Copyleaks humanizer: what Copyleaks publishes and the tools listed on one set of fields

The table lists eight tools in order of the largest request a cheapest paid plan or free tier accepts, largest first, because Copyleaks labels each section of a document and a tool that rewrites the document whole gives you one output to read against the original. No column is a Copyleaks reading. The Copyleaks column of the published run will be added to this table when the first run publishes. Every figure comes from the vendor's own pricing or product page, checked on 8 November 2026.

None of those lines mentions Copyleaks' reading, because none can yet.

What does Copyleaks publish about its own detector?

Copyleaks' consumer page at copyleaks.com sits behind a browser check and did not load for this article, so this page reads the company's developer documentation instead, which describes the same detector as sold through its API (Copyleaks, AI Text Detection API documentation, read 9 November 2026). The documentation says the detector "flags passages that are likely AI-generated", returns a classification of human or AI for each section of text plus an overall summary, and explains flagged passages with a feature Copyleaks calls AI Logic, which points to specific sentences, compares the text to known human and AI patterns and finds content copied from AI-generated websites.

Three things about that table matter for a humanizer search. First, Copyleaks gives no accuracy or false positive figure on the page read, only the statement that independent researchers verified it; a reader who wants the number has to find the study Copyleaks refers to. Second, the detection level is a setting the reviewer chooses, so the same text can be labelled differently by two Copyleaks users, and a "Copyleaks reading" without the level is incomplete. Third, Copyleaks states in plain terms that "text spinners" and "heavily modified AI text that's meant to look human" are targets of its manipulation checks, which is the category a humanizer's output falls into.

What does the detection level mean for a rewritten text?

Copyleaks' three levels are the most useful thing on its documentation for anyone comparing humanizers. Level 1 is described as catching text "copied straight from AI tools like ChatGPT", Level 2 as finding "AI text with small changes like different tenses", and Level 3 as spotting "heavily edited AI content that's been changed a lot", read 9 November 2026. A humanizer's output is by definition changed AI text, so the level a school or publisher sets decides how that output is read before any tool is involved.

That is why the published method records the detector's settings alongside its label on the run date, with a screenshot, and scans every input before humanizing so the change is visible. A reading without the level, the date and the baseline cannot be compared with another.

Why does a Copyleaks reading need a date and a baseline?

Because every detector on the market changes and every study of detectors shows a spread. Weber-Wulff and colleagues tested 12 public detectors plus Turnitin and PlagiarismCheck on human, machine, machine-paraphrased and machine-translated text and found all 14 scored below 80 percent on their accuracy measure, with obfuscation such as paraphrasing lowering performance further (Weber-Wulff et al. 2023, International Journal for Educational Integrity, read 9 November 2026). Copyleaks' current model was not in that study and a 2023 result cannot speak for a 2026 reading, which is the point: a run carries a date or it carries nothing.

No detector score should be the only evidence in a misconduct case. Copyleaks' own documentation frames its output as evidence that "gives reviewers the evidence to decide", read 9 November 2026, which puts the decision with a person, not the label.

Copyleaks and Turnitin are often searched together because both sell to schools. Turnitin states that its checker reports likely AI-generated text "including likely AI-paraphrased or bypassed content" and that its detector is built into its own submission workflow (Turnitin, AI writing detection, read 9 November 2026); Turnitin's detector is not available to the public, so the evidence on it is of a different kind and sits on the Turnitin page.