CoChat

Teachers vs. Chat GPT: identifying and naming acts of online sexual abuse

Open access
Dolev-Cohen, Michal, Henry, Nicola · 2025 · Information, Communication & Society

The DOI matches its Crossref record; 89 of the 91 references it cites were found in a registry or an index; 4 of the 15 sources its claims cite were checked for support.

Abstract

Digital media and social networks offer adolescents a space for various social and sexual interactions. While the internet fosters healthy sexual exploration, it also facilitates online sexual abuse, such as non-consensual intimate image creation or dissemination, sextortion, sexual harassment, grooming, and gender-based hate speech. This abuse often goes underreported, especially when parents and teachers minimize or blame victims, or fail to recognize the abuse. This study examined how Israeli teachers and artificial intelligence (AI) identify and attribute blame in cases of sexual abuse. Six vignettes, depicting various sexual scenarios, were presented to 153 teachers and analyzed by ChatGPT-3.5 and ChatGPT-4. Findings revealed that teachers struggled to identify online sexual abuse and often blamed victims. ChatGPT-3.5 also tended to blame victims, but ChatGPT-4 accurately identified the abuse without victim-blaming. The study suggests GPT-4 can be an effective tool for identifying and naming sexual abuse when teachers lack proper training.

Overview

CoChat Summary

TL;DR

This study compared Israeli teachers with ChatGPT-3.5 and ChatGPT-4 in identifying, naming, and assigning blame in six vignettes about adolescent sexual scenarios. Teachers struggled to identify online sexual abuse and often blamed victims; GPT-3.5 showed similar victim-blaming, while GPT-4 accurately identified the abuse without blaming victims.

Key findings

  • Teachers correctly named physical rape in 98% of cases but performed poorly on online abuse: sextortion 33.3%, non-consensual image sharing 60.1%, revenge-based image sharing 59.5%, online grooming 8.5%, and consensual sexting 23.5%. GPT-4 correctly named all vignettes at 100%; GPT-3.5 was mixed (e.g., sextortion 100%, revenge-based image sharing 100%, but sexting 20% and grooming 40%).
  • Teachers had significant difficulty identifying online grooming as sexual abuse: only 16.3% said yes, 63.4% said no, and 20.3% were unsure. GPT-3.5 and GPT-4 both identified grooming as sexual abuse in 100% of responses. For consensual sexting, teachers said it was sexual abuse in 21.6% of cases, GPT-3.5 in 50%, and GPT-4 in 0%.
  • Teachers assigned some responsibility to the victim in all online abuse cases: sextortion 78.4%, non-consensual image sharing 79.1%, online grooming 93.5%, and revenge-based image sharing 71.2%. GPT-4 assigned the victim responsibility in 0% of cases across these vignettes; GPT-3.5 blamed victims in 50% of sextortion, 90% of grooming, and 40% of revenge-based image-sharing responses.
  • For physical rape, teachers assigned the victim some responsibility in 28.8% of cases, GPT-3.5 in 10%, and GPT-4 in 0%.

Methods

  • Between December 2023 and February 2024, 153 teachers from across Israel were recruited via social media and WhatsApp groups and completed a questionnaire based on six vignettes: physical rape, consensual sexting, sextortion, non-consensual image distribution, revenge-based image distribution, and online grooming. All teacher participants were women; the sample was 69.3% Jewish and 30.7% Arab.
  • For each vignette, participants answered three questions: labeling the event, judging whether it was sexual abuse, and assigning responsibility to victim and abuser on a 5-point Likert scale later dichotomized.
  • In February 2024, GPT-3.5 and GPT-4 each received the same vignettes and questions 10 times in new chat windows. Chi-square tests compared teachers with the AI models using SPSS v27; tables report N = 173, reflecting 153 teachers plus 20 AI evaluations.

