GPs Ask Patients to Write Complaints Without ChatGPT

Hospitals and GP practices are asking patients to write complaints in their own words after ChatGPT letters began citing laws that do not apply. A template from Gloucestershire’s local medical committee, now circulating among practices, tells patients they do not need legal words or NHS rules for a complaint to be taken seriously.

The same line is on hospital sites and on the Parliamentary and Health Service Ombudsman’s form, in a year when more than a quarter of a million written complaints already reached the NHS in England.

Gloucestershire GPs Ask for Facts and Outcomes

Nic Wright, chief executive of Gloucestershire Local Medical Committee, said the committee wrote the statement after GPs saw a rise in complaint and feedback letters generated or heavily assisted by AI tools. She said the worry is not that patients complain, and not that they use AI to help them say what they mean.

Some AI generated letters were often lengthy and focused heavily on alleged breaches of legislation, regulations, contracts or professional standards, sometimes citing numerous legal or policy references.

Nic Wright, chief executive, Gloucestershire Local Medical Committee

Wright said that mix makes it harder for a practice to see what actually happened and to run a meaningful investigation. What they need, she said, is a clear account of the patient’s experience: what happened, why they were unhappy, how it affected them, and what outcome they want.

The template’s blunt line is the one now being copied: you do not need to use legal words or quote laws or NHS rules for us to take your complaint seriously. Lincolnshire LMC’s adapted AI complaints note, posted on 16 September 2026, says it took that work from Gloucestershire with permission. Five days later the same wording was being described as national practice advice.

What ChatGPT Puts in a Complaint Letter

The Gloucestershire note lists five ways a chatbot can wreck a complaint that started as a real event. It can give incorrect information, add details that are not true, include laws, rules or rights that do not apply, use language that is too formal or hard to follow, or make the letter longer than it needs to be.

WHAT THE BOT ADDS

• Wrong facts: AI tools can insert incorrect information about dates, staff, or what a clinic is required to do.
• Invented detail: Letters sometimes add scenes or claims the patient never made.
• Fake rights: ChatGPT will cite laws, rules or NHS rights that do not apply to the case.
• Legal tone: Formal wording makes a simple delay read like a solicitor’s letter.
• Extra length: A short account is padded until the point is hard to find.

Lee Bennett, strategic complaints lead at NHS England, who trains practices on the process, has called the same pattern a challenge across the NHS and other government bodies. He said some patients now use AI both to write letters and to reply to the practice, and that the results can be overly complicated, confusing and, in some cases, inaccurate.

The Ombudsman does not tell people to abandon the tools. It published five suggested prompts for drafting a complaint that tell the bot to keep the patient’s voice, stick to supplied facts and refuse to invent new ones. Prompt five is the clearest: rewrite the text as a clear, polite complaint, use only the information provided, do not add or invent new facts, and stay concise.

That is the official theory. Type “complain about a GP being dismissive” into ChatGPT and it will still hand you a full page in seconds, which is how a three-line slight becomes a dossier.

Written Complaints Hit 256,777 Last Year

The letters are landing on a system that was already taking more written complaints before the bots became common. NHS England’s official count for 1 April 2024 to 31 March 2025 is the latest full year on the books.

NHS WRITTEN COMPLAINTS, ENGLAND

Those 256,777 written complaints in 2024-25 compare with 207,407 in 2014-15. In hospital and community services, communications was the largest single subject, at 17.6% of new complaints. A quarter of those hospital and community complaints were fully upheld (25.0%). In primary care the fully upheld share was 30.3%, and the largest subject was clinical treatment, 15.0% in general practice and 28.9% in dental.

A bot that pads a communications grievance with extra law does its damage in the category that already dominates hospital files. The count is not an AI census. NHS Digital does not record which letters came from ChatGPT. It only shows the pile the new letters are joining.

Most People With Poor Care Never Complain

The “own words” campaign treats the problem as too much writing. The older problem is too little. Healthwatch England, looking at the 2024-25 figures, said the official total is only the people who got as far as a letter.

WHO STILL NEVER COMPLAINS

• Poor care in a year: 24% of people Healthwatch polled said they had experienced poor NHS care in the past year.
• Took no action: 56% of those who had poor care did nothing about it.
• Formal complaint: 9% of those who had poor care made a formal complaint.
• Why they stay quiet: Rebecca Curtayne, external affairs manager at Healthwatch England, said the most common reason was a lack of confidence that the NHS would take meaningful action to stop the same thing happening to someone else.

Professor Victoria Tzortziou Brown, president of the Royal College of General Practitioners, has said technology, including AI, may make it easier for patients to raise concerns and improve access for some people. She also warned it could increase the number of complaints made in the heat of the moment or without full context. Both can be true in the same waiting room: a patient who could not face a blank page now can, and a patient who is still angry at 11pm can hit send before anyone has slept on it.

