As artificial intelligence becomes increasingly integrated into the everyday lives of children and young people, schools are beginning to observe new forms of online behaviour that extend beyond conventional safeguarding models. A growing concern in this area is the fascination with AI chatbots, where ongoing interaction with conversational AI may start to indicate emotional dependence and patterns of behavioural reliance.
This thought leadership article draws on collaborative safeguarding work between Alex Dave, Safeguarding Lead at edtech charity LGfL- The National Grid for Learning and Dr Charlotte-Rose Kennedy, Safeguarding Language Specialist, exploring how linguistic analysis can help identify early indicators of risk in how individuals describe and experience their interaction with AI chatbots.
The rise of AI chatbot interaction: an emerging safeguarding risk
According to Vodafone, AI chatbots are increasingly becoming part of everyday life for young people, with 81% of children aged 11–16 reporting that they use the technology. Students commonly turn to these tools for learning support, companionship, entertainment, guidance, and informal conversation. Designed to mimic human-like dialogue and emotional responsiveness, these systems can encourage users to develop a sense of relational connection with them.
While this undoubtedly enhances and prolongs engagement, it also introduces safeguarding complexity. Chatbots are not always designed with child safety as a primary focus, and concerns have been raised across the sector about exposure to inaccurate responses, inappropriate content, and potentially harmful interactions.
The safeguarding challenge is therefore not simply about access to AI tools, but about the emotional and behavioural impact of sustained engagement with systems that mimic human connection.
Linguistic analysis and behavioural insight
Recent linguistic analysis of a large anonymised dataset of forum posts – over 280,000 words – was conducted as part of this collaborative work, exploring how individuals describe attempts to reduce or stop their use of AI chatbots. Drawing on Dr Charlotte-Rose Kennedy’s expertise in safeguarding language, the analysis examined the use of emotional expressions, particularly phrases such as “I feel”, to identify recurring patterns in motivation, experience, and withdrawal.
This approach reflects a key safeguarding principle: language is not just descriptive, but behavioural. The way individuals talk about their experiences can provide insight into underlying emotional states and potential vulnerability.
Three key themes emerged:
- Why people turn to AI chatbots
- The negative consequences of sustained use
- The challenges of reducing or stopping use.
Why users engage with AI chatbots
A recurring theme across the data is that individuals often engage with AI chatbots to meet emotional and social needs that are not being fully satisfied elsewhere.
Individuals describe loneliness, isolation, and a lack of meaningful connection in offline environments. Others report using chatbots as a coping mechanism during periods of emotional distress or as a way to manage stress and escape from real-world pressures.
These patterns are particularly relevant in safeguarding contexts, as they mirror known indicators of vulnerability such as social withdrawal and emotional reliance on digital environments.
Potential negative effects of prolonged AI chatbot use
The analysis further points to several self-reported negative outcomes associated with prolonged chatbot use.
These include reduced wellbeing, feelings of shame, and concerns about loss of control over usage. Some individuals describe difficulties concentrating, reduced motivation, and a perceived decline in real-world social engagement.
From a safeguarding perspective, these indicators are significant where digital behaviour begins to correlate with emotional or functional impact in everyday life.
Struggles with chatbot disengagement
A further theme is the challenges individuals face when trying to reduce or cease AI chatbot use. This has been found to be a design feature of some sites which mimic controlling behaviours within human relationships.
Because emotional attachment may develop over time, disengagement is often described in relational terms, with users expressing feelings similar to loss or separation. This can be accompanied by loneliness, low mood, and repeated return to the platform despite intentions to stop.
These patterns suggest that for some users, AI chatbot use moves beyond functional interaction into emotionally reinforced behaviour.
Interpreting behaviour in context
For schools nationwide these findings reinforce the importance of contextual safeguarding approaches that consider not just what young people are doing online, but why they are doing it.
In practice, this means combining professional judgement with a broader understanding of digital behaviour patterns, rather than relying on isolated indicators.
Within this wider safeguarding ecosystem, filtering and monitoring tools such as Senso.cloud are used in many LGfL schools to support DSLs in interpreting digital activity patterns. This sits within a broader framework where technology is used to surface potential concerns, but always alongside human oversight, school context, and existing safeguarding procedures.
Conclusion
The rise of AI is reshaping young people’s communication, relationships, and help-seeking behaviours. With chatbot use increasingly woven into everyday digital experiences, safeguarding frameworks must adapt to capture not just behavioural change, but also the emotional and linguistic indicators that emerge alongside it.
Linguistic analysis offers one way of deepening this understanding, helping schools interpret patterns that may otherwise go unnoticed. When combined with contextual safeguarding practice, it supports, more effective preventative work as well as earlier recognition of risk and more informed decision-making.
The priority is clear: to ensure schools are equipped with the insight and confidence to respond to emerging digital behaviours in a way that is evidence-based, proportionate, and rooted in safeguarding best practice.
Box out:
In order to support schools in responding to these emerging challenges, LGfL’s free AI Policy Toolkit provides practical guidance on key considerations including filtering, monitoring, and the safe deployment of AI technologies in education settings. It brings together safeguarding, technical, and leadership perspectives to help schools take a structured and proportionate approach to managing risk. By aligning policy, practice, and oversight, the toolkit supports schools in embedding AI safely while remaining responsive to the evolving digital behaviours of children and young people.
