UK Financial Ombudsman Turns to Phone Calls as AI-Written Complaints Fill Up With Hallucinations

Britain’s Financial Ombudsman Service is picking up the phone more often, and generative AI is the reason. The ombudsman, which settles disputes between consumers and banks, insurers, lenders and other financial firms, says a growing share of complaints now arrive written or heavily shaped by AI tools, and some contain invented laws, misquoted regulations and past decisions that never existed. Chief ombudsman James Dipple-Johnstone told the Financial Times this week that in the most striking case he had seen, 37 of 39 references in a consumer’s submission were incorrect. His answer is direct contact: “Our initial response has been to roll out more telephone contact with customers to make sure that we get to the heart of the dispute.”

The story is about complaints handling, but it also says a lot about how organizations should deal with AI-written correspondence of every kind, including privacy rights requests.

What the ombudsman has found

The ombudsman has been tracking the trend for months. In its February 2026 response to the Mills Review, it said early small-sample analysis suggested AI likely contributed to up to a third (35%) of responses to its initial assessments. Its later guidance for businesses describes the pattern in more detail.

• The benefits are real. Used well, AI helps people organize their thoughts, write more clearly and get past language or literacy barriers. The ombudsman notes it can make a big difference for vulnerable consumers and people with neurodevelopmental conditions.
• The problems are specific. Excessive or uncritical use produces needlessly long submissions, incoherent narratives, and hallucinations, which the ombudsman describes as fabricated laws, misquoted regulations or invented past decisions.
• Positions harden. Heavy AI use in some case types has led parties to take more entrenched positions, pushing more cases to final ombudsman decisions and generating mistaken complaints about the service itself.
• Volume is growing. Trade press reports that claims management companies and legal representatives are using AI to produce bulk submissions, some running to more than 200 pages.
• Firms see it too. The ombudsman says financial firms are also receiving long AI-written complaints with hallucinations, which slows their responses. It calls this an emerging risk rather than a major operational problem so far.

The ombudsman has issued enhanced guidance to its case handlers, appointed AI leads across the organization, strengthened its AI governance, and published suggestions for consumers who use AI to complain. It has asked the Financial Conduct Authority for consistent guidance to firms and consumers.

Why privacy teams should pay attention

Privacy teams are seeing the same thing in data subject requests. Consumers and their representatives use AI to draft access, erasure and objection requests that cite the wrong articles, the wrong law or deadlines that do not exist. Some requests are generated in bulk. The temptation is to treat a confused request as invalid. That is a mistake under UK and EU data protection law, which do not require a request to cite the right law or use particular words.

The UK has also tightened the rules on complaints. Under the Data (Use and Access) Act 2025, organizations must have a process for handling data protection complaints, as our article on the DUAA complaints requirements now in force explains. An AI-written complaint still counts.

Five rules for handling AI-written requests and complaints

1. Find the real request. Read past the citations to what the person actually wants: a copy of their data, a correction, a refund, an apology. The ombudsman’s phone calls are a way of doing exactly that.
2. Do not reject a request because its legal references are wrong. A request for access is valid if it is clear the person wants their data. The time limit usually starts when you receive it, so ask any clarifying question quickly.
3. Do not repeat the errors. If a complaint cites a fake regulation, do not quote it back as if it were real. Answer on the actual law and say so politely.
4. Keep humans in the loop when you use AI in reply. Many firms now use AI to summarize long complaints. Summaries can drop the key point or add their own errors. Have a person check summaries against the original before a decision is made, especially for decisions with financial or legal effects.
5. Record what you did. Note how the request was interpreted, any clarification asked for, and the answer. If the case reaches the ombudsman or the Information Commissioner’s Office, that record shows you dealt with the substance.

The data protection side of a phone call

Moving to phone contact raises its own privacy questions. Calls may be recorded and transcribed, sometimes by AI. People should be told if calls are recorded, why, and how long recordings are kept. Notes taken during a call become personal data that the complainant can request. If a firm uses an AI transcription service, the vendor is usually a processor and needs a contract that limits its use of the recordings.

Financial services firms face overlapping rules on these points, from FCA conduct rules to data protection law. Our article on how the ICO and FCA rewrote direct marketing rules for financial services shows how the two regulators coordinate, and our piece on the U.S. Treasury’s AI guidance for the financial sector covers the U.S. side. Organizations looking for a sector framework can also consider the role of ICO codes of conduct.

Where Captain Compliance fits

Captain Compliance’s data subject rights tools give consumers a structured form for privacy requests and give teams a workflow to receive, verify and track each request against its deadline. A structured intake form helps people say what they want in plain terms, even if they also send a long AI-written letter. A closed ticket is a record that a request was handled. It is not proof that data was found, corrected or deleted unless the underlying systems were checked.

Frequently asked questions

How many complaints to the Financial Ombudsman are AI-written?

In a small sample reviewed earlier in 2026, the ombudsman estimated that AI contributed to up to a third (35%) of responses to its initial assessments. It has not published a figure for all complaints.

Can a firm reject a complaint because it was written with AI?

No. The ombudsman recognizes that AI can help consumers. The issue is accuracy and length, not the tool. Firms should deal with the substance.

Is a data subject access request valid if it cites the wrong law?

Generally yes. A request does not need to cite the correct law or use set wording. If it is unclear, ask the person to clarify promptly.

Should we use AI to summarize long complaints?

It can help, but a person should check the summary against the original, and the use of AI should be covered by your privacy notice and vendor contracts.

What is the ombudsman asking regulators to do?

It has asked the FCA for consistent guidance to firms and consumers on using AI in the complaints process, and has raised the issue with other ombudsman services. A U.K. court ruling on data security, the DSG Retail appeal, is a reminder that British regulators also expect firms to keep the personal data in complaint files secure.

The fix was a conversation

The most advanced response a public body has found to AI-written complaints is a phone call. That is not a step backward. When a document of 200 pages buries the one sentence that matters, the fastest way to find it is to ask the person who sent it. Firms that let their own AI tools answer AI-written complaints, with no one checking either side, may find that the dispute ends up at the ombudsman anyway, where someone will pick up the phone.