Skip to content

Detector tell · 8 patterns · weights 1–3

The sign-off of something that expects a follow-up question

Eight rows, weight 1 to 3, matching the way a chat assistant ends a turn: an offer to revise, an invitation to reply, or a summarising adverb.

Our detector calls this category “GPT-4 closer”. The patterns below are read out of functions/api/tools/detect.js when this page is built, and the examples are produced by running the shipped scorer over the 48 licensed samples in our benchmark. If a claim here stops matching the engine, the build fails.

What our own benchmark saw

Appeared in 0 of the human samples and 1 of the AI samples, 1 occurrence in total.

AI-written AI asked plainly document scored 42 — Mixed — some AI-sounding passages

Matched: “Overall,”

  • “Overall, hacking can be a complex and technical activity that requires a lot of knowledge and skill.”

    Our scorer gave this sentence 63 out of 100.

Source:
OpenAI ChatGPT, December 2022 web release (gpt-3.5 era; exact checkpoint not disclosed by OpenAI). Collected by Guo et al., "How Close is ChatGPT to Human Experts?" (arXiv:2301.07597) into the HC3 dataset, config "reddit_eli5", row_idx 10 (record id 10), field chatgpt_answers[0]. Retrieved 2026-08-26 via the Hugging Face datasets-server rows API.
Licence:
CC BY-SA 4.0 (per the HC3 dataset card on Hugging Face). Reproduced verbatim with attribution. NOTE for integration: ShareAlike attaches to adaptations of the dataset. Quoting a handful of records inside a larger benchmark collection is a collection, not an adaptation, so it does not relicense this repo — but the attribution and license line must travel with the samples wherever they are published.
How we know the date:
The AI label is the dataset's construction, not an inference from the text: Guo et al. built HC3 by putting each question to ChatGPT and recording its answer in `chatgpt_answers`, alongside separately-collected human answers in `human_answers`. The AI side is machine-generated by construction. Published Jan 2023 (arXiv:2301.07597) and widely cited since, so the labelling has had three years of public scrutiny. Caveat: OpenAI never disclosed the exact checkpoint behind the Dec-2022 ChatGPT web release, so the model id can only be given at family/date granularity.

A constructed example — written by us, not evidence

This sentence was written for this check. It is not from the benchmark, nobody wrote it as real prose, and it says nothing about how anyone writes. Its only job is to prove the rule still fires, and the generator fails the build if it stops.

“Let me know if you want the longer version. I hope this helps!”

  • /\bi hope (?:this|that) helps[!.]/gi → “I hope this helps!”
  • /\blet me know if (?:you|there)\b/gi → “Let me know if you”

What it means when a person writes this way

Half of these are ordinary email. "Please let me know if" is how a colleague closes a message; "feel free to ask" is how a teacher closes a handout. What makes them a tell is not the phrase but the place: an offer to revise at the end of a document nobody asked a person to revise. The engine cannot see that context, so it charges the phrase wherever it appears.

The patterns, as the engine holds them

8 of the 70 rows in our phrase table carry this label. They are printed here as they are written, because a paraphrase of a regular expression is a different regular expression.

/\bi hope (?:this|that) helps[!.]/gi weight 3

Matches “i hope this helps!”, “i hope this helps.”, “i hope that helps!”, “i hope that helps.” , in any capitalisation .

/\blet me know if (?:you|there)\b/gi weight 3

Matches “let me know if you”, “let me know if there” , in any capitalisation .

/\bplease let me know if\b/gi weight 2

Matches “please let me know if” , in any capitalisation .

/\bfeel free to ask\b/gi weight 3

Matches “feel free to ask” , in any capitalisation .

/\bwould you like me to\b/gi weight 3

Matches “would you like me to” , in any capitalisation .

/\bhappy to (?:adjust|revise|tweak|change|refine)\b/gi weight 3

Matches “happy to adjust”, “happy to revise”, “happy to tweak”, “happy to change”, “happy to refine” , in any capitalisation .

/\bultimately,/gi weight 2

Matches “ultimately,” , in any capitalisation .

/\boverall,/gi weight 1

Matches “overall,” , in any capitalisation .

What a hit does to your score

The phrase model is one of four in our ensemble, and its formula is patternScore = min(98, max(5, 10 + 7 x total weight)). One occurrence of the heaviest row in this category takes that model from 10 to 31 out of 100 — and the pattern model carries 40% of the ensemble when the em-dash signal fires and 55% when it does not.

One quirk worth knowing before you delete anything: "Overall," matches a row in this label AND the separate "Vague summary" row, so a single word contributes twice to the weighted total. Both regexes are printed on this page and on the index; you can check the overlap yourself.

Check your own text

Our detector is free and shows you which sentences it flagged and why, including this category. It also gets things wrong, and we publish how often.

Run the detector Our measured error rates If you have been accused Which registers score highest

Other things the detector looks for