Samples

Try the free AI detector on these samples.

Six real-world text samples — two genuine AI outputs, three genuine human texts and one genuine mix. Every sample is verified against the live engine; scores are printed on each card. Click any sample to load it into the detector and run the analysis. Image, video and audio examples with visual previews are shown below the text samples.

AI

Heavy AI (GPT-style)

A real ChatGPT answer (HC3, personal finance) — 246 words, scores 95 AI. The hedged explainer register does the work, not caricature markers.

Co-signing a personal loan for a friend or family member can be a risky proposition. When you co-sign a loan, you are agreeing to be responsible for the loan if the borrower is unable to make the payments. This means that if your friend or family member defaults on the loan, you will be on the hook for the remaining balance.There are a few things to consider before co-signing a personal loan for someone:Do you trust the borrower to make the payments on time and in full? If you are not confident that the borrower will be able to make the payments, it may not be a good idea to co-sign the loan.Can you afford to make the payments if the borrower defaults? If you are unable to make the payments, co-signing the loan could put your own financial stability at risk.What is the purpose of the loan? If the borrower is using the loan for a risky or questionable venture, it may not be worth the risk to co-sign.Is there another way for the borrower to get the loan without a co-signer? If the borrower has a good credit score and is able to qualify for a loan on their own, it may not be necessary for you to co-sign.In general, it is important to carefully consider the risks and potential consequences before co-signing a loan for someone. If you do decide to co-sign, it is a good idea
AI

AI listicle

A real ChatGPT answer (HC3 ELI5) built as enumerated reasons — 240 words, scores 89 AI with 12 hot sentences.

There are a few reasons why some websites might hide the price of a product until it is added to the cart: 1. To encourage customers to add more items to their cart: By hiding the price until the item is added to the cart, the website may be trying to encourage customers to add more items to their cart before they see the total cost. This can be especially effective if the website offers discounts for larger orders. 2. To reduce comparison shopping: By hiding the price until the item is added to the cart, the website may be trying to make it more difficult for customers to compare the price of the product with prices offered by other websites. This can be especially effective if the website has a unique or hard-to-find product that is not available elsewhere. 3. To create a sense of exclusivity: Some websites may hide the price of a product until it is added to the cart as a way of creating a sense of exclusivity or scarcity. This can make customers feel like they are getting a special deal or that the product is in high demand. 4. To reduce distractions: Some websites may hide the price of a product until it is added to the cart in order to reduce distractions or distractions for the customer. This can be especially effective if the website has a lot of products and wants to
HUMAN

Human — casual

A real human forum answer (HC3, medical explainer) — 214 words, scores 7 Human. Conversational, specific, full of ordinary words.

So, with cancer, what is actually happening is a few things. First of all its not just one mutation. Your body mutates constantly, DNA is always messed up, and even then your RNA gets spliced up and you are making faulty proteins all the time. Mutation is actually how things evolve partly. The concept with cancer is actually different than just mutation. Think of a car. If you cut the gas, well it wo n't move. Sometimes DNA will mutate and your body will find it and wo n't make it duplicate or be transcribed during transcription. Now DNA also can actually not be picked up by " scanners " when it mutates sometimes. We can call this cutting the breaks. So during transcription lots of crap happens that 's not supposed to, and so we just let it go inside the cell and push everything that has been made out of the nucleus to get eaten up by lysosomses (cell stomachs). Now normally for actually cancer to happen, we need about 5 breaks cut and 10 gas pedals broken. It takes a lot of DNA fuckup to actually cause cancer. And if it does happen, we have this awesome enzyme called p53, which Is actually one GIGANTIC and I mean FUCKING HUGE protein made to
HUMAN

Human — academic

A real pre-LLM arXiv introduction — 245 words, scores 20 Probably human. Dense and formal with inline LaTeX, and the engine clears it.