Limitations

  • The paper states the sample was relatively small, recruited by convenience sampling, and all participants were women; future research should include larger diverse samples, male and non-binary educators, and analysis by religion or other identity factors.
  • The paper notes all vignettes had boys as abusers and girls as victims, so future research should examine abuse against boys and non-binary youth by offenders of different genders. It also notes possible self-selection bias, since respondents were teachers willing to engage with sexual content, and that more conservative populations might show lower naming ability and higher victim-blaming.
  • Evident from the text: AI performance was based on only 10 measurements per vignette per model, and the study used vignettes developed specifically for this research rather than a broader validated stimulus set.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1080/1369118x.2025.2492585

Cites

89 of 89 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Image 1
CC-BY-NC-ND · From the PDF
Image 2
CC-BY-NC-ND · From the PDF
Image 3
CC-BY-NC-ND · From the PDF
Figure 1 Differences between teachers, GPT-3.5, and GPT-4 in correctly naming the phenomena pre­ sented in the vignettes.
Figure 1 Differences between teachers, GPT-3.5, and GPT-4 in correctly naming the phenomena pre­ sented in the vignettes.CC-BY-NC-ND · From the PDF
Figure 2 Difference between teachers, GPT-3.5, and GPT-4 in identifying sexual abuse in phenomena presented in the vignettes.
Figure 2 Difference between teachers, GPT-3.5, and GPT-4 in identifying sexual abuse in phenomena presented in the vignettes.CC-BY-NC-ND · From the PDF
Figure 3 Difference between teachers, GPT-3.5, and GPT-4 in assigning responsibility for sexual abuse.
Figure 3 Difference between teachers, GPT-3.5, and GPT-4 in assigning responsibility for sexual abuse.CC-BY-NC-ND · From the PDF

Verification

Claims the paper makes

4 Valid · 0 Not valid · 6 Not checked · 48 Own results and interpretations

30 more claims

Create a free account to see all 58 claims and to verify the 3 references still open.

See all claims — free account
  • Probably not a substantial claim. 13
  • It may state two separate claims as one. 16
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

5 of 6 · the rest aren’t loaded

1 more finding

Create a free account to see every finding of this check.

See all findings — free account

Citations

Cites

Show all 89 references

Research works printed but not linked: 1

  1. Ezioni, L. (2020). The crime of grooming. Child and Family Law Journal, 8(1), 1–18. https://lawpublications.barry.edu/cgi/viewcontent.cgi?article=1041&context=cflj

Cited by

See all 8

Metadata

Authors
Dolev-Cohen, M., & Henry, N.
Affiliations
  • Oranim Academic College of Education (IL)
  • RMIT University (AU)
Corresponding author
Dolev-Cohen, Michal
Venue
Information, Communication & Society
Published
2025-04-17
Volume
28
Issue
16
Pages
2976-2994
ISSN
  • 1369-118X (ISSN-L)
  • 1468-4462
OpenAlex
W4409531005
Cited by
8
References
91
Citations by year
2026: 6 · 2025: 2
FWCI
14.58
Type
journal article
Language
en
Open access
Locations
Information Communication & Society (publishedVersion, cc-by-nc-nd)
Topics
  • Social Sciences › Gender Studies › Sexual Assault and Victimization Studies
  • Social Sciences › Education › Artificial Intelligence in Education
  • Social Sciences › Psychology › Clinical Psychology › Child Abuse and Trauma
Keywords
online sexual abuse, teachers, ChatGPT, blame attribution, victim blaming, sextortion, sexual harassment, online grooming, gender-based hate speech, teacher professional development, adolescents, Israeli teachers

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Teachers vs. Chat GPT: identifying and naming acts of online sexual abuse. https://app.cochat.ai/papers/10.1080/1369118x.2025.2492585
The paper
Dolev-Cohen, M., & Henry, N. (2025). Teachers vs. Chat GPT: identifying and naming acts of online sexual abuse. Information, Communication & Society. https://doi.org/10.1080/1369118x.2025.2492585

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.