AI Writing Pattern Checker

What the checker actually measures

Large language models are trained to produce the most probable next word, and that habit leaves fingerprints. This tool looks for four of them and shows you exactly where they are, so you can edit rather than guess.

1. AI-typical vocabulary

A curated list of about 80 words and phrases that appear far more often in model output than in ordinary writing: delve, tapestry, testament, moreover, furthermore, leverage, robust, seamless, ever-evolving, in today’s fast-paced world, it’s important to note, and chat-style leftovers such as Certainly! or I hope this helps. Every hit is counted and highlighted in the preview. None of these words is wrong on its own; the problem is density. Two in a 600-word article is normal, twelve is a pattern.

2. Sentence rhythm (burstiness)

Humans write in bursts: a twenty-word sentence, then a four-word one. Models tend to produce sentences of similar length. The tool computes the mean sentence length, the standard deviation, and their ratio, which researchers call burstiness. Below about 0.3 the text is unusually uniform; above 0.5 it has the natural variety of hand-written prose.

3. Formatting habits

• Em-dashes per 100 words. Models love the em-dash as a connector. More than two or three per 100 words stands out.
• Triads. “Fast, reliable, and affordable.” Grouping in threes is a classic rhetorical device that models overuse.
• Repeated sentence openers. If 40% of sentences begin with “This” or “It”, the paragraph reads mechanically.
• Paragraph count, so you can spot the one-idea-per-tiny-paragraph layout that chat interfaces encourage.

4. The score

The 0–100 score is a plain weighted sum, and the breakdown table shows every component: phrases (max 40), uniform rhythm (max 25), em-dashes (max 15), triads (max 10) and repeated openers (max 10). Under 30 is Low, 30–59 Moderate, 60 and above High. Because the formula is fixed and visible, the same text always gets the same score, which is not true of probabilistic detectors.

Worked example

Click Try sample to load a paragraph written in typical assistant style. It scores High: about thirty listed phrases in roughly 110 words, burstiness around 0.35, two em-dashes and a “connect, engage, and thrive” triad. Now edit it by hand:

After a few such changes, the phrase count drops to one or two, the sentence lengths vary from 4 to 22 words, and the score falls into the Low band. The point is not to game a number; it is that each edit also made the text shorter and more specific.

Practical tips

• Fix the highlighted words first, but replace them with something concrete, not a synonym. “Leverage our robust platform” becomes “use the platform”; better still, say what the platform does.
• Break the rhythm deliberately. After a long explanatory sentence, add a short one. Ask a question. Use a fragment if the tone allows.
• Cut most em-dashes. A full stop or a comma usually works and looks less mannered.
• Watch the pattern, not the score. A legal notice or a scientific abstract may legitimately use comprehensive and facilitate; the score cannot know your genre.
• Combine with the Readability Checker: AI drafts often have a high grade level and low burstiness at the same time. Bringing both down usually improves the writing more than either alone.
• Non-native writers are sometimes flagged by detectors because school English favours the same formal connectors (moreover, furthermore, in conclusion). Replacing those with plainer links such as also, and, so helps both readability and perception.

Read next

• Best Free AI Humanizer Tools: An Honest Comparison Framework — How to compare free AI humanizer tools on privacy, quotas, quality and detector claims, the red flags to avoid, and where our own free tool fits.
• 60+ ChatGPT Words and Phrases to Avoid (With Plain Alternatives) — The definitive list of AI-typical words and phrases, grouped by type, with plain replacements, why models over-use them, and how to scan for them.
• Does Turnitin Detect ChatGPT? How the AI Indicator Works — What Turnitin's AI writing indicator actually measures, why it produces false positives, and what to do if you used ChatGPT on an assignment.

Frequently asked questions