Researchers and applied teams need faster, reproducible ways to turn interview transcripts and network diagrams into actionable findings; the primary keyword is AI qualitative analysis criminal justice. According to Connell et al., PLOS One (2026), people in contact with the criminal justice system living in the community face intersecting mental health and substance use problems that are shaped by individual, relational, and contextual forces. This post explains, for qualitative researchers and program teams, how the PLOS One study (published 30 July 2026) can be re-analyzed and scaled using AI-enabled qualitative research methods to surface relational drivers of help-seeking faster and with stronger segment cross-tabs.
Key Takeaways
Direct answer: According to PLOS One, help-seeking for mental health and substance use among people with community criminal justice contact is shaped by three interlocking domains (desire, ability, and context) and those domains are strongly influenced by social networks and local culture.
- Connell et al., PLOS One (published 30 July 2026) interviewed 50 people between 1 November 2023 and 31 July 2024 (23 in LA1, 27 in LA2) and derived a framework of 'desire', 'ability', and 'help-seeking context'.
- Connell et al., PLOS One (2026) note population prevalence estimates cited in their introduction commonly report over 40% meeting criteria for a substance use disorder, more than 10% for major depressive or psychotic disorders, and about 75% for any personality disorder in similar justice-involved groups.
- Connell et al., PLOS One (2026) report that most participants had at least one unsuccessful help-seeking attempt and that relational influences across networks often normalized problems or discouraged help-seeking.
What happened: study design and core findings
Answer-first: Connell et al., PLOS One (2026) collected qualitative interviews from 50 adults with recent criminal justice contact and used framework analysis to show help-seeking results from interactions between desire, ability, and context, with strong relational influences.
Connell et al., PLOS One (2026) explain that interviews were audio-recorded and conducted using Network Canvas software, and that data collection ran from 1 November 2023 to 31 July 2024.
Connell et al., PLOS One (2026) report 50 participants were interviewed (23 in LA1, 27 in LA2) and that interviews lasted up to two hours, producing pseudonymised transcripts and linked social network maps used in framework analysis.
Connell et al., PLOS One (2026) conclude that help-seeking is rarely purely individual: participants described network norms that normalized substance use, epistemic mistrust toward services, and local stigma that made seeking help risky.
Direct participant evidence: participant P13 in Connell et al., PLOS One (2026) said, "you just kind of can’t get away from the social, " and participant P12 said, "I have dealt with it for almost twenty odd years by myself… I don’t like going to doctors because I think they are a waste of time...."
Findings snapshot
| Date / Source | Metric | Value | Implication |
|---|---|---|---|
| Published 30 July 2026, PLOS One | Interviews completed | 50 participants (23 LA1, 27 LA2) | Framework built from first-hand accounts across two contrasting localities |
| Interview period 1 Nov 2023–31 Jul 2024, Connell et al., PLOS One (2026) | Data type | Audio-recorded interviews plus social network maps | Enables combined thematic and relational analysis |
| Connell et al., PLOS One (2026) literature summary | Typical prevalence (cited studies) | Over 40% substance use disorder; >10% major depressive/psychotic; ~75% personality disorder | High unmet need reinforces urgency for effective help-seeking interventions |
Implications for qualitative researchers and program teams
Answer-first: Researchers and program teams should treat help-seeking as a multi-level, network-shaped outcome and design qualitative studies and interventions that capture relational data, as Connell et al., PLOS One (2026) demonstrate.
Connell et al., PLOS One (2026) show that social network structure, shared norms, and local service availability meaningfully alter whether participants desire or are able to seek help, so mixed-methods designs that combine interview transcripts and network data produce richer, actionable findings.
Connell et al., PLOS One (2026) recommend multi-level interventions that go beyond individual-focused approaches; program teams should therefore prioritize community-level trust-building, peer-credible messaging, and service navigation supports that accommodate unstable housing, variable schedules, and mistrust.
