The challenges

AI-generated complaints tend to be longer than complaints written by humans, so take more time to read and respond to.
Rather than provide clarity, the crux of the complaint can be lost in verbose, repetitive letters.
They often contain references to incorrect, non-existent or irrelevant legislation, making it hard for healthcare staff to respond to.
Some AI models overstate facts or use overly dramatic language, which can exaggerate the impact of relatively minor issues and can make it hard for the reader to understand what the complainant’s priorities are when resolving the issues.
It can leave the complainant with unrealistic expectations about the outcome of the complaint process. AI-generated complaints that sound plausible, eloquent and legally robust may lead the complainant to conclude prematurely that their stance is irrefutable.
It offers vexatious complainants a means of generating multiple complaints about the same issue, which can overload the healthcare provider before they’ve had a chance to investigate the first complaint.
Generating these letters requires little time or effort, so it facilitates ‘boomerang’ complaints: sending follow-up complaints about the issues already addressed but wording them slightly differently to make them look like new complaints.

When complainants use generative AI, it can create an unsustainable workload and stress for staff tasked with analysing and responding to these complaints. It also risks the diversion of resources away from the clinical care of other patients.

The ramifications of AI complaints are well recognised in many sectors, not just healthcare. The Parliamentary and Health Service Ombudsman (PHSO) recognises 'there has been a significant increase in the use of generative AI to produce complaint submissions. These can increase volume, raise expectations of rapid resolution, and increase complexity if the content is unclear or includes inaccuracies.' 

To this end, they have suggested how AI prompts may be phrased to optimise the output.

Please see the footnote below for comments from the Scottish Public Services Ombudsman (SPSO) and the Public Services Ombudsman Wales (PSOW) on the use of AI in complaints¹.

Practical tips

Most organisations want to take positive steps in the face of negative feedback by effectively resolving concerns and implementing any lessons learnt. However, achieving this is much harder where the use of AI results in unfocussed complaints that prioritise sounding legalistic over identifying the key issues of concern.

The strategies below are ordered according to where on the spectrum these complaints fall, from a single AI-generated complaint from a complainant who has used it to help them write, through to a complainant who is engaging in ongoing unacceptable behaviour with the assistance of AI.

Our intention is not to give you additional work, but rather to support you in processing lengthy correspondence and to help ensure your first response is suitably comprehensive, so that you can avoid future circular communications about the same issues.

Initial steps

As a starting point, you cannot decline to answer a complaint simply because it is written by AI. Regardless of how the complaint was generated, you are still obliged to answer it if the subject matter falls within the NHS complaint regulations for your jurisdiction.
It is not appropriate to ask the complainant to resend a shorter version, although there may be situations where this might be necessary if the letter is undecipherable. MDU members can approach us for advice on this.
When you acknowledge the complaint, you don’t necessarily need to mention that it appears to be AI-generated. You can highlight that the letter is long and the key issues are unclear, so the response may take longer than usual. You can add that if the reason for this was due to it being AI-generated, the complainant may wish to redraft the complaint and they may find PHSO guidance useful in making it more succinct.
It is important that every complaint is acknowledged, but if you receive several in short succession, you can reference all of them in one acknowledgment letter.
As many AI complaints include repetition and extraneous detail, it can help to go through and highlight what the themes are. You can then cross out points that are duplicated. If you’ve received more than one complaint, you only need to respond to each issue once, so you can also cross out points in subsequent letters that are repeated from the first letter.
Writing a comprehensive first response will ultimately save you time. As long as it fully addresses each subject in the complaint, you’re not required to respond to them again even if they’re phrased differently in subsequent correspondence. Instead, you can direct the complainant to where it has already been addressed in your previous response, and confirm you will not respond to that particular point again. You can find our complaint response guidance here.
lnclude a list at the start of the response of what you understand the themes to be. Here you can mention that, due to the length/nature of the complaint, you hope you have not misunderstood or missed their key concerns. If multiple correspondence has been sent prior to your first response, you can state the dates of each complaint and what you understood to have been added to the concerns at each point. This will not only guide what concerns you need to respond to in your first response, it will also allow you to spot whether the same issue is being revisited in subsequent correspondence.
Use headings to organise your response. You can also explain at the outset that you have used headings to identify what your understanding of their complaint is. This can be helpful to pre-empt any suggestions that you have ignored aspects of the complaint.

