Qualitative Analysis of AI-Assisted Abuse

Evidano is an AI-powered qualitative data analysis platform that helps teams securely code, analyse, and turn disclosures of AI-assisted abuse into action. Fast, defensible qualitative analysis of AI-assisted online sexual abuse is now urgent: a July 6, 2026 nationally representative Australian study of 1, 894 16–18 year olds found about 1 in 25 teens have been directly affected by AI-generated sexual imagery (ABC News). For researchers, UX teams and policy analysts, this post explains how to run rigorous qualitative analysis of AI-assisted abuse disclosures, identify themes, compare segments (gender, peer group), and produce secure, shareable outputs. Read on to learn a 7-step workflow and how Evidano maps to each step so you can turn sensitive disclosures into action without sacrificing ethics or data security.

Key Takeaways

Evidano is an AI-powered qualitative data analysis platform that helps teams securely code, analyse, and turn disclosures of AI-assisted abuse into action. The July 6, 2026 Australian study of 1, 894 16–18 year olds found about 1 in 25 teens were directly affected by AI-generated sexual imagery, and 19% of respondents disclosed to AI chatbots versus 15% to police or teachers.

Use explicit coding for AI involvement, measure disclosure pathways (including chatbots), and run a secure workflow to convert sensitive data into referrals and policy-ready outputs.

About 15% of respondents reported some form of online child sexual victimisation in the 2026 study, and roughly one-quarter of those incidents involved AI-generated imagery, giving the approximate headline of 1 in 25 teens affected.
Disclosure channels matter: 66% disclosed to friends, 43% to parents, 19% to AI chatbots and 15% to authorities, with chatbots more commonly used than police or teachers.
Gender and clustering patterns are important: 19% of girls and 11% of boys reported victimisation overall, while AI-involvement was proportionally higher among boys (27% vs 9% for girls) and peer clusters show increased risk among friendship groups.

Findings Snapshot

Metric	Value	Source / Note
Published	6 July 2026	Australian nationally representative survey (see original)
Sample	1, 894 respondents aged 16–18	University of Adelaide collaboration with ICMEC & AFP
Any online child sexual victimisation	15% of respondents	Includes non-consensual image sharing & solicitation
AI involvement (of victimised cases)	≈25% (≈1 in 25 of all teens)	Deepfakes/nudify services; likely a minimum estimate
Friend experienced AI-assisted victimisation	≈1 in 9 (~11%)	Peer clustering noted
Disclosure recipients (most common)	Friends 66% · Parents 43% · AI chatbots 19% · Authorities 15%	AI chatbots more common than police/teachers
Weekly AI help-seeking	>20% asked for psychosocial counselling from AI weekly; 10% disclosed hurt weekly	Study-reported frequencies

What happened (plain English)

The study surveyed 1, 894 Australian teens (aged 16–18) in a nationally representative sample completed in 2026 and asked about non-consensual sexual image sharing, adult sexual solicitation, and whether AI was involved.

15% reported some form of online child sexual victimisation.
About one-quarter of those incidents involved AI-generated imagery or services (deepfakes, nudify tools), producing the headline figure of approximately 1 in 25 teens.
Gender patterns shifted: 19% of girls and 11% of boys reported victimisation overall, while a larger share of boys' experiences involved AI (27% vs 9% for girls).
AI chatbots are now a common help-seeking channel: 19% disclosed to AI, higher than the 15% who disclosed to police or teachers.

So what for researchers, UX teams, and policy analysts

For qualitative researchers

Qualitative researchers should treat AI involvement as a coded variable in every study of online harm.

This is a signal to add explicit probes: ask Was AI used? Which service? Did you seek help from a bot? and code for disclosure channel.

Researchers should expect cluster effects: if one teen in a friendship group is affected, others are likelier to be too, and design recruitment and consent processes accordingly.

For UX & product teams

UX and product teams should design chatbot help-flows that escalate disclosures to human support and include clear referral pathways.

Users are already disclosing trauma to chatbots, therefore design help-flows that escalate disclosures to human support and include clear referral pathways rather than a generic canned response.

Test chatbot responses with survivors in controlled research settings and record qualitative feedback to refine scripts and escalation.

For policy & safeguarding teams

Policy and safeguarding teams should require service-level duties of care for referrals and prioritise measurement of help-path effectiveness.

Reporting-by-design is incomplete: new eSafety rules require reporting mechanisms but not mandatory referral for disclosures, so policymakers should require service-level duties of care for referrals.

Prioritise measurement: track not just prevalence but help-path effectiveness (did disclosure to AI lead to services?).

Do more, faster with Evidano

Evidano speeds workflows by ingesting diverse inputs

Evidano ingests interview transcripts, chat logs, survey spreadsheets, and incident reports into a single corpus with metadata (age, gender, disclosure channel).

Capture AI involvement as a coded dimension

Evidano lets teams import a codebook or use AI-assisted code suggestions to tag 'AI involvement', 'disclosure channel', 'requested support', and other variables reliably across documents.

Transcription, translation & PII controls

Evidano can auto-transcribe interviews with a custom dictionary and perform PII redaction to protect survivors.

Evidano can translate multilingual disclosures while preserving coded tags.

Thematic and cross-segment analysis

Evidano runs thematic and frequency analyses (themes, co-occurrence networks, hierarchical code → subcode views) and slices results by gender, peer-cluster, or disclosure channel to surface patterns such as higher AI-involvement among boys.

Secure, research-first model

Evidano stores data encrypted and does not use sensitive data to train third-party models, which supports institutional ethics requirements when handling disclosures of sexual harm.

From insight to referral

Evidano provides AI chat-over-documents to synthesise guidance for frontline staff and to export stakeholder-ready visuals and quote packs for warm handoffs to services.

FAQ: Qualitative analysis of AI-assisted abuse

How do I reliably detect AI involvement?

Ask explicit survey and interview questions and code mentions of tools to reliably detect AI involvement.

Ask participants Was AI used? and which service, and code for keywords such as deepfake or nudify; supplement self-reports with pattern searches over uploads and image metadata where available.

Can I compare segments, for example boys versus girls?

Yes, you can compare segments using side-by-side thematic and frequency analyses.

Export segment filters to run side-by-side thematic and frequency analyses; the Australian study found a higher share of boys' cases involved AI.

How should we handle disclosures made to chatbots?

Treat disclosures to chatbots as data about help-seeking behaviour and evaluate whether chatbot responses include referrals.

Evaluate whether chatbot responses include referrals and whether users followed up with human services, and consider designing chatbot flows that escalate to human support.

Topics

AI-assisted abuse qualitative analysis
AI-generated imagery teens
disclosure to chatbots
qualitative coding workflows
secure qualitative data

Keep reading

Browse all articles

Qualitative Analysis of AI-Assisted Abuse

Key Takeaways

Findings Snapshot

Findings Snapshot

What happened (plain English)

So what for researchers, UX teams, and policy analysts

For qualitative researchers

For UX & product teams

For policy & safeguarding teams

Do more, faster with Evidano

Evidano speeds workflows by ingesting diverse inputs

Capture AI involvement as a coded dimension

Transcription, translation & PII controls

Thematic and cross-segment analysis

Secure, research-first model

From insight to referral

Ethics note

FAQ: Qualitative analysis of AI-assisted abuse

How do I reliably detect AI involvement?

Can I compare segments, for example boys versus girls?

How should we handle disclosures made to chatbots?

Wrapping up & next steps

Topics

Keep reading

Company

About

Preview Evidano

Account

Newsletter