In the following we connect a number of old and venerable topics related to symmetries and conservation laws, such as Lie algebras and Noether’s theorem, with differential geometric structures like Lie derivatives and Killing vectors. Although most of the basic ideas are well-known 4 4 4 See for example refs. [ 2 ] - [ 4 ]., we present extensions and generalisations of interest in the description of certain physical systems; in particular we apply our methods to study the motion of spinning particles in a curved space-time. According to Einstein’s equivalence principle the world line of a massive scalar point particle in curved space-time is a time-like geodesic, described by the equation D 2 ​ x μ D ​ τ 2 = x ¨ μ + Γ λ ​ ν μ ​ x ˙ λ ​ x ˙ ν = 0, \frac{D^{2}x^{\mu}}{D\tau^{2}}\,=\ddot{x}^{\mu}\,+\,\Gamma_{\lambda\nu}^{\ \ \ \mu}\,\dot{x}^{\lambda}\dot{x}^{\nu}=0, (1) with overdots denoting proper time derivatives. Its time-like nature is expressed by the condition (d ​ s d ​ τ) 2 = g μ ​ ν ​ d ​ x μ d ​ τ ​ d ​ x ν d ​ τ = − c 2 0, \left(\frac{ds}{d\tau}\right)^{2}=g_{\mu\nu}\,\frac{dx^{\mu}}{d\tau}\frac{dx^{\nu}}{d\tau}=-c^{2} 0, (2) where the universal constant c c (the light velocity) is a real number. In the following we usually take c = 1 c=1 and consider particles of unit mass, but occasionally we re-instate the explicit mass dependence
HUMAN

Human — technical

The English Wikipedia article on the Internet — 241 words, scores 44 Inconclusive. Encyclopedic register is the detector’s hardest human case: this one stays below the 60% decision line, but the engine declines to call it human outright.

The Internet (or internet) is the global system of interconnected computer networks that uses the Internet protocol suite (TCP/IP) to communicate between networks and devices. It is a network of networks that comprises private, public, academic, business, and government networks of local to global scope, linked by electronic, wireless, and optical networking technologies. The Internet carries a vast range of information services and resources, such as the interlinked hypertext documents and applications of the World Wide Web (WWW), electronic mail, discussion groups, internet telephony, streaming media and file sharing. Most traditional communication media, including telephone, radio, television, paper mail, newspapers, and print publishing, have been transformed by the Internet, giving rise to new media such as email, online music, digital newspapers, news aggregators, and audio and video streaming websites. The Internet has enabled and accelerated new forms of personal interaction through instant messaging, Internet forums, and social networking services. Online shopping has also grown to occupy a significant market across industries, enabling firms to extend brick and mortar presences to serve larger markets. Business-to-business and financial services on the Internet affect supply chains across entire industries. The origins of the Internet date back to research that enabled the time-sharing of computer resources, the development of packet switching, and the design of computer networks for data communication. The set of communication protocols to enable internetworking on the Internet arose from research and development commissioned in the 1970s by the Defense Advanced
MIXED

Mixed human + AI

A human narrative with a whole generated answer pasted into the middle — 510 words, scores 64 AI-assisted with roughly half the words reading as generated and the sentence map marking the pasted block. That split is the honest answer for a mixed document.

The first thing you notice at the Kolkata flower market is that nobody is looking at the flowers.

It's four in the morning and the marigolds are stacked in sacks the size of small cars, but the men moving them have their eyes on each other, on the narrow lane, on the wet brick underfoot. A garland the length of a cricket pitch goes past on someone's shoulder. Nobody stops. I stood there with my notebook getting steadily damper and understood, after about twenty minutes, that I had misunderstood the assignment. There is no scene here to describe. There's just work.

