Pangram
A trained classifier that explicitly targets humanized text as well as raw model output. Pangram publishes its own comparisons and reports detection above 90 percent on high-quality humanized content.
What it is
Pangram trains a classifier on large sets of human and model text, and since early 2025 has included humanizer output in that training. Its position is that a humanizer leaves its own statistical fingerprint, so a detector can learn to spot rewritten model text even when the rewriter was tuned to beat a specific detector. Pangram publishes comparison tables against other vendors and claims over 90 percent detection of high-quality humanized content. Those are vendor numbers, but the blog is more specific than most about method, and the free checker is a reasonable second opinion.
Known limits
Those figures are vendor-reported. No independent study of the same claim had been published when this entry was checked.
When to use this entry
Use this entry for a detector aimed at humanized text and for a vendor that publishes its reasoning.
Questions
Is Pangram free?
A free checker with limits exists; volume use and the API are paid.
Has Pangram's humanizer claim been independently tested?
Not in a published study that this directory could find on its check date. Treat it as a vendor claim.