You are reading an article and something feels off. The phrasing is smooth but empty, the examples are generic, and you wonder whether a person wrote it at all. Search for how to check, and you will find long lists of "tell-tale signs" and an invitation to try a paid detector. Most of those lists mix strong clues with weak ones and treat them all as equal. This guide ranks the evidence, so you know what to trust, what to double-check and when to say "I can't tell".
First, ask why you are checking
The right check depends on what you need to decide. Many guides skip this, and it changes everything.
| You are a... | What you really need to know | Best focus |
|---|---|---|
| Reader | Can I trust this article? | Accuracy, sources and who is accountable for it, not how it was typed |
| Editor or site owner | Did this writer follow our rules, and is it original and correct? | Agreed policy, process evidence, fact-checking and plagiarism checks |
| Teacher | Did this student do the thinking? | Process evidence, a conversation about the work and a clear policy |
| Shopper | Are these reviews or testimonials genuine? | Signs of fake or repeated reviews, and whether the reviewer really exists |
For readers especially, "was it AI?" is often the wrong question. A careful, accurate article written with help from AI is more useful than a careless human one. The questions that protect you are: is it true, is it sourced, and who stands behind it?
The evidence ladder: what to trust, from strongest to weakest
| Strength | Type of evidence | Examples | How much it proves |
|---|---|---|---|
| 1. Strongest | Direct evidence | A disclosure that AI was used. Draft or version history showing generated text pasted in. Chatbot leftovers such as "As an AI language model" or an offer to regenerate the response. | Good evidence that AI was used, though not how much or whether a person reviewed it |
| 2. Strong for reliability | Verifiable facts | Cited sources or quotes that don't exist or don't say what the article claims. Wrong names, dates or numbers. | Proves the article is unreliable. It does not prove who wrote it, because humans make errors too |
| 3. Moderate | Provenance | No real author, no contact details, no track record, a publication date that doesn't fit the content, a site that publishes huge volumes with no editorial process | Raises questions about accountability. It does not prove AI use |
| 4. Weak | Style clues | Generic phrasing, even sentence rhythm, stock phrases, no personal detail | Very little. People write this way too, and AI text can be edited to avoid it |
| 5. Weakest | Detector scores | A percentage or label from an AI detector | A guess about style patterns that can be wrong in both directions |
Why style clues and detector scores sit at the bottom
The usual advice is to look for stock phrases like "delve into" or "in today's fast-paced world". Those lists age quickly as models change, and a person can delete them in seconds. Detectors face the same problem. The FTC ordered the maker of an AI content detector to stop making accuracy claims it could not back up with reliable evidence. A 2023 Stanford-led study published in the journal Patterns found that several detectors often flagged writing by non-native English speakers as AI-generated. Our guide on free AI content detectors shows how to test one yourself.
How to check an article, step by step
- Look for disclosure. Check the byline, the About page, the editorial policy and the footer. Some publishers say when and how they use AI.
- Look for chatbot leftovers. Scan the start and end for phrases that belong in a chat reply, such as "Certainly! Here is a draft", "As an AI language model", "I hope this helps" or placeholder text in brackets.
- Check who wrote it. Is there a real person with a profile, other published work and a way to contact them? Does the article match the expertise they claim?
- Verify the sources. Open the links and look up the studies, books and quotes. Do they exist, and do they say what the article says? Missing or mismatched sources are the clearest practical red flag.
- Check the specifics. Test two or three facts, names, numbers or dates. Generic articles often avoid specifics, and generated text can state wrong ones confidently.
- Compare with the author's other work. A sudden change in voice, depth or quality can be worth asking about, but it has innocent explanations too, such as an editor or a different topic.
- Only then consider a detector, and treat any result as a prompt to look harder, not an answer.
Citation checker
Look up up to five citations or quotes from the article and record what you found. It runs in your browser only, and nothing leaves the page.
What have you found?
Tick what you have actually observed. The tool shows what that evidence can and cannot support. It runs in your browser only.
Clues worth taking seriously
- Chatbot leftovers. Phrases that make sense only in a chat, such as an offer to adjust the answer, a note about the model's limits or a "regenerate" prompt, are hard to explain any other way.
