how your draft reads before you submit.
Read articleFree 500 words/day, no signup. Paste your essay, see the score, click humanize.
Try Refrazr free