A passenger of the Delta Diamond Medallion program submitted a complaint about a "frustrating" experience and the airline's response revealed the format in which the complaint letter was being composed. The email began with a remnant from an artificial intelligence tool: "here's a more polished and professional version with a warm, empathetic tone. The chatbot's reply was simply pasted into the reply without deleting the top instruction, and the incident soon made headlines among frequent flyers of the airline who are known to pay close attention to airline customer service.

What Happened

This story was flagged by JonNYC. The Diamond Medallion member was an elite and holds Delta's highest grade and had written about problems with Delta representatives using the airline's chat. The grievance was precise: They were given inconsistent information and at least one chat session prematurely closed despite their still trying to find assistance.

The reply expressed regret for the inconvenience and frustration and recognized those inconveniences in the first place and informed the member that being a valued Diamond, they deserve knowledgeable, attentive and seamless service. It also promised them that the airline was actively listening to them and the feedback was being presented to Delta's Flight Operations senior leadership team for consideration.

There's a lot of irony in this. A customer complained about the lack of personal, helpful service and received a response that was clearly from a chatbot, but before hitting send, the sender did not notice the giveaway line.

Is AI's use here the scandal?

Honestly, probably not. For 20 years or more, airlines have been automating their customer relations. Before generative AI, agents used to copy and paste pre-written paragraphs and modify a few elements.Before generative AI, agents would take templates and copy and paste the stock paragraphs, and make some alterations. Everyone who has ever interacted with airline customer service has encountered the classic errors, such as an email with a name in the salutation portion that doesn't have the recipient's name.

Some of the following suggestions are important to consider:

• No one thinks an agent will compose every response from scratch, so a person who uses an AI tool to edit a copy and paste response is just a little further ahead than the old copy/paste technique.
• The true error was keeping the prompt text in, which is more of a quality-control error than a sign the feedback has not been followed.
• The bottom line is whether the grievance gets to those who can correct the problem.

With regard to the latter, the e-mail states that the feedback was passed to senior leadership. If it's like that, the most helpful thing was that it did work, but the words that went with it were a bit embarrassing.

Why the Wording is Not as Important as Feedback Routing

The frequency of complaints form patterns that are used to inform operational decisions made by airlines. When many high-level flyers report the same problem, for instance, different chat agents providing different answers or sessions falling out of the conversation in the middle, this can shape the way the airline staffs, trains and develops its support channels. Individual e-mail messages don't usually make a difference; they just get added to the stack, and that is what leadership sees.

After all, the more significant question is not whether or not a chatbot assisted in the wording of the apology. It is if anyone at Delta is doing anything about the core issue of its chat support. It is nice to get a polished and thoughtful response, but if it's not followed by a sincere attempt to repair the damage then it means nothing at all. When it comes to frequent travel, reliability counts when it comes to a customer service team, particularly when flights are disrupted and quick customer service is needed, and that's where the Delta Chat may come into play.

The auto-hack is by no means a first

This isn't the first airline to be caught with its automation showing. American Airlines also had the attention in 2023, when they were criticized for their reply to complaints using a fake name "Gemma Flint.American Airlines was also in the news in 2023 for their automated reply to complaints signed by a fake name, “Gemma Flint.” I suspect that's still used in some form, with names switched around in messages to ensure that they don't come up too much.

A human read and sent an AI-improved response is probably a better option than a completely automated one. At least the sender of the e-mail is in the loop, even if he/she did not catch the top of the message. However, the Delta incident will certainly be a bad look and comes at a bad time, especially if it ends up in the in-box of a customer who's already an unhappy one for the quality of the airline's support.

What travelers can take away

If you've ever written to an airline regarding issues, this episode will remind you to have some realistic expectations. Expect that a response might be somewhat or entirely scripted and be polite instead of being sincere; polished and empathetic responses should not be interpreted as sincere. 

To ensure your complaint is heard, here are a couple of tips:

• In your message provide the issue, date, flight number & confirmation code to ensure correct routing.
• Specify the resolution that you want such as a refund, miles, or a solution to a process.
• Make a note of chat logs and emails in case you need to escalate later.

Adhering to these steps won't ensure your message gets a better reply, but it makes it more difficult for an email to be dismissed with a cookie-cutter message, human or otherwise.

The Bottom Line

A Delta Diamond Medallion got a reply from customer service that also included the AI instruction used to create it and a humorous and somewhat embarrassing mistake for an airline that prides itself on its superior service. But few people think that airlines compose every reply from scratch, and templates and automation of responses is a well established practice for years. 

The bigger lesson here is that the airline's quality control failed to spot a glaring problem and the true measure of the complaint's success would be its impact on those who could make a difference for chat support. Maybe Delta should build on its AI process with one more instruction: Avoid using the word "AI" in the response.