Artificial Involvement? Why AI Does Not Replace Lived Experience in Psychosis Research.

We write as researchers who do lived experience work in psychosis. We are concerned about a recent article which proposes using large language models (LLMs) to provide Patient & Public Involvement (PPI) in areas such as grant proposals. The development appears designed to address longstanding concerns about difficulties in doing PPI work, such as a lack of time and capacity for researchers to do the work that successful PPI requires (Evans et al., 2025). While psychosis is not mentioned, PPI in psychosis faces a lot of the same issues, and this approach may appear tempting in the face of often challenging research infrastructure (Husain, 2025). 

Beyond just giving opinions and feedback on research, PPI involves people with personal experience of mental illness bringing their lived experience, identity, and reflections about encounters with systems where they may have had difficult and frightening experiences of being disempowered. This contextualised knowledge, borne out of sometimes painful personal experiences, is important for researchers to understand when they want to do research which is acceptable to people with mental health problems and respects their rights. An LLM cannot be a person – it will not be able to contribute to research in a human way, which is the basis for our scepticism. For further concerns on the critical adoption of LLMs within research, please see (Guest et al., 2025)

Furthermore, lived experience in mental health is a source of phenomenological insight and has drawn attention to topics overlooked in psychosis (Woods et al., 2015) and changing conduct in clinical research (Owrid et al., 2024).  These insights appearing in published research emerge through people with lived experience challenging established ways of doing things, not via an LLM doing prediction and pattern matching. 

PPI has already been accused of “elite capture” where patients who get involved in research are often selected by academics as being most likely to contribute to research in a way that does not challenge existing hierarchies (Okoroji et al., 2023). Since an LLM cannot generate anything new because it is trained on existing data, elite capture becomes even more likely because researchers will have even more power to choose what they want to listen to, and marginalised voices will be further excluded.

In summary, using LLMs to simulate lived experience risks reducing meaningful involvement to a technical process that reinforces existing hierarchies rather than challenging them. PPI in psychosis is not a problem to be automated, but a human relationship to be cultivated. 

References

Evans, J., Lewin, C. D. C., Fabian, H., Alam, M., Calder, P., Hill, R., Khayri, E., Markham, S., Sweeney, A., & Wykes, T. (2025). Facilitators of and barriers to patient and public involvement in mental health research within university settings: A systematic review and meta-synthesis. Psychological Medicine, 55, e297. 

Guest, O., Suarez, M., Müller, B., van Meerkerk, E., Beverborg, A. O. G., de Haan, R., Elizondo, A. R., Blokpoel, M., Scharfenberg, N., & Kleinherenbrink, A. (2025). Against the Uncritical Adoption of’AI’Technologies in Academia. 

Husain, M. (2025). If patients only knew how bureaucracy obstructs their involvement in research and clinical trials. In Brain (Vol. 148, Issue 10, pp. 3423–3424). Oxford University Press UK. 

Okoroji, C., Mackay, T., Robotham, D., Beckford, D., & Pinfold, V. (2023). Epistemic injustice and mental health research: A pragmatic approach to working with lived experience expertise. Frontiers in Psychiatry, 14, 1114725. 

Owrid, O., Richardson, L., Allan, S., Grant, A., Gogan, S., Hamilton, N., Yanga, F., Sirey, N., Syrett, S., & Garety, P. (2024). “T here’s no us vs. Them, it’s just us”: A creative approach to centring lived experience within the AVATAR2 trial. BMC Psychiatry, 24(1), 807. 

Woods, A., Jones, N., Alderson-Day, B., Callard, F., & Fernyhough, C. (2015). Experiences of hearing voices: Analysis of a novel phenomenological survey. The Lancet Psychiatry, 2(4), 323–331.