More and more clients are using tools like ChatGPT and Gemini to summarise their grievances and manipulate agents' responses, resulting in back-and-forths that often never reach a resolution.
Abta confirms it has been made aware of the trend, telling TTG it has “heard from members they are seeing more complaints that appear to have been drafted by AI tools".
The subject of these complaints can vary widely. A client might be frustrated about a long-haul flight delay or disappointed by the quality of their hotel room’s bedsheets. They may feel underwhelmed by the check-in experience, or annoyed at construction noise outside of their resort.
But what doesn’t vary is the objective of these complaints: compensation. “We are living in a very litigious world,” says Claire Smith, Head of Legal at Not Just Travel. “Clients think – 'let’s use AI to get a full refund’.”
'Gaslit by a robot'
Marie Rowe is one of 12 agents who told TTG they have been impacted by this recent development. The Travel Counsellor says she was left “exhausted” after a client, who had been cordial during the enquiry stage, turned to ChatGPT to dispute a third-party invoice for damages.
The client had been charged for a chipped windscreen on their hire car and, despite having declined additional insurance at the rental desk, were now insisting they weren’t liable to pay.
They proceeded to send a series of long, convoluted messages to Rowe demanding a refund – and accusing her of failing to properly communicate their rights as a consumer.
“They’d reference comments I’d made out of context by way of reframing what is a third-party contract matter into an alleged miscommunication on my part,” says the Twyford-based advisor. “I felt like I was being gaslit by a robot.”
Initially, Rowe tried to reason with the client by forwarding them a copy of the insurance policy. When this didn’t work, she decided to address their points one-by-one, only to find herself caught in a “paper warfare” with a robot representative.
“Instead of prompting ChatGPT accurately, they’d copy and paste my [human-written] responses back into ChatGPT with a different prompt, resulting in an endless [AI] prompt loop of new arguments,” she explains.
Rowe has since escalated the matter to the support team at Travel Counsellors, which told TTG it “recognises that AI-generated complaints can sometimes make already sensitive customer conversations longer and more complex".
The homeworking agency also encourages members to use its in-house AI assistant TC Co-Pilot and to “draw on the experience and judgement of our wider support teams when they may need additional guidance".
Why are AI-generated complaints so problematic?
Not Just Travel's Smith says one of the biggest problems with the rise of consumers using AI to formulate complaints is the potential for misinformation to spread.
Many consumers, she says, omit important details, such as the terms and conditions of an invoice, when prompting their AI tool of choice. And because the AI does not have access to a personal database or transaction records, it will often fill in any blanks by inventing facts or "hallucinating".
“Clients are not feeding the AI all of the information,” Smith explains. “They will go, ‘I’m really upset, so I’m going to tell my AI tool about it'. So the AI is like, ‘Yes, absolutely. You have all of these rights so let’s formulate a response. Let’s articulate all of your rights so you can get what you're entitled to'.”
Smith warns AI models often then overstate the consumer’s rights, because it isn’t using all of the relevant facts to build its argument. It also routinely cites laws and acts that aren’t applicable to the client's case, resulting in reams of legal jargon that only derail the user’s initial point.
Abta agrees AI’s tendency to complicate the argument is a growing frustration for agents.
“Feedback from members has shown that sometimes, AI supported complaints can include references to legislation or regulations that are not relevant or accurate, or expectations around remedies that may not be realistic,” Charles Fachiri, Senior Customer Information Manager at Abta, tells TTG.
'Mentally draining'
For agents, navigating these AI complaints can take up a huge amount of time. "It took several emails from the client to even understand what they were trying to say,” Rowe admits. “It was so mentally draining.”
It’s not just agents who stand to lose from this over-reliance on AI, though. Consumers are also suffering the consequences of taking everything AI platforms say at face value.
“I had a client whose villa burnt to the ground before they arrived due to an electrical fire,” one agent told TTG. “USAirtours proceeded to give them an upgrade worth £6,000 more in a nicer resort.
"But instead, out of their own pocket, they booked a separate villa and rejected the USAirtours one. They got nothing back. They paid for the holiday plus an extra £4,000. It’s a real lesson in why you shouldn’t listen to ChatGPT."
What is the solution?
There are currently no industry-wide guidelines on how travel agents should navigate this issue. But with AI becoming embedded in most consumer processes, there is likely to be a reckoning sooner or later.
While Abta did not share any plans with TTG to introduce an official policy, it “recommends customers ensure any information they include is accurate” when using AI to draft complaints.
It also warns AI’s susceptibility to “irrelevances and inaccuracies” can prolong the process because “businesses need to seek clarification before they can fully respond".
Rowe, though, believes the industry should be proactive. She suggests “clear customer-facing clauses” that would establish fair usage boundaries for AI, as well as mandatory disclosure for AI-generated complaints. “If a business is legally obliged to be transparent, clients should be too," she remarks.
'Stick to the bare facts'
In the meantime, Smith says it’s critical agents are trained in spotting AI-generated complaints. She believes agents should familiarise themselves with typical characteristics of AI script, such as long "em" dashes, as well as lists and subheadings, and an overly formal tone.
“A lot of agents just don’t know how to handle this right now,” Smith says. “They see this really long email and they panic.”
She urges any agent who feels intimidated by this style of message to “step back and move away” before replying. “Once you’ve come back refreshed, tease apart the facts from the noise," she advises. "You might find that only three sentences are real.”
Agents should then only address these specific sentences, and completely disregard anything that isn’t relevant or true. “The problem is the more you reply, the more you’re feeding the AI bot,” Smith adds. “So by only replying with the bare facts, you’re giving them less to come back with."
