Skip to main content
Comparisons

How the tools you already know compare

What each tool costs, what it limits and what it says about itself, read from the vendor's own pages with the date we read them. Where a vendor publishes an accuracy figure, we report it as their claim and keep it apart from ours, because nobody outside the vendor can see the texts it was measured on.

All comparisons

ToolFreePaid fromPlagiarism check
TextlumeOne run per day on each word-limited tool, with no account and no card.6.99 USD a month billed yearly, or 9.99 USD month to month, for 10,000 humanizer words, 10,000 paraphraser words and 50,000 AI detector words a month.Coming soon
GPTZeroScans of up to 10,000 characters without an account. Longer scans need a free account.Shown in the visitor's local currency. Yearly billing is marked as a 45% saving.Sold as Plagiarism Check, listed in its product menu.
QuillBotParaphrasing up to 100 words in 2 modes, basic grammar fixes, a limited AI detector and humanizer, summaries up to 1,200 words.Premium at $4.17 USD a month, billed annually, on the plan page we read.Included in Premium, 25,000 words a month.
GrammarlyA free AI detector on its website. The free plan covers spelling, grammar and tone, with 100 AI prompts a month.Pro at $12 USD a month billed yearly ($144 a year), or $30 USD month to month.In Pro, checked against billions of web pages.
TurnitinNone for individuals. Students reach it through their institution's licence.No public price. Sold to institutions.Similarity Report against Turnitin's own database.
  1. Each text language the detector reads has its own calibration. A text in any other language gets no score, only a note that the language is not calibrated.

  2. Our accuracy page lists how often human texts were flagged in a held-out test, per language, including writing by non-native speakers, with every sample size.

  3. Every reading shows a score with a range, a confidence rating and a mark on each sentence. It is a probability, never proof.

  4. Weak spots are on the accuracy page too: French, and texts where a model continues a human opening. Very short texts always get a low confidence rating.