01How AI content detectors work
AI detectors look for statistical and stylistic differences between human and machine writing. Commercial tools usually rely on classifiers trained on large collections of human and AI-generated text. Research methods such as Binoculars compare how predictable a passage is to two different language models. Both approaches need substantial computing power and run on remote servers.
This detector takes a lighter approach that runs entirely in your browser. It measures eight stylistic signals linked to AI-generated text, including the explainer framing typical of newer models, the excess style vocabulary that Kobak and colleagues (2024) traced across millions of research abstracts, low variation in sentence length, formulaic transitions and a lack of concrete detail. It then combines them into a likelihood and shows exactly which sentences contributed.
02How accurate are AI detectors?
Less accurate than their scores suggest. OpenAI withdrew its own AI text classifier in July 2023, citing a low rate of accuracy: it identified only 26% of AI-written text as likely AI and wrongly labelled 9% of human writing. A widely cited Stanford study (Liang et al., 2023) found that seven popular detectors classified more than half of essays written by non-native English speakers as AI-generated, while rating native speakers' essays almost perfectly.
Detection also gets harder as models improve and as people edit generated text. Short passages, formulaic genres such as lab reports, and heavily edited writing are especially likely to be misjudged in either direction.
03Common signs of AI-generated writing
None of these proves anything on its own, and good human writers use all of them occasionally. What stands out is when many appear together and the writing could apply to almost any topic.
- Polished but generic statements that make no specific claim and cite no specific evidence.
- Style words such as delve, pivotal, multifaceted, underscore, tapestry and landscape.
- Framing phrases like it is important to note, in today's fast-paced world and plays a crucial role.
- Every paragraph opening with Furthermore, Moreover or Additionally.
- Sentences and paragraphs of strikingly similar length and shape.
- Neat lists of three and balanced contrasts in almost every paragraph.
- Confident references to studies or statistics that cannot be traced to a real source.
04What to do if your work is flagged as AI
- 1
Stay calm and ask what the concern is based on
A detector score alone is weak evidence. Ask whether other concerns were raised and how the decision will be made under your institution's academic integrity policy.
- 2
Gather evidence of your process
Version history in Google Docs or Word, earlier drafts, research notes, saved sources and browser history all show how the work developed over time.
- 3
Be ready to talk about the work
Explaining your argument, your sources and why you made particular choices is one of the most convincing ways to show the writing is yours.
- 4
Get support
Most universities have independent advice services and student unions that help with integrity cases. Contact them early.
05Using AI tools responsibly in assignments
Policies on generative AI vary widely between universities, departments and even individual modules. Some ban it for assessed work, some allow it for specific tasks, and many require you to declare how you used it.
- Read the AI policy for each module before you start, and ask your tutor if it is unclear.
- Declare any AI use your institution requires you to acknowledge, in the format it specifies.
- Never submit generated text as your own writing, even after editing it.
- Check every fact, quotation and reference an AI tool gives you: they regularly invent sources that do not exist.
- Keep your drafts and notes, so you can show how your work developed if you are ever asked.
