Signs of AI writing: 12 AI tells in writing, measured

Which AI tells in writing actually give a text away? We compared 300 AI answers with human answers to the same questions. These are the patterns that actually separate them in 2026, with the numbers. Paste any text below to see its slop score and which tells it contains.

0 words · try a sample

Free: 3 AI rewrites a day (up to 300 words) and unlimited quick clean. Need more? A $5 pack of 50 rewrites never expires, or go Pro at $12/month · $79/year. See plans.

The 12 signs

An AI tell in writing is a habit that shows up far more often in model output than in human text on the same topic. One tell proves nothing; several together are what readers notice. For the patterns that changed in 2026, see Wikipedia's updated list and our em dash test, or paste text into the AI slop detector.

  1. Bold labels everywhere. 32.68 bold spans per 1,000 words in AI answers vs 0 in human ones.
  2. Bullet lists for everything. 29.24 list lines per 1,000 words vs 0.03.
  3. Headings in a short answer. 6.48 per 1,000 words; humans: 0.
  4. Em dashes. 4.5 per 1,000 words vs 0.21. Data by model.
  5. The closing offer. "If you tell me…", "If you want, I can…": 34.1× the human rate, in 11.3% of AI answers.
  6. "It's important to…" 16.6× the human rate.
  7. Stacked hedges. "may" (4.3×), "often", "typically", "can sometimes", "especially if".
  8. "Here are some…" list intros: 12.2×.
  9. Abstract helper verbs. "can help", "can lead to", "can provide", "contribute to", "ensure".
  10. Generic qualifiers. "specific", "common", "key", "practical", "crucial", "beneficial", "tailored".
  11. Colons as scaffolding. 19.52 colons per 1,000 words vs 3.24.
  12. A summary nobody asked for. "Overall," and "In summary," at the end of short answers.

What's no longer a reliable sign

"Delve", "tapestry" and "testament" were the 2023 tells. In 67,773 words of 2026 AI answers we found no "delve" at all. Sentence length was also nearly identical (AI 20.1 words on average, humans 19.8). See the full AI words list.

FAQ

Can you tell for sure if something was written by AI?

No. Every sign on this page also shows up in human writing, and AI detectors make frequent mistakes, especially on non-native English writers. Treat signs as hints, never as proof.

What is the most reliable sign of AI writing?

In our data, formatting: bold labels, bullet lists and em dashes. Modern AI answers had 32.68 bold spans per 1,000 words; human answers had 0.

Do AI detectors work?

Not reliably enough to accuse anyone. They produce false positives on human text and are easy to fool, which is why gounslop focuses on making writing better rather than on detection.

How do I make my writing sound less like AI?

Cut the hedges and offers, drop bold labels and bullet lists where prose works, replace em dashes, use specific details, and vary sentence length. The unslop tool does a first pass for you.