How do I spot AI-generated content?
The signals worth noticing, why detection tools are unreliable, and what to do instead of trying to detect.
- Difficulty
- beginner
- Time
- 10 min
- Read
- 2 min
Short answer
There is no reliable way to detect it, and detection tools produce false accusations regularly. Judge the content on whether its claims check out, not on whether it feels machine-written.
This is worth being honest about: detection does not work well. Detectors have flagged human writing — including, notoriously, well-known historical texts — as machine-generated. Verifying claims is a far better use of effort than guessing at authorship.
Step by step
- Do not trust detection tools.They have high false-positive rates and are known to misclassify writing by non-native English speakers particularly badly. Accusing someone on that basis is unsafe.
- Check the claims instead.Whoever wrote it, the question is whether it is right. Verify specifics against primary sources — that works regardless of authorship.
- Look for fabricated citations.The most reliable single signal. References that look correct but lead nowhere, or to a paper that says something else, are a well-known failure mode.
- Notice suspicious evenness.Uniform paragraph lengths, balanced 'on the one hand' structure throughout, tidy summaries of everything, no specific anecdotes and no strong opinions.
- Look for missing specificity.Generated text often covers a topic broadly and correctly while containing no particular detail — no dates, names, numbers or first-hand observation.
- Check images for the usual tells.Hands, text in the background, jewellery, repeated patterns and inconsistent lighting. Improving quickly, so this is becoming less reliable too.
- Consider the provenance.Who published it, do they have a reputation to protect, is there a named author with a history. This is more informative than the text itself.
- Accept the limits.For most purposes it does not matter who wrote it. It matters whether it is accurate and whether the source is accountable.
Tips
- Never accuse someone of using AI on the strength of a detector. The false-positive rate is high enough that it will eventually be wrong about a real person.
- Reviews and testimonials are worth extra scepticism — that is where generated content has the clearest commercial incentive.
- The useful question is not "was this written by a machine" but "is this true and who is accountable for it".
Questions people ask
Do AI detectors work?
Not reliably. They produce false positives frequently, particularly on writing by non-native English speakers, and should not be used as evidence.
What is the most reliable signal?
Citations that do not check out. Beyond that, judge the content on accuracy rather than trying to determine authorship.