Empathy and it’s Role in Therapeutic Settings
“If you want others to be happy, practice compassion. If you want to be happy, practice compassion.”- The Dalai Lama.
“Why do I overthink everything?”
“How do I calm down during panic attacks?”
According to ChatGPT, these are two of the most frequent questions that it is asked relating to mental health. It should be noted that in relation to these questions, many people who ask them aren’t looking for a diagnosis, but they want to understand their experiences. More importantly, they want to know that someone else understands their experiences. In other words, people are seeking an empathic response from these language learning models.
Sadly, it is a mark of the times that more and more people are turning to artificial intelligence models to seek guidance relating to mental health issues. According to OpenAI, there were one million enquiries per week to ChatGPT about potential suicide planning or intent. In the UK, the NHS advises those who have these thoughts to call 999 or the Samaritans to seek immediate help. Calls to those services have continued to increase, but with limited resources, the average person in the UK with mental health issues is left waiting up to 18 months for help, according to Rethink Mental Illness.
Why are so many people turning to AI for help? Can AI replace therapists, and where does empathy fit in this equation? At My Family Psychologist, we wanted to find out more.
Types of Empathy
It is common to assume that empathy is being able to feel what someone else is feeling. However, research by Daniel Goleman and Paul Ekman breaks empathy down into three distinct groups and, based on the setting and the aim of the therapy session, a therapist may employ one or all three to help someone make sense of their experience or emotions.
Emotional
Emotional empathy is the skill and ability to share another person’s emotions, experiencing their feelings as if they belong to you. This is what most people think of when asked to define empathy, and it is heavily emphasised in counselling/therapeutic training.
Cognitive
Cognitive empathy is a bit different. Here, the emphasis is on being able to understand someone’s thoughts and feelings, as well as their perspective, without sharing their emotions. Cognitive empathy is the ability to understand another person’s thoughts, feelings, and perspective without necessarily sharing their emotions.
Behavioural
You can tell someone that you feel for them, but do you show it? Behavioural empathy focuses on how a person expresses empathy through behaviours, rather than focusing solely on how they feel internally.
There would be too many studies to list here, in a nutshell, but when it comes to therapy, counselling, or even being a support worker, empathy is a core predictor of positive outcomes for the person seeking help. Studies by Bourke, Barker and Fornells-Ambrojo (2021), as well as Tschuschke et al (2021), noted that empathy and responsiveness were not only a core building block of the therapeutic alliance but also significantly influenced outcomes.
For both studies, tools like the Working Alliance Inventory (WAI), the California Psychotherapy Alliance Scales (CALPAS) and the Helping Alliance Questionnaire (HAQ) were given to participants to explore how they felt about their therapy or intervention. These tools have statements such as “my therapist understands me” and “I trust my therapist”, with participants able to score how much these statements applied to them.
It was determined that the more participants felt supported and understood by their therapists, the better the outcome for psychological intervention.
Can AI Empathise?
On to the question of the hour: is it possible for an LLM to show genuine empathy to someone in distress?
To analyse this, you would need to break down what an LLM is. According to GeeksforGeeks, this is how they work-
“At their core, LLMs are neural networks designed to predict the next token in a sequence. They do not “understand” language like humans but learn statistical patterns, grammar, and context from enormous amounts of text, including books, websites, code, and conversations. By repeating this prediction billions of times during training, LLMs develop the ability to generate coherent, contextually relevant text.”
So, when a question about anxiety, depression, or even feeling overwhelmed is put into Open AI ChatGPT, its response is learned from it being able to access billions of data points in a millisecond, look at statistical patterns and grasp context. The truly concerning part, however, is that an LLM does not understand language as a human does, so how exactly can it empathise with someone in crisis?
There has been a wealth of research exploring whether AI can replace humans, specifically the empathy of humans, in therapeutic settings. One recent piece from Kumar et al (2025) argues that AI models may eventually become a useful tool for helping humans communicate more empathetically. However, it still cannot replace genuine human connection or emotional understanding. This paper found that over 200 collected conversations (where one person shared a problem and another responded), the AI model was almost as accurate as the experts at spotting empathic communication and was also more consistent than non-experts when evaluating empathy.
So, while AI cannot “feel” empathy for those who are asking it questions, it knows how to spot and respond with empathic words and phrases. This begs the question of how this may differ from a real therapist or counsellor.
Empathy and Outcomes in Therapy
It turns out that it differs greatly, more so from the perspective of the person seeking therapy.
Yes, AI models can empathise, but in a therapeutic setting, this equates to more than repeating empathic wording. Abargil and Tishby (2021) measured therapists’ empathic abilities through emotional recognition tasks and found that the higher a therapist’s empathic skillset, the more likely they were to have a stronger working alliance with their client, as well as help with emotional regulation and be more effective in reducing the client’s distress.
In a one-to-one setting, it also seems that body language and non-verbal cues play a large part in the process, which is why many people believe that trained therapists and counsellors can’t be replaced by AI. Scheidt et al (2021) and Salonen et al (2021) found that synchrony between counsellors/therapists and their clients related to head movement and body language (called mirroring) equated to better client-well-being ratings. Does this mean that the therapists felt what their clients were feeling? Perhaps not, but they were able to show all of the non-verbal cues that they did, which helped to build rapport faster and have better long-term outcomes for their clients mental health.
Attachment and Therapy
Then, there is evidence about attachment style and therapy preference. As many as 10-20% of people are estimated to have either an anxious or avoidant/resistant attachment style, according to The Private Therapy Clinic, which can greatly impact how they interact with and perceive others. This attachment style is usually caused by childhood issues with parental figures or caregivers.
