AI Detector

Paste an essay, article or email to estimate how likely it was written by AI. You get an overall AI score plus every sentence highlighted, and your text never leaves your device.

Updated 100% free Text stays on your device No sign-up

AI Detector – free online tool
0 words

The AI models (about 210 MB) download once on first use and are then cached by your browser.

How to check if text was written by AI

  1. Paste the text into the box above. Use at least 80 words; a few paragraphs give the most reliable result.
  2. Click Check for AI. The first time, the AI models (about 210 MB in total) download and are saved in your browser, so later checks start in seconds.
  3. Read the AI score and the verdict, then look at the highlighted sentences to see which parts look machine-written.
  4. Click Copy report if you want to share or save the result.

How this AI detector works

The detector runs two AI models inside your browser and combines their results:

  1. An AI detection model. A RoBERTa model fine-tuned by Fakespot AI on large sets of human-written and AI-generated English, including text from modern chatbots. It reads your text in sections of about 300 words and estimates how likely each section is to be machine-written.
  2. A predictability check. A small language model (GPT-2) measures how surprised it is by each word, called perplexity. AI chatbots pick very likely words, so their text is predictable. When a section is unpredictable, typical of real people, the AI score is lowered. This cuts false alarms on casual, personal and technical writing.

The report also shows two extra signals that help you judge the result yourself:

  • Sentence variety – people mix short and long sentences; AI text often keeps a steady rhythm.
  • Typical AI phrases – expressions chatbots overuse, such as “in today’s fast-paced world”, “delve into”, “plays a crucial role” and “in conclusion”.

How to read the result

  • 0–34% – likely human. The wording is varied and less predictable.
  • 35–64% – mixed or unclear. The text may be partly AI-written, AI-assisted or heavily edited.
  • 65–100% – likely AI. The text is very predictable and reads evenly, like typical chatbot output.

The coloured highlights show the score of each sentence, so you can see exactly which parts drive the result.

Important: no AI detector is 100% accurate

Every AI detector, including well-known paid tools, only estimates probability. Simple or formal human writing, lists, legal text and texts by non-native English speakers can look predictable and get a high score, while AI text that was rewritten by hand can pass as human. Never use a score alone to accuse someone of cheating or plagiarism. Treat it as one signal, and look at drafts, sources and the writer’s other work as well.

Who uses an AI content checker?

  • Teachers and lecturers checking essays and assignments.
  • Editors and website owners reviewing articles from freelance writers.
  • Students and writers making sure their own work does not sound robotic before they submit it.
  • Recruiters reading cover letters and job applications.

Tips to make your writing sound more human

Use your own examples and experiences, vary the length of your sentences, cut filler phrases such as “it is important to note that”, and write the way you would explain something to a friend. Then check your word count with the word counter and your keyword use with the keyword density checker.

Is my text private?

Yes. The AI model runs in your browser, on your own device. Your text is never uploaded or stored, so it is safe to check unpublished work, student essays and confidential documents.

We tested it: classics vs AI text

On 12 November 2026 we ran five texts through this detector: three famous passages written by people long before AI existed, and two paragraphs written by an AI assistant — one generic, one plain and specific.

TextWho wrote itAI scoreVerdict
Pride and Prejudice (1813), openingJane Austen12%Likely human
A Tale of Two Cities (1859), openingCharles Dickens61%Mixed or unclear
Declaration of Independence (1776)Thomas Jefferson and others63%Mixed or unclear
“In today’s fast-paced digital world…” guide introAn AI assistant100%Likely AI
Short paragraph of plain test results with numbersAn AI assistant0%Likely human

What the test shows:

  • Two classics scored “mixed”. Dickens and the Declaration repeat sentence patterns, which looks predictable to a detector.
  • Generic AI wording was caught at 100% (phrases like “fast-paced digital world” and “let’s dive in”).
  • Plain, specific AI writing passed as human (0%). Detectors measure style, not who typed the words.
  • So use a detector score as a hint to look closer, never as proof — especially before accusing a student or writer.

See how to check if text is AI written.

Frequently asked questions

Is this AI detector free?

Yes. It is completely free, with no sign-up and no daily limit.

Can it detect ChatGPT, Gemini and Claude?

It detects the typical patterns of text written by large language models, including ChatGPT, Gemini, Claude and Copilot. It does not identify which chatbot wrote the text.

How accurate is the AI detector?

In our tests it flagged typical ChatGPT-style text at 90–100% and scored casual human writing low. It works best on 150 words or more of plain English prose. Like every AI detector it can be wrong: very formal, encyclopedia-style or heavily edited human text can look AI-like, so use the score as a guide, not as proof.

Is my text uploaded anywhere?

No. The analysis runs in your browser and your text never leaves your device.

Does it work for Urdu, Hindi or other languages?

The model is trained on English, so results are reliable only for English text.

What is perplexity?

Perplexity measures how predictable a text is for a language model. AI-written text is usually very predictable (low perplexity), while human writing tends to be less predictable (higher perplexity).