Independent advocates already exist for people who cannot write a letter. The Ombudsman tells people who do not want to use AI that they can call 0345 015 4033, ask a friend, family member or advocate, or request the form in another format. “Own words” is easy advice if you already have words. It is a wall if you do not.

Practice Managers Now Read for Hallucinations

Dr Sarah Townley, deputy medical adviser at Medical Protection, told a GP conference in Birmingham in June 2026 that the bots have moved past ordinary complaints. People are bringing claims in their own name, without solicitors. The letters can be 25 pages long, she said, quoting legal arguments that maybe do not even exist because they were generated by AI.

Even weak claims, she said, create phenomenal workload, because staff have to sift far more text. Length no longer tracks severity. A simple matter can arrive at five or six pages. Practices told the same project they were seeing AI complaint letters of seven pages and longer, produced in seconds. Townley said most practices would much prefer to hear the patient in their own language, because that makes it easier to tell what the complaint is about and what the patient actually wants.

That preference is already turning into a filter. In GP discussion of the same letters, some doctors said they stopped reading after the second page. Some reception teams treat an AI-looking letter as junk on sight, on the logic that a real grievance would have been written by a person. Once style becomes a test of seriousness, a fluent bot-written account of a true delay can be filed with the fake-law pile. The patients who needed the tool most, including people whose first language is not English, then pay twice: they needed help to write, and the help makes the letter look untrustworthy.

In January 2026 the Independent Sector Complaints Adjudication Service described a private-care case in which a patient used AI, then would not confirm the “heads of complaint” (the list of issues to be investigated) after the letters filled up with legal arguments that were not part of the original concern. The process stalled on the extra law, not on the care.

Trusts Copied the Ombudsman’s AI Prompts

The hospital notices now going up are, in large part, the Ombudsman’s winter advice with a trust logo. The Parliamentary and Health Service Ombudsman told people as early as 24 November 2025 that AI can help, if they keep it simple, stick to the facts and use their own words. It repeated that on 9 February 2026.

Thinking of using an AI tool to help you submit your complaint to us? Make sure:

– you keep it simple
– you stick to the facts
– it’s in your own words.

For more advice on using AI visit: https://t.co/3vkquTlrkB pic.twitter.com/KLNRmm1Zaq

— Public Service Ombudsman (@PublicServOmb) February 9, 2026

The live form still says the office wants to hear the complaint in your own words. It names ChatGPT. It warns that AI text can make a complaint too complicated or add legal references that are not relevant, which may slow the review or lead staff to ask for a resubmit. It also warns that tool-makers may use prompts to train their systems and may process the text in other countries, so patients should keep NHS numbers and other identifiers out of the chat and add them only on the form.

HOW THE OWN-WORDS ADVICE SPREAD

1. 24 November 2025: The Ombudsman tells complainants AI can help if they keep it simple, stick to the facts and use their own words.
2. January 2026: ISCAS publishes a patient checklist and a case study of AI letters that derailed a stage-two investigation with extra legal argument.
3. 9 February 2026: The Ombudsman repeats the three rules on its public feed and points to the form.
4. 11 June 2026: Medical Protection describes 25-page AI legal letters, and BMA GP chair Dr Katie Bramall floats word-limited web forms.
5. 16 September 2026: Lincolnshire LMC posts the Gloucestershire-based practice note.
6. 21 September 2026: Hospital trusts including Stockport and Mid Cheshire are cited as giving patients the same own-words warning.

Stockport NHS Foundation Trust told patients that while there can be benefits to using AI, it encourages complainants to describe their experience in their own words wherever possible. It also said it will not reject a complaint simply because it is complex or because AI has been used, and asked people to keep the letter focused on the issues that matter most. Mid Cheshire’s page on AI complaints tells patients not to ask AI to add legal arguments, use formal legal language or reference laws to try to make a complaint sound more official, and to check that any AI text matches their own experience and the facts. It says the wording is taken from the Ombudsman.

A Word Cap Will Not Catch Fake Rights

Dr Katie Bramall, chair of the BMA’s GP Committee, has asked why practices cannot put a complaints form on the website with a limited number of words, “getting around a lot of the AI word soup and salad.” A cap would stop the 25-page claim letter. It would also stop the patient who needed two pages because English is a second language, or because the story runs across three clinics and a missed referral. A short letter can still cite a right that does not exist. The hallucination is not a word-count problem.

Bramall also argued for a national, central way of handling complaints as part of the whole passage of care, pointing to patients who cannot find a secondary-care appointment and cannot get through after reception closes. That is a different failure from ChatGPT’s tone. If the bot is filling a vacuum, the vacuum is still there when the letter is cut to 400 words.

Wright’s list remains the practical test. Practices need what happened, why the patient was dissatisfied, how it affected them, and what they want done. The Ombudsman’s prompts try to drag a chatbot back to that list. The Gloucestershire template tries to talk patients out of dressing it in law. Neither will work if the person reading the envelope has already decided that a polished page is not a real complaint.