Co-signing a personal loan for a friend or family member can be a risky proposition. When you co-sign a loan, you are agreeing to be responsible for the loan if the borrower is unable to make the payments. This means that if your friend or family member defaults on the loan, you will be on the hook for the remaining balance.There are a few things to consider before co-signing a personal loan for someone:Do you trust the borrower to make the payments on time and in full? If you are not confident that the borrower will be able to make the payments, it may not be a good idea to co-sign the loan.Can you afford to make the payments if the borrower defaults? If you are unable to make the payments, co-signing the loan could put your own financial stability at risk.What is the purpose of the loan? If the borrower is using the loan for a risky or questionable venture, it may not be worth the risk to co-sign.Is there another way for the borrower to get the loan without a co-signer? If the borrower has a good credit score and is able to qualify for a loan on their own, it may not be necessary for you to co-sign.In general, it is important to carefully consider the risks and potential consequences before co-signing a loan for someone. If you do decide to co-sign, it is a good idea

I'd come because a friend said it was beautiful, and it is, in the way that a container port is beautiful — not because of what's being moved but because of the fact that it moves at all, every day, before anyone else is awake. Forty thousand people, more or less, depending on who you ask and the season. Nobody has counted properly. The flowers arrive from Midnapore and Howrah on the night trains, and they're gone by nine, and by ten the lane smells only of the river.

What stays with me isn't the colour. It's a man near the eastern end who sold nothing but the wire you use to bind the garlands: spools of it, thin and grey, in a shop about the width of a wardrobe. He'd been there thirty-one years. I asked him if business was good and he tilted his head in that way that means neither yes nor no, and said the flowers change but the wire doesn't. Then he went back to what he was doing, which was nothing in particular, and I went to find tea.

Image detection samples — what to try

No stock-photo guessing: the honest test is the original camera file versus a generated export. The two deterministic test images below were generated locally (seeded, reproducible) to illustrate the two ends of the pixel physics — drop either into the Image tab above for a live reading, or try it yourself with any photo you own.

Camera-like test image for the free AI image detector — independent per-channel sensor noise, reads camera-like
Camera-like pattern. Independent per-channel noise (seed 20260214): cross-channel residuals disagree the way a Bayer sensor forces them to. Expect a camera-leaning index. Download PNG · Open Image tab
Synthetic-like test image for the free AI image detector — shared R/G/B micro-texture, reads AI-leaning
Generated-like pattern. One shared micro-texture painted into R, G and B (seed 77031): residuals agree almost perfectly, the diffusion-decoder signature. Expect an AI-leaning index with low reliability on this small file — the report tells you why. Download PNG · Open Image tab

Video and audio detection samples — how to test

The AI video detector needs an MP4, WebM, MOV or MKV file: original phone footage versus an Sora, Veo, Runway, Pika or Kling export. The AI audio detector needs a WAV, MP3, FLAC, OGG or M4A voice clip of at least one second: natural speech versus ElevenLabs, OpenAI TTS or similar synthesis. Drop either into the matching tab above — every frame and acoustic measurement is printed with the exact arithmetic, and short or recompressed files honestly return low reliability.

VIDEO
MP4 · WebM · MOV · MKV
Video sample guidance. Use original camera footage for the camera-like baseline, then a generated clip for the AI-leaning case. Check provenance first, then temporal grain, flicker and motion coherence. No download needed: the Video tab has a built-in Animation sample button that records a locally generated clip and runs the full engine on it.
AUDIO
WAV · MP3 · FLAC · OGG
Audio sample guidance. Use a clear 3–10s voice recording versus a TTS clone of similar content. Compare spectral entropy, bispectrum anomaly and ZCR variation in the acoustic table. The Audio tab also has a built-in Test-tone sample button for a one-click pipeline check.

Audio pipeline test files — procedural, not voice

Two seeded, reproducible WAV files that exercise the full audio pipeline (decode → 14 acoustic signals → exact arithmetic). They are not real voice and carry no promised verdict — unlike the image patterns above, whose physics is directly constructed, these only prove your file decodes and every number in the report is live. For a real test, use genuine speech versus a genuine TTS clone.

Voice-like pattern. Wandering pitch, vibrato, irregular pauses and breath noise (seed 20260214, 4s, 16kHz mono). Download WAV · Open Audio tab
Synthetic-like pattern. Flat locked harmonics with metronomic gating (seed 77031, 4s, 16kHz mono). Download WAV · Open Audio tab