How to Tell if Someone Used ChatGPT: The Detection Checklist Teachers Use

AI detection tools like GPTZero and Turnitin's AI checker have improved but still hit false positives and false negatives every day. The most reliable detection in 2026 is still human — a checklist teachers, editors, and hiring managers actually use.

The high-signal writing tells

Related: GPT-5 Is Here: Everything You Need to Know About OpenAI's Most Powerful Model Yet →

These show up in AI-written text far more often than in human writing:

• "It's important to note that..." and its cousins ("in today's fast-paced world", "in the ever-evolving landscape of...")
• Bullet lists with exactly the same length — every point three lines long
• "Delve", "leverage", "tapestry", "myriad", "landscape", "boasts" — the AI vocabulary tell
• Perfect grammar with zero personality — no contractions, no fragments, no rhythm changes
• Even section lengths — every paragraph clocks in near the same word count

Any two of these together is a soft flag; three or more is strong.

The structural tells

Related: Will AI Coding Agents Replace Developers? We Asked 100 Engineers →

• Introductions that restate the prompt back to you
• Conclusions that begin with "In conclusion" or "Overall"
• Section headings phrased as questions when nothing else in the doc asks questions
• Numbered lists that suddenly appear in the middle of prose

The knowledge tells

Related: ChatGPT vs Claude 4: Which AI Should You Actually Pay For in 2026? →

• Facts that are correct but suspiciously general (no specifics, no examples, no named sources)
• Citations to books/papers that don't exist (hallucinated references)
• Dates that are consistently "in recent years" or "as of the current time"
• No personal anecdotes when the assignment asks for them

Compare to their earlier work

Related: Google Gemini 3 Ultra Review: Has Google Finally Caught Up? →

The single most reliable check: compare to something the person wrote before. Look for:

• Vocabulary suddenly a level higher
• Sentence structure suddenly more uniform
• Voice suddenly more formal or more generic
• Errors they always make (a specific typo, a favorite phrase) suddenly absent

Use detectors as evidence, not proof

Detectors are useful but should never be the whole case. Best practice:

1. Run through GPTZero, Turnitin, and Copyleaks — they disagree often, and agreement is meaningful.
2. Cross-check with the human tells above.
3. If it matters, ask the writer to explain a specific paragraph in person or in writing. Real authors can. AI-only work usually can't.

Common false-positive traps

• Non-native English writers often score as "AI-generated" because their sentence structure is more uniform. Do not use detectors as the sole evidence in these cases.
• Templated formats (lab reports, legal briefs, resumes) naturally look "AI-like" because they follow strict conventions.
• Short samples (under 200 words) are unreliable for any detector.

The pragmatic policy

For educators and managers, the useful stance isn't "prove they used AI" — it's "ask them to demonstrate the underlying work." A brief oral defense, a live edit, or a follow-up question makes AI-only submissions collapse quickly and doesn't punish honest work.

That's the checklist. Combine the writing tells, structural tells, comparison to previous work, and detector cross-checks — and you'll be right far more often than any single tool.