AI Content Detector — Pattern Analysis
An honest look at how machine-like a text reads. Eight transparent signals, each explained, with the evidence highlighted in your text. It runs in your browser and never claims to know who wrote something.
Machine-like patterns
Signal breakdown
Sentence lengths
Highlighted text
Phrases found
Related Tools
How to Use
- Paste at least 80 words — Paste the text you want to check. 80 words is the minimum; 300 words or more gives the signals much more to work with.
- Click Analyze — The text is analysed in your browser. You get a 0–100 “machine-like patterns” score, a breakdown of eight separate signals, and the evidence behind each one.
- Read the evidence, not just the number — Overused phrases, hedges, sentence-opening transitions and dashes are highlighted in your text, and a chart shows how much your sentence lengths vary. Each signal explains what it measured and why it matters.
- Use it as a writing aid — Rewrite highlighted clichés, vary sentence length and cut filler transitions if you want your writing to sound more natural. Never treat the score as proof of who wrote a text.
About the AI Content Detector
Most “AI detectors” give you a single percentage and a confident label without showing how they got there. Independent research has repeatedly found such tools unreliable: OpenAI withdrew its own AI text classifier in July 2023 because of its low accuracy, and a Stanford study the same year found popular detectors misjudging most essays by non-native English writers. WritePadPro’s AI Content Detector takes a different approach. It measures specific, explainable writing patterns that are common in text from large language models, shows you the evidence for each one, and reports a machine-like patterns score rather than a verdict about who wrote the text.
The eight signals
| Signal | What it measures | Why it matters |
|---|---|---|
| Sentence-length variation | Coefficient of variation (standard deviation ÷ mean) of words per sentence | People mix short and long sentences; language models tend to produce sentences of similar length. A CV of 0.25 or less scores 100, 0.70 or more scores 0. |
| Overused AI phrases | Weighted matches per 1,000 words from a list of phrases such as “delve into”, “tapestry”, “a testament to”, “multifaceted”, “it’s important to note”, “navigate the complexities”, “in today’s fast-paced world” | These phrases appear far more often in model output than in ordinary human writing. |
| Sentence-opening transitions | Share of sentences that begin with words such as “Moreover”, “Furthermore”, “Additionally”, “Overall”, “In conclusion” | Models signpost every step; most writers use a handful of transitions. |
| Repetitive openings | Share of sentences starting with the most common first word | Strings of sentences opening with “This”, “It” or “The” read as formulaic. |
| Hedging | Hedges per 100 words (“typically”, “generally”, “may”, “to some extent”, “it can be argued”) | Models qualify claims constantly; heavy hedging makes text sound non-committal. |
| Dashes and semicolons | Em dashes (and dashes used like them) plus half-weighted semicolons per 1,000 words | Recent chat models use em dashes heavily. Many good human writers do too, so this signal has a low weight. |
| Paragraph uniformity | Coefficient of variation of words per paragraph (3+ paragraphs) | Model answers often come in neat paragraphs of nearly equal size. |
| Vocabulary diversity | Moving-average type–token ratio over 50-word windows (MATTR) | Low diversity means repetitive wording. Research is mixed on whether AI text is more or less varied than human text, so this signal counts least. |
Each signal is converted to a 0–100 score, and the overall score is a weighted average of the signals that had enough text to measure (phrases ×2, sentence-length variation ×1.5, transitions ×1, openings, hedging, dashes and paragraphs ×0.75, vocabulary ×0.5). The result is described as few (0–24), some (25–49), many (50–74) or strong (75–100) machine-like patterns.
A worked example
Consider: “In today’s fast-paced world, effective communication is crucial. Moreover, it plays a pivotal role in fostering collaboration. Additionally, it is important to note that clear writing serves as a testament to professionalism.” Its three sentences are 8, 9 and 15 words long, every one of them opens with a transition or stock opener (“In today’s…”, “Moreover”, “Additionally”), and seven listed phrases appear in just 32 words: “in today’s fast-paced world”, “crucial”, “Moreover”, “plays a pivotal role”, “fostering”, “it is important to note” and “serves as a testament to”. A full text written like this scores near the top of the phrase and transition signals. Compare: “Write clearly. Your readers are busy, and a muddled email costs them time they would rather spend on the actual problem.” Sentences of 2 and 19 words, no stock phrases and no transitions: writing like this triggers none of those signals.
What the score can and cannot tell you
- It cannot prove authorship. A high score means the text shares patterns with typical model output. A person writing formally can produce the same patterns, and edited AI text can avoid them.
- It can be unfair to some writers. Non-native speakers, people who learned from templates, and writers of technical or legal prose often score higher through no fault of their own.
- It should never be the only evidence. If you are a teacher or editor, talk to the writer, look at drafts and version history, and ask them about the content before drawing any conclusion.
- It is useful for revision. The highlights point at exactly the habits that make writing sound generic, whoever wrote it.
Privacy
The analysis is ordinary JavaScript running in your browser. There is no AI model, no upload and no storage, so you can check student essays, unpublished articles and confidential documents without sharing them with anyone.
Frequently Asked Questions
Can this tool prove that a text was written by AI?
No, and no tool can. It measures writing patterns that are more common in text from large language models, such as very even sentence lengths and stock phrases. People write that way too, especially in formal, technical or templated writing, and AI text can be edited until the patterns disappear. Treat the result as a description of the writing style, never as evidence about who wrote it.
How is the score calculated?
Eight signals are each scored from 0 to 100: sentence-length variation, overused AI phrases, sentence-opening transitions, repetitive sentence openings, hedging, dash and semicolon density, paragraph-length uniformity and vocabulary diversity. The overall score is their weighted average, with the phrase and sentence-rhythm signals weighted most and vocabulary diversity least. Signals that need more text than you provided are left out and marked as such.
Why did my own writing get a high score?
Because the signals describe style, not authorship. Careful, formal writing with even sentences, frequent “however” and “moreover”, and polished vocabulary scores higher than casual writing. Writing taught from templates, such as exam essays and business reports, often does too. If you wrote it, the highlights simply show which habits make prose sound generic.
Are AI detectors unfair to non-native English writers?
They can be. A 2023 Stanford study published in the journal Patterns tested seven popular GPT detectors on 91 TOEFL essays written by non-native English speakers: on average the detectors labelled 61 percent of them as AI-generated, while they judged essays by US eighth-graders almost perfectly. Simpler vocabulary and predictable phrasing look “machine-like” to these tools. This is one reason a detector score should never be used as the only evidence in an academic or employment decision.
Does it work for languages other than English?
Partly. Sentence-length variation, paragraph uniformity, repetitive openings, vocabulary diversity and punctuation work for any language that uses spaces between words. The phrase, hedging and transition lists are English only, so for other languages those three signals are skipped and the page tells you so.
Is my text uploaded or stored?
No. The analysis is plain JavaScript that runs in your browser; there is no AI model and no server involved. Your text is not sent anywhere, saved or logged, so you can safely check unpublished drafts, student work or confidential documents.
How long does the text need to be?
At least 80 words, and ideally 300 or more. Sentence-length variation needs at least five sentences, repetitive openings at least six, and paragraph uniformity at least three paragraphs. Short texts give unstable results, so the page shows a reliability note based on length.
How can I make my writing sound less machine-like?
Mix short and long sentences. Replace stock phrases such as “delve into”, “a testament to” and “in today’s fast-paced world” with plain, specific words. Cut sentence-opening transitions you do not need, start sentences in different ways, and use concrete details, examples and opinions that only you could provide. The Readability Improver and Paraphrasing Tool can help with the rewriting.