The research here is still in the early stages but the results have been surprising. A study conducted by Wu, Liew and Dorahy in 2025 found that higher attachment anxiety predicted greater willingness to adopt AI counselling. Why? It ticks a lot of boxes for these types of attachment, in that it is immediately available, offers no fear of rejection, is responsive and can feel less invasive than meeting someone face-to-face.
People who have an anxious attachment disorder are more likely to have higher levels of social anxiety, according to Zhang 2025. As a result, they are deemed to be more vulnerable, which can impact their preference for using AI therapy. A more recent pre-release study (2026) from Bekes and Doorn highlighted that people who have insecure attachment, lower interpersonal trust and higher levels of psychological distress showed higher acceptance of AI therapists/digital interventions. In other words, people who may need higher levels of empathy are turning to outlets that offer faux empathy to seek help. The long-term impacts of this have not been established, but when looking at those who have anxious preoccupied attachment who can attend in-person therapy, or therapy via Zoom, there is minimal research looking at how impactful therapy can be.
So, could it be that the focus of AI-based research looks too much at anxious attachments and how those who have these gravitate towards AI, rather than exploring an intersection of how to get those who present with anxious attachments to attend in-person or digital therapy?
If you have been talking to ChatGPT and feel it’s not giving you what you need to feel better, then we can help. Contact My Family Psychologist for a confidential chat.
Whether you’re experiencing problems issues with relationships and are getting stuck in loops with LLM’s, or you just need someone to talk to, we’re here to help. Contact My Family Psychologist for a conversation with someone who can offer support, understanding and real empathy.
FAQs
What is Empathy?
Empathy is the ability to sense and understand other people’s emotions and feelings. However, research has shown that empathy can be broken down into three types: emotional, cognitive and behavioural, which all can (and should) be utilised in therapy by a trained professional.
Can AI Feel Empathy?
In a word, no. However, AI models can accurately detect empathy in conversations and replicate it, which can make it appear empathic.
Does Attachment Type Impact Therapy Choices?
Research by Wu, Liew and Dorahy in 2025 and Bekes and Doorn in 2026 found that people who displayed signs of anxious attachment were more likely to seek AI-based therapy. The reasons have been theorised to be linked to fear of rejection as well as interpersonal contact.
References
Békés, V., & Aafjes-van Doorn, K. (2026). The most vulnerable are prone to use AI therapists: The role of attachment, epistemic trust, and mental health symptoms in acceptance of digital mental health interventions. Journal of Psychoeducational Assessment. https://doi.org/10.1080/10503307.2026.2615388
Fraga Domínguez, S., Storey, J. E., & Glorney, E. (2026). Older victims of abuse: Understanding help-seeking in an understudied population. Victims & Offenders. https://doi.org/10.1080/15564886.20251337887
GeeksforGeeks. (2025). Large language model (LLM). https://www.geeksforgeeks.org/artificial-intelligence/large-language-model-llm/
Hoermann, S., McCabe, K. L., Milne, D. N., & Calvo, R. A. (2021). Application of synchronous text-based dialogue systems in mental health interventions: Systematic review. Journal of Medical Internet Research, 23(7), e26704. https://pubmed.ncbi.nlm.nih.gov/33569885/
Kumar, A., Poungpeth, N., Yang, D., Farrell, E., Lambert, B. L., & Groh, M. (2026). When large language models are reliable for judging empathic communication. Nature Machine Intelligence. https://doi.org/10.1038/s42256-025-01169-6
Li, X., Zhang, Y., & Chen, J. (2021). Article 696448. Frontiers in Psychology, 12. https://doi.org/10.3389/fpsyg.2021.696448
Marcus, G. (2025). The appeal of human empathy in an age of AI. Psychology Today. https://www.psychologytoday.com/us/blog/urban-survival/202507/the-appeal-of-human-empathy-in-an-age-of-ai
Nyman-Salonen, P., Kykyri, V.-L., Tschacher, W., Muotka, J., Tourunen, A., Penttonen, M., & Seikkula, J. (2021). Nonverbal synchrony in couple therapy linked to clients’ well-being and the therapeutic alliance. Frontiers in Psychology, 12, 718353. https://doi.org/10.3389/fpsyg.2021.718353
Rethink Mental Illness. (2025, February). New analysis of NHS data on mental health waiting times. https://www.rethink.org/news-and-stories/media-centre/2025/02/new-analysis-of-nhs-data-on-mental-health-waiting-times/
Roose, K. (2025, October 27). OpenAI says over a million people talk to ChatGPT about suicide weekly. TechCrunch. https://techcrunch.com/2025/10/27/openai-says-over-a-million-people-talk-to-chatgpt-about-suicide-weekly/
Rubin, M., et al. (2021). Article 34719825. Journal of Medical Internet Research. https://pubmed.ncbi.nlm.nih.gov/34719825/
Salonen, P., et al. (2021). Article 1050-3307 / 2615388. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2021.718353
Samaritans. (2025). Samaritans reports third consecutive year of rising calls from those facing financial hardship. https://www.samaritans.org/news/samaritans-reports-third-consecutive-year-of-rising-calls-from-those-facing-financial-hardship/
The Private Therapy Clinic. (n.d.). Attachment styles. https://theprivatetherapyclinic.co.uk/blog/attachment-styles/
Top of Form
Bottom of Form