NHS trusts across England are encountering a new administrative problem: formal patient complaints drafted by artificial intelligence that are lengthy, factually unreliable, and increasingly difficult to process efficiently.
Healthcare administrators report that submissions which once ran to one or two pages are now regularly reaching dozens of pages. The complaints frequently cite statutes that do not exist, reference clinical encounters with fabricated details, and in some cases name members of staff who have no record of involvement in a patient's care. Legal teams within trusts are spending considerable time verifying claims that turn out to have no basis in law or fact.
The pattern is linked to the growing use of large language models by patients drafting formal grievances. These tools lower the barrier for people who struggle to articulate complaints in writing, particularly those with literacy difficulties or for whom English is not a first language. That is a genuine benefit. The problem is that the same tools generate text with a tendency to invent. They cite plausible-sounding legal precedents that do not exist, repeat arguments in circular structures that inflate word counts, and produce a professional appearance that masks substantive inaccuracy.
For staff responsible for patient relations, the practical consequence is a bottleneck. A complaint that once required an hour to read and assess may now demand a full working day of verification. Administrators must first establish which elements of a submission correspond to real events before they can address the clinical concerns at the centre of it. That process is not straightforward when a twenty-page document intersperses genuine grievances with fabricated detail and invented statutory obligations.
There is a secondary concern that is harder to quantify but no less significant. Handwritten complaints, submitted by patients without access to or familiarity with AI tools, may face longer waits for response simply because the volume of high-word-count AI submissions occupies more administrative time. The risk is that the patients least likely to use these tools are also the patients most likely to experience delays as a result of others using them.
NHS trusts have existing obligations under the Parliamentary and Health Service Ombudsman framework to respond to formal complaints within defined timeframes. Whether AI-generated submissions affect compliance with those timelines has not yet been assessed systematically, but the operational pressure described by administrators suggests the question will need to be examined.
The challenge presents those responsible for healthcare administration with a difficult position. Restricting the format of complaints would risk excluding patients who rely on these tools to communicate effectively. Accepting submissions without modification leaves staff to manage the verification burden alone.
Some administrators have begun discussing the use of AI screening tools to summarise incoming letters before they reach human review. The logic is functional, if uncomfortable: if patients are using machines to write complaints, trusts may be forced to use machines to read them. The result would be a process in which artificial intelligence sits at both ends of a formal grievance procedure, with clinical staff and administrators operating in the middle without full visibility of what has been filtered or condensed.
That outcome raises questions that go beyond administrative efficiency. Formal complaints serve a regulatory function within the NHS. They are one of the primary mechanisms by which patient safety concerns reach trust boards and, where necessary, external regulators. A system in which those complaints are summarised by AI before human review introduces a layer of interpretation that could affect which concerns are escalated and which are not.
The current situation is in part a product of speed. AI tools became widely accessible faster than institutions developed policies to manage their use in administrative contexts. Healthcare providers are now adapting to a problem they did not anticipate. Whether they can do so without disadvantaging the patients these complaints are meant to protect remains an open question.
