Guide — a heuristic first check for AI-written text
8 weighted writing signals, flagged phrases and a 0–100% score. Runs in your browser, no AI model, no upload.Paste into the INPUT pane, press paste to read the clipboard, or load the 🤖 AI-written or ✍️ Human-written sample. Use at least 100 words.
With Auto-analyze on, the score updates shortly after you stop typing; otherwise press ▶ analyze. The Sensitivity slider switches between Conservative, Medium and Aggressive.
The ring shows an AI likelihood from 0 to 100% with one of five verdicts, from Likely Human to Very Likely AI. Under 100 words the verdict carries a low-word-count warning.
The Signal Analysis table scores each of the 8 signals. Flagged Phrases lists up to 30 distinct hits, and the highlight tab marks them in your text: red strong signal, orange transition or filler, yellow buzzword.
Press copy report for a plain-text summary with the score, verdict, word and sentence counts and every signal. Treat it as one signal among several, never as proof.
Quick reference
- 0–24% / 25–49%
- Likely Human / Probably Human
- 50–69%
- Uncertain / Mixed
- 70–84% / 85%+
- Likely AI-Generated / Very Likely AI
- phrase density
- 28%
- sentence variety · transitions · vocab richness
- 20% · 15% · 12%
- passive · length · starters · strong phrases
- 10% · 8% · 4% · 3%
- length
- under 100 words scaled down, under 50 halved
- language
- English word lists only
- method
- hand-tuned heuristics, no trained model, no measured accuracy
AI text detection, explained
This detector does not ask an AI whether it wrote the text. It measures writing patterns that are common in language-model output, such as uniform sentence lengths, stock phrases and heavy use of transitions, and combines them into one score. That makes it fast and private, but also easy to fool and prone to false positives.
How does this AI text detector work?
This tool analyzes text using 8 heuristic signals that research and observation suggest differ between human and AI-generated writing. They are: phrase density (AI-typical buzzwords and phrases per 100 words), sentence variety (burstiness), transition density per paragraph, vocabulary richness (type-token ratio against the length of the text), passive voice frequency, average sentence length, AI-typical sentence starters, and strong AI phrases.
Each signal produces a score (0–100%). A weighted sum gives the final AI likelihood score: phrase density counts for 28%, sentence variety 20%, transitions 15%, vocabulary richness 12%, passive voice 10%, sentence length 8%, starters 4% and strong phrases 3%. The sensitivity slider loosens or tightens the thresholds. No machine learning model or external API is used — all analysis runs locally in your browser.
What AI writing styles does this detect?
The phrase lists and thresholds are hand-picked from patterns often seen in output from large language models such as ChatGPT, Claude, Gemini, Llama and Mistral; they were not trained or benchmarked on a labelled test set. Typical patterns are lower sentence length variance, overuse of transition words ("Furthermore", "Moreover"), and reliance on certain buzzwords ("nuanced", "delve", "leverage", "pivotal").
The detector works best on English text. It may give unreliable results for other languages, highly technical text, poetry, dialogue, or very short passages.
How accurate is the AI detection?
Heuristic AI detection without a trained model is inherently probabilistic. This tool will produce false positives (human text scored as AI) and false negatives (AI text scored as human). There is no published accuracy figure for this tool. Consider it one signal among many, not a definitive verdict, and never use the score on its own to accuse someone of using AI.
Accuracy improves with longer texts (100+ words) and degrades for heavily edited AI text, paraphrased content, or text in non-English languages. The detector is designed to be a quick first-pass screen, not a forensic tool.
Is my text stored or sent anywhere?
No. All analysis runs entirely in your browser using JavaScript. Your text never leaves your device — there is no server, no API call, and nothing is saved in your browser. You can even run this tool offline once the page has loaded. The page only loads the privacy-friendly Simple Analytics script, which counts visits and never receives your text.
What is "sentence burstiness" and why does it matter?
Burstiness refers to the variance in sentence lengths. Human writers naturally alternate between short punchy sentences and long complex ones — this creates high burstiness. AI models tend to produce more uniformly-sized sentences, resulting in lower burstiness.
The detector measures the coefficient of variation (CV = standard deviation / mean) of sentence lengths. A high CV suggests human writing; a low CV suggests AI.
What is the vocabulary richness (TTR) signal?
Type-Token Ratio (TTR) is the ratio of unique words to total words. A higher TTR means more diverse vocabulary. AI models trained on massive corpora often settle into frequent, "safe" word choices, producing slightly lower TTR scores — though this signal is weaker than phrase density and burstiness.
Can I use this on non-English text?
The buzzword and phrase databases are built for English text only. The phrase-based signals will be largely inactive for other languages. Word counting also only recognises the unaccented letters a–z, so accented words are split up and non-Latin scripts are not counted at all, which distorts sentence variety and vocabulary richness as well. Treat scores for non-English text as unreliable.
What is the minimum text length for reliable results?
We recommend at least 100 words for meaningful results. Below that, statistical signals get noisy: under 100 words the verdict shows a low-word-count warning and the score is multiplied by 0.75, and under 50 words it is halved.