For subsequent correspondence

If subsequent correspondence raises reasonable follow-up points, then it is appropriate to respond to these.
Offer a resolution meeting early if you feel it may be a more productive way of resolving the complaint than sending further written responses. You can find our guide on these meetings here. You will still need to send the complainant a written summary afterwards, but a conversation may allow the complaint to be resolved much earlier.
Compare the themes in the new correspondence with your previous list. Are you satisfied you have fully investigated them and responded clearly? If so, as advised above, it is perfectly reasonable to state that particular point has already been addressed, and you will not respond further to it.
It may help to direct the complainant to an advocacy service to support them in complaining effectively, which may avoid further AI-based correspondence.

For ongoing, repetitive complaints despite an appropriate investigation and response

We recognise that not all vexatious complaints involve the use of AI and that not all AI complaints are vexatious, but we are seeing a rise in unreasonable behaviour when making complaints combined with the use of AI, which is a potent mix.

The bar to deem a complaint as vexatious is relatively high and it is important that organisations can justify a decision to not reply to a complainant on the basis their complaints are vexatious.

Deciding what constitutes a vexatious complaint can be challenging. While we use the term vexatious, this is shorthand for complaint behaviour that is unacceptable. The PHSO guide on managing challenging situations in complaint handling defines unreasonable or unacceptable behaviour as 'any behaviour by a person that raises substantial health, safety, resource or fairness issues for the people involved – either because of the type of behaviour or how often it happens.'

Deeming a complainant’s behaviour as unacceptable will require the organisation to consider the context and history of the complaint. While the initial complaint might be problematic, the practice still needs to answer it, as it cannot be deemed a vexatious complaint until a pattern of follow-up behaviour has confirmed the underlying motive for the complaints.

In cases where there is repetitive correspondence, the following strategies can help manage the situation.

Send a de-escalation letter to the complainant to flag:
- aspects of their behaviour that are problematic and why
- what you would like them to change going forward
- the outcome if they don’t. For example, restricting their communication or stopping considering their complaint. You can find a template letter in the PHSO’s complaint guide. You can also suggest a communication agreement, as detailed on page 13 of the guide.
Appoint a single route of correspondence and a named individual who will be reading their correspondence. This allows you to process any future correspondence more efficiently. It also prevents new complaints coming via email, telephone and post from getting lost or read by different members of staff who may not be privy to the context.
Make a timeline of what has been sent and when with numbered letters (ie, letter of complaint 1, letter of response 1, etc). You may also wish to include a list of what the points of complaints were each time. This allows you to quickly check what has been received and sent. You can also highlight in any final letter that over a certain period the complainant sent X number of letters and received Y number of responses.
Send a final letter of response which contains:
- a summary of the correspondence sent and received to illustrate the workload and effort made to resolve the issue
- an explanation that attempts at local resolution have been exhausted and you will not be corresponding further on the matter
- the impact this has had on resources which has diverted time from patient care. This may allow the complainant to develop insight into the ramifications of their behaviour
- reiteration of their right to escalate the complaint to the PHSO
- a statement that further correspondence will be read and filed but not responded to unless it contains new points of complaint.
Consider sending a warning letter regarding their behaviour. It would be appropriate to clarify that these steps are not intended to prevent them raising new complaints, but rather to stop the same topics being revisited and in a way that is unacceptable.
Consider advising the complainant that their email address will be blocked with reassurance they can still contact the practice in the usual ways to arrange care. Before any action is taken on this step, it would be appropriate to:
- discuss this as a practice
- establish that it won’t be detrimental to the patient’s ability to access care
- liaise with your ICB and LMC
- check the practicalities of doing this within the system you use.

Many practices feel uncomfortable about these last steps, and we would strongly advise that MDU members involve us early in these challenging cases for objective advice.

It may be reassuring to note that, while not related to AI complaints specifically, both the GMC and the PHSO have policies on vexatious complaints and respectful behaviour respectively. These do not provide guidance for healthcare professionals about complainants, but relate to the conduct of individuals using these organisations’ services. They contain many principles that are broadly applicable for practices dealing with problematic complainants. In addition, healthcare providers may find it reassuring when organisations that oversee their work recognise the need to place boundaries in such cases.

Footnote

The Scottish Public Services Ombudsman (SPSO) and the Public Services Ombudsman Wales (PSOW) provide advice for complainants on the possible inaccuracies AI can introduce to their complaints and the importance of checking the output for accuracy prior to submitting it. The SPSO also has advice on supporting respectful engagement, which is intended to 'help organisations and individual staff approach the most common problems that can occur when managing complaints.'

Managing AI-generated complaints

More on complaints

Related articles

Membership in your pocket