Check Russian or English text for AI patterns. Analysis needs at least 120 words.
AI scoring runs on the SEOKit server. Short excerpts are sent to Serper/Google for source discovery; retrieved pages are compared locally. Text is not stored on the server. Local history is opt-in.
Up to 10 recent reviews, including full text, are stored only in this browser. Do not enable on a shared device. History is not stored on the server.
These markers appear together in the indicated paragraphs. This groups editing tasks; it does not strengthen a claim of AI authorship.
Numbers and links are details, not evidence of truth. Check the source, date and context; other factual claims may not be detected.
Not checked automatically. Use this checklist before publishing; no item is marked as passed.
Switch between your draft and its highlighted version without losing your place.
Six editing dimensions, priority filters and actionable explanations.
Rule-based analysis, not an authorship verdict. Review meaning and sources yourself.
No authorship or factual-accuracy verdict. A text with no matches is not certified human-written, accurate or ready to publish.
Phrase matching uses small EN/RU/UK dictionaries. Repetition means a four-word sequence appearing at least three times; up to 12 distinct sequences are selected. Structure notices use length variation (8+ sentences or 3+ paragraphs). Thresholds are editorial heuristics, not calibrated probabilities. Sentence boundaries and language detection may be imperfect. Specialist terms, quotes and deliberate repetition may be appropriate. No semantic analysis or web fact-check is performed.
Ruleset 4 adds literal RU/UK/EN phrase families and bounded sentence patterns, including repeated formulaic constructions. Repeated-word families require at least 3 occurrences (reassurance: 2), and at least one per 100 words. Openers, signposting, with-ing, participial tails, contrasts and paired dashes require 2; parentheses and formulaic constructions require 3. Isolated hedges or a single dash are not flags. Heading checks require explicit Markdown ##/###; structural comparisons and both indexes require 120 words. With-ing is English-only. New markers ignore Markdown code and URLs. No stemming, semantic similarity, citation verification or scientific authorship model is used. Combination cards group markers in the same paragraph; document-wide rhythm may contribute. Thresholds may misfire on technical, quoted or deliberately repetitive text.
Provisional automatic index, NOT an AI probability or a quality verdict. Only detectable markers count; meaning and source reliability still require an editor. A zero does not certify the text.
Each dimension: 0 = no supported markers; 1 = one rule with 1–2 matches; 2 = at least two rules or three matches. Total: 0–3 light review, 4–7 substantial editing, 8–12 consider a deep edit. These are workflow hints, not validated cutoffs. Repetition across categories may raise several dimensions. The index is available from 120 words.
Next step