What Is an AI Detector?

If you have sent an essay through Turnitin recently, you have already met an AI detector face to face. Behind the "AI Generated" label sits a piece of software that tries to guess whether every sentence was typed by a human or spat out by a model. Understanding how the guessing works is the first step toward knowing how much weight to give it.

What detectors actually measure

Most commercial detectors rely on two core signals: perplexity and burstiness.

Perplexity

Perplexity captures how predictable each word is given the words that came before. A sentence like "The quarterly revenue increased by 12 percent" is highly predictable to a language model, so it scores low on surprise. Human writers constantly reach for unusual phrasing, odd metaphors, or sentence fragments that would make a language model pause.

Think of it this way: if you can guess the next word in a sentence before you finish reading it, the perplexity is low. If the writer surprises you, the perplexity is high.

Burstiness

Burstiness tracks the rhythm of a paragraph. People tend to write in bursts: a long clause, then a short one, then a run-on sentence that trails off. Language models produce a more uniform cadence. Detectors look for that flattening.

Key insight: Human writing is rhythmically uneven. AI writing is rhythmically flat. That difference is one of the strongest signals detectors use.

Token probability

A third signal that many systems now incorporate is the probability distribution the model assigns to its own top tokens. Researchers call this log probability or token-level entropy. If the model is extremely confident about every token, the text is likely something the model could have produced itself. If the model hesitates, swapping between many possible next words, the text is more likely to be human.

A 2024 study from researchers at Stanford found that token-level entropy is one of the strongest single features for classification, outperforming perplexity alone on several benchmarks.

The machine-learning layer

Perplexity and burstiness feed into a classifier—usually a logistic regression model or a small neural network. That classifier was trained on pairs of human-written and AI-generated texts.

• Turnitin has said its training corpus includes tens of thousands of essays across multiple disciplines
• GPTZero uses a mix of academic papers, news articles, and its own synthetic data
• Originality.ai trains on content from the open web

The training data shapes what the detector considers "human" and "AI." If a detector was trained mostly on academic prose, it may flag blog posts or conversational copy simply because the writing style falls outside its training distribution. This is the root cause of many false positives.

Where false positives come from

A 2025 Stanford study analyzed over 1,200 essays across six universities and found that detectors labeled roughly 6.1 percent of human-written work as AI-generated.

The gap exists because ESL prose often uses more standardized grammar, shorter sentences, and a narrower vocabulary—the exact patterns detectors associate with AI.

• Turnitin has published its own internal false-positive rate of 3.8 percent on English-language essays
• ZeroGPT's rate in independent tests has exceeded 14 percent

These numbers matter because a single false positive can trigger an academic integrity investigation.

Genres that cause problems

Detectors also struggle with certain genres:

• Legal contracts follow rigid templates
• Medical abstracts use standardized structures
• Press releases follow the inverted-pyramid format
• Literature reviews repeat the same paragraph structure

When a student writes a literature review using the same structure every time, the detector sees low perplexity and flags it. The detector is not wrong about predictability; it is wrong about the cause.

The missing context problem

AI detectors give you a score, not an explanation. They do not:

• Tell you which sentences triggered the flag
• Account for Grammarly or AI autocomplete usage
• Distinguish between copying an entire ChatGPT output and using AI to brainstorm then rewriting by hand

Both scenarios might produce text that looks statistically similar to AI output because the student internalized the model's phrasing during the drafting process.

Why the tools still matter

Despite their flaws, detectors serve a real purpose. At scale, they give educators a triage tool.

A professor with 200 essays cannot read every one with the same level of scrutiny. A detector that flags 20 of those essays lets the professor focus attention where misconduct is more likely.

Universities that treat detector output as one signal among many—alongside writing samples, drafts, and oral examinations—get reasonable results. Problems arise when institutions treat the score as a verdict rather than a hint.

Types of AI detectors

AI detectors come in two main forms: standalone tools and integrated platforms. Understanding the difference matters because they serve different use cases and offer different levels of reliability.

Standalone detectors

Standalone detectors are independent tools you access through a website or API. You paste text in, get a result out. Examples include GPTZero, ZeroGPT, and Copyleaks. These tools are designed for quick checks and are typically used by individuals or small teams.

Advantages:

• No setup or integration required
• Free or low-cost tiers available
• Quick enough for routine checks

Limitations:

• Character limits on free tiers
• No historical tracking
• Limited to the text you paste in

Integrated detectors

Integrated detectors are built into larger platforms. Turnitin is the most prominent example—it is embedded directly into learning management systems like Canvas, Blackboard, and Moodle. When a student submits an essay through Turnitin, the detector runs automatically.

Advantages:

• Seamless workflow for educators
• Historical data across submissions
• Bulk processing capabilities

Limitations:

• Requires institutional adoption
• Higher cost per check
• Less control over detection parameters

Hybrid approaches

Some tools combine both models. Originality.ai offers a standalone web interface plus an API for integration into content management systems. This flexibility makes it popular with content teams who need to check both individual articles and bulk uploads.

How to interpret detection scores

A detection score is not a verdict. It is a probability estimate. Understanding what the numbers mean prevents misinterpretation.