- Invented sources. References that look plausible but cannot be found, or real papers described incorrectly. This is a sign of careless generation or careless copying.
- Confident errors. A smooth, assured paragraph that gets basic facts wrong.
- Hollow specifics. Many words with no named examples, dates, numbers, places or first-hand detail where a human expert would naturally include them.
- Mismatched dates. Content that refers to events that could not yet have happened, or ignores something recent and obvious.
- Mass production. A site that publishes huge volumes of similar articles with no named editors and no corrections policy.
If you need to raise it with someone
- Start with questions, not accusations. "Can you walk me through how you wrote this?" is better than "AI wrote this".
- Ask for process evidence: outline, notes, sources, drafts and version history.
- Use a policy that was agreed beforehand. It should say what AI use is allowed and what must be disclosed.
- Avoid relying on a score. If a detector is part of your process, write down its limits, and never make it the only basis for a decision.
- Be extra careful with non-native English writers, where research has found detectors to be unreliable.
What should you do next?
Why "written by AI" is not the whole story
Google's guidance for site owners says generative AI can be useful for researching a topic and adding structure to original content, but using it to generate many pages without adding value for users may violate its spam policy on scaled content abuse. In other words, the concern is value and intent, not the tool. Google also says that sharing how content was created can give readers helpful context.
Reviews are a different matter. The FTC's rule on fake reviews and testimonials took effect on October 21, 2024. It bans fake or false reviews, including testimonials that claim to come from someone who does not exist or from a person without the experience they describe, and the FTC said AI-generated reviews are covered. The rule does not ban AI use in general, but it matters if you are trying to work out whether a product review is real.
Conclusion
To check if an article is written by AI, start where the evidence is strongest: disclosure, draft history and chatbot leftovers. Then verify the sources, quotes and facts yourself, and consider who is accountable for the piece. Treat style and detector scores as hints that deserve a closer look, never as proof. And remember that accuracy, originality and honesty matter more to readers than the tool that produced the first draft.
Related guides: free AI content detectors for blog posts and AI-generated fake news checkers. Got a message or link you are unsure about? Paste it into our free checker for a quick first read.
Frequently asked questions
What is the most reliable way to check if an article was written by AI?
There is no single reliable test. The strongest evidence is direct: a disclosure, draft or version history, or chatbot leftovers in the text. Next come checks you can verify yourself, such as whether the cited sources and quotes exist and say what the article claims. Writing style and detector scores are the weakest evidence.
Can I tell just by reading an article?
Sometimes you can spot clues, such as generic phrasing, stock phrases or an offer to regenerate left at the end, but they are unreliable. Skilled editing can hide them, and plenty of human writing is generic too. Do not accuse anyone of using AI on style alone.
Do AI detectors work?
They can give you a clue, but they can be wrong in both directions. The FTC has ordered one AI content detector's maker to stop making accuracy claims it could not support, and a 2023 Stanford-led study found several detectors often flagged writing by non-native English speakers as AI-generated. Our guide on free AI content detectors explains how to test one.
Are fake citations a sign of AI?
They can be, because generated text may invent sources or attach real-sounding details to the wrong ones. But a source that does not exist is a problem whoever wrote the article. It means you should not trust the article until its claims are verified.
Is it a problem if an article was written with AI?
Not automatically. Google's guidance says generative AI can be useful for research and structure, and the concern is generating many pages without adding value for users. The real problems are inaccuracy, misleading readers, mass-produced low-value content, and hiding AI use where honesty is expected, such as in schoolwork, reviews or journalism.
How can I check a student's or freelancer's work fairly?
Do not rely on a detector score alone. Ask for process evidence such as outlines, notes, sources and draft or version history, talk to the writer about their work, and follow a clear written policy that was agreed beforehand. Treat a score as a reason to ask questions, not as proof.
Are AI-generated reviews legal?
The FTC's rule on fake reviews and testimonials took effect on October 21, 2024. It bans fake or false reviews, including testimonials that claim to come from someone who does not exist or from a person without the experience they describe, and the FTC said AI-generated reviews are covered. It does not ban all use of AI. If you suspect fake reviews, you can report them to the FTC.
This guide is general information. Tools, policies and rules change, so check official sources for the latest.
Got a message or link you want checked?
Run it through the checker