What you're actually checking for
When you ask "how do I check for AI," you usually mean one of three things: figuring out whether a tool you're using runs on artificial intelligence, spotting AI-generated text or images someone else created, or understanding whether you're talking to a chatbot instead of a person. The answer depends on which one you're trying to do, because the methods are completely different.
If a company built the tool you're using, they almost always say so in their terms of service or product description — not because they have to, but because AI is now a selling point. If you're trying to spot AI-generated content someone posted online, there are detection tools, but they're imperfect. If you're trying to know whether you're talking to a human, the honest answer is: sometimes you can't be sure, and that's becoming more true every month.
Key Takeaways
- Companies that use AI in their products usually advertise it, so check the product description, help section, or terms of service before looking elsewhere.
- AI detection tools exist for text and images, but they're not reliable — they produce false positives and false negatives regularly.
- If a chatbot or customer service contact is AI, the company is usually required to tell you at the start of the conversation, though enforcement varies.
- AI-generated text often has patterns: repetitive phrasing, overly formal tone, lists instead of natural flow, and paragraphs that don't quite connect to each other.
- No single method catches all AI content, so combining multiple checks — reading carefully, using detection tools, and asking direct questions — works better than relying on one.
Checking whether a tool you use contains AI
Start with the source. If you're using a writing tool, image generator, search engine, or anything that processes language or images, the company's own website is the fastest place to look. Search their help documentation for "AI" or "machine learning" — most companies have a page explaining how their product works. Microsoft Copilot, ChatGPT, Google Gemini, and Adobe Firefly all have public documentation about their AI systems.
If the company doesn't mention AI anywhere, check the terms of service. Search for "artificial intelligence," "machine learning," "neural network," or "large language model." If those terms don't appear, the tool probably doesn't use AI in the way you're thinking — though it might use simpler automated systems that aren't considered AI. When in doubt, contact the company directly through their support email or chat. A straightforward question like "Does this tool use artificial intelligence?" gets a straightforward answer.
Spotting AI-generated text
Read it like a human first. AI-generated text has recognizable patterns, even when it's good. Look for: paragraphs that start with the same structure ("Furthermore," "Additionally," "To summarize"), lists instead of flowing sentences, repetition of the same adjectives, and transitions that don't quite fit. AI often writes in a formal, cautious tone — it hedges with phrases like "may be," "could potentially," and "it is possible that" more than a human would naturally speak.
The text also tends to be perfectly organized in a way that feels artificial. Real writing meanders, circles back, changes its mind. AI writing marches forward in neat sections. If every paragraph is exactly three sentences, or if the whole piece follows a rigid formula, that's a sign.
For a more technical check, you can use AI detection tools. Originality.AI, Turnitin, and GPTZero are the most widely used. You paste in the text and they give you a percentage likelihood that it was AI-generated. The catch: these tools are wrong frequently. They flag human writing as AI and miss AI writing that's been edited. They work better on longer pieces than short ones, and they struggle with text that mixes human and AI writing. Use them as one signal, not the final answer.
Detecting AI-generated images
AI images have gotten much harder to spot with your eyes alone. Early AI images had obvious tells — weird hands, strange text, impossible physics. Modern AI (DALL-E 3, Midjourney, Stable Diffusion) produces images that look real. You can't reliably spot them by looking anymore.
Image detection tools exist, but they have the same problem as text detectors: they're imperfect. Reverse image search (Google Images, TinEye) can tell you if an image has been posted elsewhere, which sometimes reveals whether it's AI or a real photograph. If you're trying to verify whether a photo is real, look at the metadata — the hidden information embedded in the file. Real photos from cameras contain EXIF data (camera model, date taken, location). AI images don't. You can check EXIF data with free tools like Jeffrey's Exif Viewer or by right-clicking the image and checking properties.
The most reliable method is context. Where did the image come from? Did the person claim to have taken it themselves? Does it match their other work? If someone posts a "photo" of something that would be extremely difficult or expensive to photograph, and they have no history of that kind of work, it's worth questioning.
Knowing if you're talking to a chatbot
Most companies are required to tell you when you're interacting with a bot rather than a human. If you start a chat on a website and it's powered by AI, there's usually a disclosure at the beginning — something like "You're chatting with an AI assistant" or "This is an automated response." Read the first message carefully.
If no disclosure appears, ask directly: "Are you a human or a chatbot?" A real person will answer when ready. A chatbot might dodge the question, give a vague answer, or explain what it can help with instead of answering. Some chatbots are programmed to say they're human, which is deceptive but happens.
Chatbots also have behavioral patterns. They respond when ready, never say "I don't know" without offering alternatives, and don't get frustrated or impatient. They're extremely polite and formal. They don't make typos or use casual language. If you're testing whether something is a bot, try asking something outside its training — a very recent news event, a local question, or something absurd. Humans adapt and say "I have no idea." Bots either refuse to answer or make something up.
Why detection isn't perfect
AI systems are improving faster than detection tools. A detector trained on older AI models often fails on newer ones. Companies also deliberately make AI output harder to detect by having humans edit it, mixing it with human writing, or using techniques that make it look more natural. Someone motivated to hide AI content can do so fairly easily.
The other direction is also true: detection tools flag human writing as AI, especially writing that's formal, well-organized, or written by non-native English speakers. A student writing a carefully structured essay might get flagged as AI. A translated document might get flagged. This is why no detection tool should be used alone to make important decisions — like accusing someone of cheating or rejecting a job process.
What to do if you're unsure
Ask the source directly. If you think someone used AI to create something, ask them. If you're evaluating a tool and can't find information about whether it uses AI, email the company. If you're reading something online and suspect it's AI-generated, check the author's other work — do they have a consistent voice? Do they write about the same topics regularly?
For high-stakes situations — hiring, academic integrity, publishing — combine multiple methods. Use detection tools, read carefully for patterns, check metadata, and ask questions. No single method is reliable enough to stand alone, but together they give you a much clearer picture.
Frequently Asked Questions
Can AI detection tools tell the difference between human and AI writing?
They can make an educated guess, but they're not reliable. These tools flag human writing as AI and miss AI writing regularly, especially if the AI text has been edited by a human. They work better on longer pieces and worse on short ones. Use them as one signal among several, not as proof.
What should I do if a detection tool says something is AI but I think it's human?
Ask the person who wrote it. If they say they wrote it themselves and you believe them, that's your answer. Detection tools are wrong often enough that a false positive is common, especially with formal writing or non-native English speakers.
Is it illegal to use AI to write something without telling people?
It depends on the context. In schools, using AI without permission usually violates academic integrity policies. In publishing and journalism, undisclosed AI use can violate editorial standards. In business writing, it's often fine as long as the work is accurate. There's no single law against it, but specific organizations have their own rules.
Why do chatbots sometimes claim to be human?
Some are programmed that way by their creators, either as a test or because the company wants to see how users interact without knowing it's a bot. It's deceptive, but it happens. If you ask directly and get a vague answer, you're probably talking to a bot.
Can I trust AI detection tools to catch cheating in school?
Not as the only method. These tools produce false positives — they flag human work as AI — and false negatives — they miss AI work. Schools that rely solely on detection tools end up accusing innocent students. Combine detection tools with reading the work yourself, comparing it to the student's previous writing, and asking questions about their process.