For writers and editors

AI text detection for writers and editors

Editors and writers are asked more and more whether a piece was written by a person. A detector cannot answer that. It can point at passages that read as templated, which is often worth an editor’s attention anyway.

Last reviewed 20 September 2026

The six signals, in editorial terms

Litimus combines six measurements into one score. Each one corresponds to something an editor already looks for.

Sentence rhythm. Human prose swings between short and long sentences. A passage where every sentence runs to about the same length reads as monotone.
Stock transitions. “Moreover”, “it is important to note” and “in conclusion” are the connective filler an editor would cut.
Vocabulary spread. A low ratio of distinct words to total words means the same safe word choices recur.
Phrase repetition. The same three-word sequence turning up again and again is a copy-edit flag in any draft.
Punctuation habits. Litimus currently counts em dash frequency. Heavy use is a habit of a lot of generated text, and of plenty of human writers.
Structural symmetry. Sentences of near-identical length give a passage a templated feel.

The full explanation, with the thresholds, is in how AI detectors work.

A second opinion, not a gatekeeper

Write your policy down first. Decide what counts as acceptable AI assistance for your publication or client and tell contributors before you scan anything.
Never reject a piece on a score alone. A borderline result is inconclusive by design. Ask the writer about their process: notes, sources, earlier drafts.
Use the reasons as edit notes. If the report flags stock transitions and repeated phrases, those are things you would want tightened whoever wrote them.
Remember what a score cannot tell you. It cannot separate AI-assisted from AI-written, and heavily edited AI text can land in the middle of the range. Formulaic house styles can look templated to any statistical detector.

Plans

One-off checks are free: 2,000 words a day, no account, run in your browser. Pro is $15 a month (or $12 a month billed yearly) and adds 25,000 words a day, .docx and .pdf files, batch scans of up to 20 files and shareable reports. Team & API starts at $49 a month with five seats and a REST API. A $5 day pass adds 10,000 words for 24 hours. See the pricing page.

Recent articles

What magazines mean by “no AI-generated work”Submission guidelines rarely define the term. Reading a few carefully shows the line usually falls at drafting, not at every use of a tool.Technical writing always reads as templatedDocumentation, API references and how-to guides are supposed to be uniform and predictable. That’s also exactly what a statistical detector flags.Translation work and AI detectionA translated text is, by definition, constrained by someone else’s original — which makes it an even messier case for a detector built to read prose as if it started in English.Building an editorial AI-use policy in an afternoonA short, usable policy for contributors beats a long one nobody reads before submitting. Four decisions and a place to put them.Ghostwriting and AI detection — an awkward combinationA ghostwriter is already writing in someone else’s voice under someone else’s name. Where an AI detector fits into that arrangement, and where it doesn’t.When a client asks for an AI-detection guaranteeMore briefs now ask freelancers to “guarantee” a low AI score. What you can honestly promise, and what no detector — including this one — can.

Sources

Keep reading

Litimus vs Originality.aiIf you scan content in bulk, how a content-team suite differs.How AI detectors workWhat a score is built from, and why it is a probability.