Every AI detection tool claims high accuracy. GPTZero says 99%. Turnitin says 97%. Copyleaks says 99.1%.

But these numbers come from controlled benchmark tests on unedited AI output. Real-world accuracy is a different story.

What "accuracy" actually means

When a vendor says "99% accuracy", they usually mean: on a test set of clearly AI-generated and clearly human-written text, the tool correctly classified 99% of samples.

That sounds impressive until you consider what's missing:

Mixed text — essays that are partly human, partly AI. This is the most common real-world scenario, and it's much harder to detect.

Edited AI text — AI output that's been manually revised, paraphrased, or run through a humanizer tool. Accuracy drops dramatically.

Formal academic writing — human writing that happens to be structured and polished. This triggers false positives.

Non-native English speakers — ESL students write in patterns that overlap with AI patterns.

What does independent testing show about AI detector accuracy?

Multiple published studies and third-party reviews paint a consistent picture, even if the exact numbers vary by study:

Unedited AI text: Most tools perform well here. This is the easiest case and the one vendors benchmark against.

Lightly edited AI text: Accuracy drops significantly. A student who spends time revising ChatGPT output can often reduce detection scores.

Heavily paraphrased or humanised text: Most detectors struggle badly. Tools that specifically target detection patterns can make AI text largely undetectable.

Formal human writing: False positive rates increase, particularly for non-native English speakers. The 2023 Liang et al. study from Stanford documented substantial bias against ESL writers across multiple detectors.

The "99% accuracy" claims from vendors typically apply only to unedited AI text — which is increasingly rare as students learn to revise their AI output.

Why can't any AI detector be 100% accurate?

AI detection works by identifying statistical patterns. AI text tends to have uniform sentence lengths, predictable word choices, and formulaic structure. Detectors measure these patterns and assign a probability.

The fundamental problem is that good human writing and good AI writing are converging. As AI models improve, their output becomes less distinguishable from human writing. And as students learn to edit AI output, the statistical signatures get weaker.

This isn't a solvable technical problem — it's an inherent limitation of the approach.

What does this mean for teachers using AI detectors?

If you're using AI detectors in your classroom:

Don't treat scores as proof. A high score means the text has patterns consistent with AI output. It does not mean the student used AI.

Look at flagged passages. A tool that shows you which sentences triggered detection and why is far more useful than a percentage.

Use detection as one signal among many. Combine it with your knowledge of the student's writing, their previous work, and a follow-up conversation.

Be especially careful with ESL students. The false positive rate for non-native English speakers is unacceptably high across all tools.

What does this mean for students?

If you wrote your essay yourself and it gets flagged:

Don't panic. False positives happen.

Be prepared to explain your writing process.

If you used writing tools like Grammarly, mention that — it can explain why your text reads more uniformly.

If you're concerned, check your own work before submitting. See what gets flagged and revise those sections.

Our approach

We built Is It AI? knowing that accuracy claims are meaningless without context. That's why we show:

Flagged passages with specific explanations of why each was flagged

Multiple detection dimensions — AI pattern analysis and statistical text analysis working together

Honest confidence levels — clear about when results are uncertain rather than forcing a verdict

A teacher who can see why a passage was flagged makes a better judgment than one who only sees "87% AI".

The bottom line

AI detectors are useful screening tools with real limitations. They're best at catching unedited AI text and worst at handling mixed, edited, or non-native-English writing.

Use them to identify text worth investigating. Don't use them to convict.

Try Is It AI? free — see flagged passages with explanations, not just a score.