How the AI Content Detector Actually Works

Modern Large Language Models (LLMs) like ChatGPT, Claude, and Gemini produce highly structured, statistically predictable text. This tool does not send your data to an external API; instead, it uses a localized mathematical heuristic approach to scan your document. By analyzing the structural variance of your sentences and identifying known "robotic" stylistic footprints, it determines the probability that a text was generated by AI.

Core Diagnostic Metrics

The detector evaluates your text using two primary mathematical models: Burstiness (Sentence Variance) and Robotic Phrase Matches (Density).

1. Burstiness (Sentence Length Variance)

Human writers naturally vary their sentence lengths. We write short sentences for impact. Then, we might write a much longer, flowing sentence to explain a complex thought or connect multiple ideas together. AI models, on the other hand, tend to produce sentences of highly uniform length.

To measure this, the tool splits your text into individual sentences using standard punctuation (periods, question marks, and exclamation points). It counts the number of words in each sentence, calculates the mean sentence length, and computes the standard deviation of those lengths. This standard deviation value is your text's "Burstiness" metric.

• High Burstiness (> 8.0): Indicates highly variable sentence lengths (very human).
• Low Burstiness (< 3.0): Indicates uniform, monotonous sentence structures (very robotic).

2. Robotic Phrase Density

AI models have a strong affinity for specific transitional phrases and vocabulary. The tool scans your entire text against a predefined lexicon of 24 specific "AI footprints," including:

• "delve"
• "a testament to"
• "it is important to note"
• "navigate the tapestry"
• "demystify"
• "pivotal"
• "multifaceted plethora"

The tool counts the total occurrences of these phrases and calculates a Flag Density (the number of flags per 100 words).

The Scoring Algorithm

The system starts with a baseline AI Probability Score of 50% and adjusts it based on the two metrics above.

Burstiness Adjustments:

• If Burstiness is greater than 8, the score drops by 30 points.
• If Burstiness is between 5 and 8, the score drops by 15 points.
• If Burstiness is below 3 (highly uniform), the score increases by 35 points.
• If it falls between 3 and 5, the score increases by 10 points.

Density Adjustments:

• If there are more than 2 flagged phrases per 100 words, the score increases by 45 points.
• Between 1 and 2 flags per 100 words adds 25 points.
• Between 0.5 and 1 flag per 100 words adds 10 points.
• Less than 0.5 flags per 100 words reduces the score by 10 points.

The final score is clamped between 1% and 99%. A final score above 50% is classified as "Likely AI Generated," while 50% or below is "Likely Human Written."

Concrete Worked Example

Let's evaluate a short, highly-varied human text snippet:

"Wow! I didn't expect to see you here today, considering the weather was supposed to be terrible. Are you staying long? Because if you are, we absolutely must grab a coffee down the street at that new place that just opened up; I heard they have the best pastries in town."

First, the text is split into four sentences by the punctuation (!, ., ?, .):

1. "Wow" (1 word)
2. "I didn't expect to see you here today, considering the weather was supposed to be terrible" (16 words)
3. "Are you staying long" (4 words)
4. "Because if you are, we absolutely must grab a coffee down the street at that new place that just opened up; I heard they have the best pastries in town" (30 words)

The mean sentence length is 12.75 words. Calculating the standard deviation across lengths (1, 16, 4, 30) yields a Burstiness metric of roughly 11.43. Since 11.43 is greater than 8, the baseline AI score of 50 drops by 30 points to 20.

Next, the tool checks for the 24 robotic phrases. There are 0 matches in this 51-word snippet. The flag density is 0 (which is less than 0.5 per 100 words), so the score drops by another 10 points.

The final AI Probability is 10% (Likely Human Written).

Frequently Asked Questions

What is the minimum text length required for an accurate scan?

The detector requires a minimum of 50 characters to execute its heuristic scan. If you paste text shorter than 50 characters, the "Scan Document" button will remain disabled. For the most accurate burstiness calculation, we recommend pasting at least a few full paragraphs so the mathematical variance can be properly measured.

Why did my human-written essay get flagged as AI?

False positives generally happen when human writers use highly rigid, uniform sentence structures. If almost every sentence in your essay is between 12 and 15 words long, your Burstiness standard deviation will be less than 3.0. According to the algorithm, a Burstiness under 3.0 automatically adds 35 points to your AI probability. To fix this, try breaking up long thoughts into shorter, punchy sentences to increase your structural variance.

Does formatting like line breaks or semicolons affect the sentence count?

Semicolons and line breaks do not count as sentence separators in this tool. The heuristic engine strictly splits sentences using periods (.), question marks (?), and exclamation points (!). If you use a lot of semicolons to connect independent clauses, the algorithm will read them as one massive sentence, which could drastically increase your Burstiness score (making the text appear more human).

How many "robotic phrases" are checked during the scan?

The tool's dictionary contains exactly 24 specific footprint phrases commonly overused by Large Language Models, such as "moreover," "in conclusion," "catalyst," "foster," "subsequently," and "embark on." The more of these exact terms you use relative to your total word count, the higher your final AI probability score will be.