How Evidano helps apply the PLOS One framework
Problem: Large interview sets + network maps are slow to synthesize
Answer-first: Evidano speeds synthesis of interviews and participant-aided sociograms by combining transcription, thematic coding, and cross-segment analysis.
Evidano is an AI-powered qualitative data analysis platform that helps researchers analyze interviews, open-ended surveys, and documents.
Feature mapping: where Connell et al., PLOS One (2026) used Network Canvas and manual NVivo coding, Evidano ingests audio and network files, performs automated transcription with custom dictionary support, and generates thematic and co-occurrence networks to show where 'desire', 'ability', and 'context' themes co-occur.
Problem: Need to compare themes across localities and network types
Answer-first: Evidano provides cross-segment frequency analysis and thematic matrices to compare LA1 vs LA2 and to link participant quotes to network positions.
Practical gain: researchers can filter transcripts by segment (for example, participants with small closed networks vs diverse networks identified in Connell et al., PLOS One (2026)) and export matrices that mirror the study’s coding matrix for rapid triangulation.
Relevant Evidano link: see Evidano features for thematic coding and visualization details.
Problem: Handling audio quality, PII, and lived experience quotes securely
Answer-first: Evidano supports secure encrypted storage, PII redaction in transcripts, and controlled access to pseudonymised quotes for publication and LEAP review.
Operational detail: teams re-analyzing Connell et al., PLOS One (2026) style datasets can upload recordings and apply custom redaction dictionaries and export quote lists for advisory panels while preserving source linkage for audit.
Technical link: see Evidano speech-to-text for transcription and redaction options.
FAQ: AI qualitative analysis criminal justice
How can AI qualitative analysis help apply the Connell et al., PLOS One (2026) framework?
Direct answer: AI qualitative analysis can accelerate coding, surface co-occurring themes, and map quotes to network attributes to operationalize the 'desire, ability, context' framework described by Connell et al., PLOS One (2026).
Supporting detail: Connell et al., PLOS One (2026) combined interviews and social network maps; AI tools can ingest both text and structured network metadata to produce cross-tabulations and node-level quote lists that researchers can validate.
Can AI reproduce the nuance of participant quotes like those reported by Connell et al., PLOS One (2026)?
Direct answer: AI can reliably identify and group verbatim participant language, but human validation is still required to preserve interpretive nuance.
Supporting detail: Connell et al., PLOS One (2026) used double-coding and team review to ensure interpretive rigor; AI outputs should be used to prioritize segments for human review rather than to replace reflexive analysis.
Is reanalysis of sensitive justice-involved interview data ethical with AI?
Direct answer: Reanalysis is ethical when data are pseudonymised, access is controlled, and participants consent permits future research, consistent with Connell et al., PLOS One (2026).
Supporting detail: Connell et al., PLOS One (2026) deposited pseudonymised transcripts under controlled access at the UK Data Service and required approvals for data access, a model AI-enabled projects should emulate.
How quickly can a team generate a matrix of themes by network-position using AI?
Direct answer: With prepared transcripts and network metadata, an AI-enabled pipeline can produce an initial theme-by-network-position matrix in hours, not weeks.
Supporting detail: Connell et al., PLOS One (2026) describe building a matrix with cases as rows and themes as columns; AI can automate extraction of coded segments into the same matrix format for researcher validation.
Conclusion & Next Steps
Recap: Connell et al., PLOS One (2026) demonstrate that help-seeking among people with community criminal justice contact is shaped by desire, ability, and context, and that relational network influences are central.
Actionable next step: Qualitative teams can combine interview transcripts and participant-aided sociograms and accelerate framework-driven analysis using AI-enabled tools to produce validated thematic matrices and targeted recommendations for services.
Learn more and experiment: if your team wants to reanalyze interview-plus-network datasets like Connell et al., PLOS One (2026), Try Evidano for free to test fast transcription, thematic extraction, and cross-segment visualizations.
Ethics note: Reanalysis of justice-involved data should follow the consent and access controls described by Connell et al., PLOS One (2026) and use secure platforms.