Online Grooming: Linguistic Patterns and Evidence

Kenneth Lanning, a retired FBI supervisory special agent, developed one of the earliest analytical frameworks for understanding child sexual exploitation, noting that offenders typically work through recognisable stages. Rachel O'Connell's 2003 study, conducted at the Cyberspace Research Unit at the University of Central Lancashire, provided the first systematic linguistic analysis of online grooming conversations and proposed a six-stage model based on chat room data.

1. Stage 1: Victim selection and access
   The offender identifies a target, often seeking signals of vulnerability : loneliness, family conflict, low self-esteem : and initiates contact through platforms where unsupervised young people congregate.
2. Stage 2: Friendship formation
   The offender presents as a peer or a trusted older friend, using mirroring language, shared interests, and consistent positive reinforcement to build rapport and emotional dependency.
3. Stage 3: Risk assessment
   The offender gathers information about the child's home environment, supervision levels, and who might notice the contact. Questions are asked indirectly: who else is home, when are parents back.
4. Stage 4: Exclusivity and trust building
   The relationship is framed as special, unique, and adult. Secrecy is introduced as a positive attribute of the relationship rather than a warning sign.
5. Stage 5: Desensitisation to sexual content
   Sexual topics are introduced gradually : jokes, then questions, then explicit content : with the offender testing the child's reaction and retreating if resistance is strong before trying again later or approaching from a different direction.
6. Stage 6: Control and maintenance
   Once abuse has occurred or is imminent, the offender uses guilt, complicity, threats, or affection to ensure silence and continued access. The language shifts to coercion and secrecy enforcement.

Not every case completes all stages, and some offenders move much faster than others. The model's value is analytic, not predictive: given a transcript, it allows an expert to locate where in the process a conversation sits, and to identify the specific linguistic moves associated with each transition.

Each stage of grooming has characteristic linguistic features that researchers have identified across multiple datasets. These are not perfect signatures : a single feature proves nothing : but their combination and sequence, read in context, provides the basis for expert analysis.

• Trust-building moves: extensive use of the child's name or chosen online handle, mirroring of the child's vocabulary and abbreviations, agreement tokens regardless of content, and expressions of being uniquely understanding ("I get you more than your parents do").
• Risk assessment questions: indirect questions about supervision embedded in apparently casual conversation: "are you by yourself?", "do your parents check your messages?", "who else is around right now?". The indirection makes them hard for a child to recognise as targeted.
• Normalisation language: claims that the sexual content or behaviour being introduced is normal and widely practiced, often with false peer-group references ("everyone does this at your age", "my other friends sent me pictures").
• Secrecy instructions: explicit requests not to tell, framed as protection ("I don't want to get in trouble because I care about you"), or as complicity ("this is just between us"). As the relationship escalates, the framing can shift to threats.
• Escalation testing: the offender introduces a slightly more explicit topic, then waits for the child's response. If the child shows discomfort or exits, the topic is dropped and returned to later; if the child responds neutrally, the next step is taken.

Defence cases in grooming prosecutions sometimes argue that the conversation was consensual fantasy role-play, that the defendant believed they were communicating with an adult who was using a minor persona, or that the content was fictional and not intended to lead anywhere. These arguments require the prosecution to show what distinguishes genuine grooming from the alternatives.

The key markers are concrete and logistical rather than abstract and narrative. Genuine grooming contains escalating attempts to establish real-world information: the child's physical address, school name, daily routine, or mobile number. It contains meeting proposals with specific times, dates, and locations. It contains coercive secrecy enforcement rather than playful fictional frame-setting. It responds to the child's actual vulnerabilities in ways that require knowing what those vulnerabilities are.

Research by Chiang and Grant (2017), drawing on the Perverted Justice corpus of chat logs from sting operations, found that certain lexical and discourse features reliably discriminate grooming conversations from other categories of adult-minor online interaction. Their work has been cited in proceedings in the United Kingdom and elsewhere as a framework for expert analysis.

A significant proportion of online grooming prosecutions in the United States, and a growing number in other jurisdictions, involve undercover law enforcement officers posing as minors. Perverted Justice, a US non-profit, conducted civilian sting operations providing transcripts and evidence to police, which fed into the Dateline NBC television segment To Catch a Predator. The quality and legal admissibility of the evidence from those operations was extensively litigated.

The entrapment argument is straightforward in principle: if the officer introduced and drove the sexual content and the suspect merely followed along, the officer created the offending behaviour rather than merely providing an opportunity for pre-existing offending intent. The defence requires showing predisposition, and language analysis of the conversation is the primary tool for showing who initiated, who escalated, and who resisted.

Jurisdictional variation matters here. US federal entrapment law focuses on government inducement and the defendant's predisposition. English law takes a different approach through the abuse of process doctrine. Australian jurisdictions vary by state. What counts as relevant language evidence differs depending on which legal framework applies, and an expert opinion that addresses predisposition directly may exceed the expert's proper remit in some systems.

The Perverted Justice dataset, consisting of thousands of published chat logs from sting operations, and the data held by the Child Exploitation and Online Protection Centre in the United Kingdom provided the first large-scale empirical basis for studying grooming language. Studies by Egan, Hoskinson, and Shewan (2011), O'Connell (2003), and later Chiang and Grant (2017) drew on these corpora to identify the discourse patterns described in this topic.

Three limitations of the corpus evidence deserve acknowledgment. First, the sting-operation data overrepresents cases where the suspect engaged willingly with an apparently available minor : cases where the offender broke off contact or showed ambivalence are underrepresented. Second, the corpora are heavily weighted toward text-based platform interactions from the early to mid 2000s; current grooming uses video calls, gaming platforms, and ephemeral messaging apps that are harder to archive. Third, the published studies describe patterns from English-language interactions in Western contexts; cross-cultural and multilingual grooming behaviour is much less well documented.