Using AI to Write a Financial Complaint Can Introduce AI Hallucinations

As generative AI becomes commonplace, many people facing a financial dispute turn first to ChatGPT or Claude. They ask it to produce a well-structured complaint in formal language. But a document that appears well researched can carry a serious risk: AI hallucinations.

The UK's Financial Ombudsman Service (FOS) recently described an AI-assisted financial complaint that cited 39 laws and previous decisions. Thirty-seven of those references were incorrect.

1. A complete-looking complaint may still be wrong

AI can be a useful writing assistant. It can quickly turn a disorganized account into a clear, formal letter.

Large language models, however, generate text by predicting likely words. They do not check every fact before answering. When a user asks for legal support, precedents, or regulatory rules, an AI system may return confident, neatly formatted references that do not exist or do not apply.

FOS chief operating officer Marc Harris said the service had seen invented laws, incorrect references to regulations, and fabricated past Ombudsman decisions. Some cited materials were real but misread or unrelated to the complaint.

2. What did the Financial Ombudsman Service find?

The Financial Times reported on the issue. In the UK, consumers who cannot resolve a dispute with a bank, insurer, or another financial business can take their complaint to the Financial Ombudsman Service. The FOS is an independent body that resolves disputes between consumers and financial firms.

As AI-assisted complaints have become more common, the FOS has received submissions that are longer and packed with more references, but not necessarily clearer. It also received a response from a professional claims representative that ran to more than 200 pages after the FOS issued a provisional decision of about six pages. The response contained numerous errors.

To get past the volume of text and identify the real complaint, FOS staff have made greater use of phone calls. They ask consumers what happened, where the disagreement lies, and what outcome they are seeking.

3. What does “37 out of 39 were wrong” mean?

That figure should not be treated as a reason to reject AI altogether. It came from one case and is not the average error rate across financial complaints.

In a separate small sample, the FOS estimated that up to one-third of consumer responses to initial assessments appeared to have been generated or heavily assisted by AI. These figures describe two different things:

• AI usage: how many complaint documents appear to have been written with AI assistance.
• AI hallucination rate: how much of the generated material contains incorrect facts or references.

The two should not be confused. Widespread AI use does not mean every document has a high error rate. But without human checking, incorrect material can enter a formal financial dispute process.

4. How do AI hallucinations enter financial complaints?

In financial complaints, hallucinations tend to appear in several forms:

• Nonexistent laws or cases: the title sounds official, but no such source can be found.
• Misread financial regulations: the AI confuses different regulators or applies a rule that does not cover the case.
• Invented past decisions: the complaint cites a similar outcome or compensation award that never occurred.
• Real but irrelevant material: the source exists, but it does not support the complaint.
• Unnecessary expansion: a straightforward billing or insurance dispute becomes dozens of pages long, hiding the facts that matter.

5. Why do financial complaints require extra care?

Financial disputes depend on precision. They involve real amounts of money, contract terms, dates, and responsibility. An AI hallucination can directly affect the way a complaint is understood:

• It can obscure the actual dispute: a simple point such as “I did not authorize this payment” may disappear under pages of invented references.
• It can delay the process: investigators must spend time checking laws and cases that are missing or wrongly cited.
• It can weaken the complaint: a document filled with incorrect references may be harder to understand and trust.

6. Check these five things before submitting an AI-assisted complaint

If you use AI to help draft a financial or legal complaint, check at least these five points before sending it:

1. Confirm that the source existsCheck every law, rule, and previous decision mentioned in the draft.
2. Check its relevanceMake sure the cited material actually supports your claim.
3. Verify the detailsReview every amount, date, and event in the timeline.
4. Remove unrelated materialDelete long passages that do not address the central dispute.
5. Return to the original documentsUse contracts, statements, and conversation records as the source of truth, not the AI's summary.

7. Where can AI FactScan help?

If you use ChatGPT, Claude, Gemini, or another supported platform to organize a complaint, AI FactScan can help you review the citations included in the AI answer:

• Find URLs, DOI references, arXiv IDs, PubMed IDs, and other citations in AI answers.
• Check whether citation links are valid and show source-quality grades from A to F, along with risk signals.
• Flag invalid, low-rated, or potentially risky sources so you can open the original material first.

AI FactScan helps you check whether cited sources exist, review their source-quality grades and risk signals, and decide which original materials to verify before relying on an AI answer.