The score breakdown

Most detectors return a percentage representing the likelihood that the text was AI-generated. Here is how to read common score ranges:

What the score does not tell you

• Which sentences triggered the detection
• How much of the text is AI-generated versus human-written
• Whether AI was used for brainstorming, outlining, or full generation
• Which AI model produced the text

A score of 60% does not mean 60% of the text is AI-generated. It means the detector estimates a 60% probability that the text was produced by an AI model. The distinction matters.

Confidence intervals

Advanced detectors like Turnitin now provide confidence intervals alongside their scores. A score of 75% with a confidence interval of ±5% means the detector is relatively certain. A score of 55% with a confidence interval of ±20% means the detector is uncertain, and the result should be treated with caution.

The 2026 regulatory landscape

AI detection is no longer just an academic concern. Governments are starting to regulate how these tools are used, and the regulations have implications for educators, publishers, and tool developers.

The EU AI Act

The European Union's AI Act, which came into full effect in August 2026, classifies AI detection tools as high-risk AI systems when used in educational or employment contexts. Key requirements include:

• Transparency: Detectors must disclose that they are AI-powered and provide explanations for their outputs
• Human oversight: Detection scores cannot be the sole basis for academic integrity decisions
• Bias testing: Tool providers must demonstrate that their systems do not disproportionately flag specific demographic groups
• Appeal mechanisms: Institutions must provide clear processes for students to challenge detection results

United States

In the US, regulation varies by state. California and New York have introduced legislation requiring institutions to disclose when AI detection tools are used on student work. Several universities have faced lawsuits from students who were incorrectly flagged, creating a growing body of case law around the admissibility of detection scores.

United Kingdom

The UK's Quality Assurance Agency published guidance in January 2026 recommending a three-signal approach to academic integrity: detector score, comparison with the student's baseline writing, and contextual evidence. The guidance explicitly states that detection scores should never be used as standalone evidence.

Practical implications

These regulations mean that institutions using AI detection tools must:

1. Train staff on the limitations of detection technology 2. Establish clear policies for handling detection results 3. Provide students with information about how their work is being evaluated 4. Maintain audit trails of detection decisions

Practical tips for educators

If you are an educator using AI detection tools, these practices will help you get the most value while minimizing harm.

Use detection as a triage tool

A detector that flags 15 out of 200 essays lets you focus your attention where misconduct is more likely. That is a useful application. Using the same detector to determine guilt is not.

Compare with baseline writing

If you have access to the student's previous work, compare the flagged text with their established writing style. A dramatic shift in vocabulary, sentence structure, or argument complexity is a more reliable indicator than a detection score.

Talk to the student

Before making any accusations, have a conversation. Ask the student about their writing process. Many students use AI tools for brainstorming or outlining, which is legitimate. The conversation often reveals whether the student engaged with the material.

Document your process

Keep records of how you used the detection tool, what score was returned, and what additional steps you took. If the case escalates, a documented process protects both you and the student.

Stay current

Detection technology improves rapidly. A tool that was unreliable in 2025 may be significantly better in 2026. Revisit your tools and processes at least once per academic year.

Comparison of major detectors

Here is a side-by-side comparison of the most widely used AI detection tools in 2026:

No single tool is best for every situation. Academic institutions generally prefer Turnitin because of its integration with learning management systems. Content teams often use Originality.ai because it handles web content well. Individual writers frequently start with GPTZero's free tier.

Key takeaways

• AI detectors use perplexity, burstiness, and token probability to classify text as human or AI-generated
• No detector achieves 100% accuracy—false positives are a real risk, especially for ESL writers and formulaic genres
• Detectors come in standalone and integrated forms, each suited to different use cases
• Detection scores are probability estimates, not verdicts—they should be one signal among many
• The 2026 regulatory landscape requires human oversight and transparency in how detection tools are used
• Educators should use detection as a triage tool, not a final judgment
• Understanding detector limitations helps you make informed decisions about when and how to use them

Frequently asked questions

Can an AI detector falsely accuse me of cheating?

Yes. False positives occur at rates between 3% and 15% depending on the tool. ESL writers, non-native English speakers, and writers who use formulaic structures are disproportionately affected. If you are falsely flagged, request a comparison with your baseline writing and ask for an explanation of which specific passages triggered the detection.

Which AI detector is the most accurate?

Turnitin is often cited as among the strongest detectors for academic writing, but published accuracy figures vary by dataset and vendor. Accuracy depends on the type of text being analyzed. Originality.ai often performs better on web content, while Copyleaks offers stronger multilingual support.

Do AI detectors work on text from GPT-4o and Claude 3.5?

Detection accuracy varies by model. GPT-4o text is generally easier to detect because it has a more predictable statistical profile. Claude 3.5 and Gemini 1.5 are harder to detect because they produce more varied phrasing. Detectors continuously retrain to keep up with new models.

Should I use multiple detectors?

Using two detectors as a cross-check can improve reliability. If both detectors flag the same text, the result is more trustworthy than a single detector's output. However, no combination of detectors eliminates false positives entirely.

How does the EU AI Act affect AI detection in schools?

The EU AI Act requires that AI detection tools used in educational contexts provide transparency about how they work, allow human oversight in decision-making, and undergo bias testing. Institutions must also provide appeal processes for students who are incorrectly flagged. These requirements are designed to prevent misuse of detection technology.

Try it yourself

Want to see how detection works in practice? Test your text with Vortixy's free AI detector and get an instant analysis of whether content reads as human or AI-generated. If you need to adjust flagged text, the AI humanizer can help you rewrite it naturally.