How to Spot AI-Generated Images, Video and Voice Clones
How to spot deepfakes and AI-generated media: visual tells in images and video, signs of voice clones, verification tools and how to protect yourself.
Why AI Text Detectors Fail (and What to Do Instead)
Why AI writing detectors are unreliable, how false positives harm students and writers, what the research says, and better ways to verify authorship.
Key takeaways
- AI text detectors produce false positives and can be fooled by light editing.
- Non-native English writers and formulaic writing are more likely to be wrongly flagged.
- Assess process and understanding instead: drafts, version history and conversation.
On this page
Teachers, editors and employers understandably want to know whether a piece of writing was produced by AI, and job seekers wonder the same about applications (see AI for job search). Detection tools promise an answer. The problem is that they are not reliable enough to base serious decisions on.
How AI detectors work
Most detectors look for statistical patterns, such as how predictable each word is given the words before it. AI-generated text tends to be more predictable; human writing tends to vary more.
Why they fail
- False positives: clear, simple or formulaic human writing can look “predictable”.
- Bias: research has found that detectors flag non-native English writers’ work as AI-generated far more often.
- Easy evasion: light editing, paraphrasing tools or prompting for a different style can defeat them.
- Moving target: new models from OpenAI, Anthropic and Google arrive faster than detectors can adapt.
- Mixed writing: most real text is now partly human, partly AI-assisted, which detectors can’t meaningfully score.
Even developers of AI models have withdrawn their own detection tools because of low accuracy.
The cost of getting it wrong
A false accusation can damage a student’s record or a writer’s reputation. Students can find advice on using AI honestly in our guide to AI for students. Treat a detector score as, at most, a prompt for a conversation, never as proof.
Better approaches
In education
- Ask for drafts, outlines and version history.
- Have short conversations about the work to check understanding.
- Design assignments that use personal experience, class discussion or in-class writing.
- Set clear rules on acceptable AI use.
In publishing and business
- Judge work on accuracy, originality and usefulness, whoever typed it.
- Require writers to disclose AI assistance under a clear policy.
- Fact-check claims and sources, the real risk with AI-assisted text.
For writers
If you’re worried about being wrongly flagged, keep drafts and notes, use version history and be open about how you used AI tools.
For synthetic images, video and voice, detection is also unreliable; see our guide to spotting deepfakes for source-based checks that work better.
Frequently asked questions
Can AI detectors prove a text was written by AI?
No. They estimate likelihood and can be wrong in both directions.
Why was my own writing flagged as AI?
Clear, simple or formulaic writing can resemble AI-generated text statistically. Keep drafts to show your process.
Is there a reliable way to detect AI writing?
Not from the text alone. Process evidence, such as drafts and discussion, is far more reliable.
Sources
Every article is edited by a human and checked against our editorial policy. Spotted a mistake? Tell us.
Keep reading
AI Voice Cloning: How It Works and How to Protect Yourself
How AI voice cloning works, how scammers use cloned voices in fake emergency calls, the warning signs, and steps like family safe words to protect yourself.
AI Video Generators: What They Can Do and How to Use Them
How AI video generators work, what they can and can’t do, prompt tips for better clips, costs, and the rules on consent, copyright and labelling.