AI Content Detection
Analyses text using 12 linguistic rules: AI clichés, sentence uniformity, lexical diversity (TTR), personal pronouns, transition words, passive voice and colloquial markers.
How AI text detection works
AI Detector analyses text without external APIs or neural networks — using linguistic patterns only. The tool checks 12 independent rules and calculates a final score from 0 to 100%. The higher the score, the more machine-text signals found. Your text is never stored or shared.
What gets checked
AI clichés
Over 75 stock phrases overused by ChatGPT and similar models: "it is worth noting", "it should be mentioned", "in today's world", "in conclusion" and so on.
Sentence uniformity
AI writes sentences of almost equal length. Humans don't. The tool measures standard deviation and flags excessive regularity.
Lexical diversity (TTR)
Type-Token Ratio — unique words divided by total words. A low TTR signals a limited vocabulary and repetition, a hallmark of AI output.
Personal pronouns
AI rarely writes in first person. Low density of "I", "we", "me" is a signal of machine authorship.
Transition words
ChatGPT overuses "however", "furthermore", "in addition", "therefore". The tool counts their density per sentence.
Passive voice
AI avoids active subjects and overuses impersonal constructions: "it can be argued", "it should be noted", passive "-ed" forms.
Emotional punctuation
Absence of exclamation marks, ellipses and dashes signals a flat, neutral machine tone. Human writing is more expressive.
Colloquial markers
Over 50 informal expressions, slang and fillers. AI rarely uses "like,", "honestly,", "y'know", contractions like "don't" or "I'm".
Word repetition
When one content word appears far too often it signals the limited lexical range typical of AI generation.
Average sentence length
AI tends to build long, complex sentences. An average above 20–28 words is a suspicious signal.
Structural patterns
Too many bullet lists, numbered items and headings in a text is a typical AI generation style.
Questions
Humans ask questions — AI answers them. Several question marks in a text lower the probability of machine authorship.
Frequently asked questions
Results are indicative — linguistic analysis without neural networks cannot guarantee absolute accuracy. Genre, style and topic all affect the outcome. The detector works best on texts of 200 characters or more and identifies patterns typical of ChatGPT, GPT-4 and similar models.
No. Text is analysed entirely on the server with no storage, no sharing with third parties and no external API calls. Nothing remains after the analysis.
Ukrainian, English and Spanish are currently supported. Each language has its own dictionary of clichés, transition words and colloquial markers. More languages are planned.
For teachers and students checking essays or assignments. For editors and journalists verifying content before publication. For SEO specialists checking content authenticity. For anyone who needs to confirm that a text was written by a human.
The tool uses no paid APIs — analysis is built entirely on linguistic rules and runs on the server. This allows us to keep it free with no limits on the number of checks.