Marketing in 3 · Issue 39 · Week of 5 October 2026
The damage splits in two. When people feel the brand let them down, they respond with stronger feelings and doubts about the brand's intentions. When they feel the chatbot let them down, they act, by recommending it less, complaining or warning others. A bad chat can also push people to quiet withdrawal and indifference. In interviews, people with a good history with the brand were more likely to complain privately and constructively.
Brands now put chatbots in front of customers at the moment they most need help, such as a changed flight, a lost bag or a refund. When the bot fails, the customer has been let down twice, once by the machine and once by the company that put it there. The researchers at HEC Liège asked whether those two letdowns lead to different reactions.
They used two linked studies. The first was a survey in which Belgian adults imagined losing their luggage on a flight and asking a well-known mid-priced Belgian airline's chatbot for help. The bot gave poor answers again and again, shown in a scripted chat. People rated how much they trusted the brand and how much they expected from the chatbot before the scenario, and again afterwards. The drop in each was the measure of how far the brand or the bot had let them down.
The second was a set of interviews with 36 people who had actually had a bad branded chatbot experience in travel. This was meant to show what the reactions looked like in real life.
Both expectations fell after the failed chat. What differed was where the damage showed up.
People whose trust in the brand fell the most reported more negative feeling and more negative thinking about the brand. They felt more anger and frustration, and they were more doubtful about the brand's intentions. The brand drop was linked more strongly to the thinking than to the feeling.
People whose confidence in the chatbot fell the most behaved differently. They were more likely to take action against the brand, with intentions such as warning others, complaining or recommending it less. The same drop was also linked to more helpful action, such as giving feedback. The harmful side was slightly stronger.
The pairing was clean. Brand letdown did not clearly predict behaviour, and chatbot letdown did not clearly predict feelings or thoughts. The authors read this as the brand being a relationship and the chatbot being a tool. When a relationship is damaged, people feel and think. When a tool fails, they act.
| Group | Value () |
|---|---|
| Brand letdown to feelings | 15 |
| Brand letdown to thinking | 22 |
| Chatbot letdown to helpful action | 18 |
| Chatbot letdown to harmful action | 20 |
Study 1, Table 3, standardised paths from the structural model (N = 238). Only the four significant paths are shown. Brand letdown to behaviour and chatbot letdown to feelings or thinking were not significant.
The interviews added a view the survey could not. Negative engagement came in an active form and a passive form, sometimes in the same person.
Customers with high expectations of the brand, such as those who thought of it as premium or who were loyal, tended to react with anger and a feeling of unfairness. Others who felt abandoned in a moment of need went the other way. They said they felt nothing toward the brand any more and just would not deal with it again.
Their thinking showed the same split. Some became cynical and believed the chatbot exists to wear customers down so they give up on refunds, and that the brand, not the technology, was to blame. Others simply lost confidence in the whole support system.
Behaviour varied too. Most interviewees wanted to warn others publicly. A smaller group with a good history with the brand stayed measured. They complained privately, told a few close people, and left the door open for a future booking.
The authors' advice follows the split. Set honest expectations before the chat starts, for example by saying what the bot can and cannot do and presenting it as an assistant and not as a stand-in for a person. That limits how far the chatbot's reputation falls, and it keeps a bot failure from becoming a brand failure.
After a failure, match the response to the reaction. Angry customers need a fast, personal apology and a human. Quiet, withdrawn customers need someone to reach out, because they will not complain and may not return. Customers who offer feedback should be thanked and shown what changed.
The consistent theme was a route to a human. Interviewees who were still positive about the brand said the chatbot should not be the only door.
Marketing in 3 steps
Yes, in this study it did. Customers who had been let down by the chatbot were more willing to warn others and complain, and those let down by the brand felt more anger and doubt (El Midaoui, Dessart & Standaert, 2026, Journal of Product & Brand Management).
Both, and for different things. Trust in the brand drove feelings and thoughts, and confidence in the chatbot drove actions. In interviews, many people held the brand responsible for the bot it chose to deploy.
Unfavourable thoughts, feelings and actions toward a brand after a bad event. It can be active, such as anger and public complaints, or passive, such as indifference and quietly leaving.
Sometimes. The survey linked chatbot letdown to helpful action as well as harmful action, and interviewees with a good history with the brand were more likely to complain privately and suggest improvements.
Set expectations about what the chatbot can do before the chat, give an easy route to a human, and respond differently to angry customers and to those who simply withdraw.
Psychology & Marketing · 2026
Olya, Nazifi, Taheri, & Li
In six studies set in live-streaming contexts, generative AI produced higher repurchase intention than human agents after a single failure, and humans did better after a second one. High empathy brought AI level with humans on repeat failures, and compensation put it ahead. Use AI for a first slip and a person for a second.
Journal of Service Research · 2026
Fürst, Trißler, & Friedrich
Across three main studies and three supplementary ones, explaining a service recovery decision raised satisfaction and choice by making the agent seem more competent and warm. Rationales beat excuses and justifications, specific explanations beat general ones, and explanations from AI agents could be more impactful than those from humans. Explain the decision, specifically.
El Midaoui, Y., Dessart, L., & Standaert, W. (2026). Unpacking negative brand engagement: the role of brand and chatbot expectation violation. Journal of Product & Brand Management, 35(7), 921-936. https://doi.org/10.1108/JPBM-03-2025-5889
This brief is written for practitioners from the published article. Numbers are as reported by the authors; read the article for full methods.
Customers are starting to delegate shopping to AI agents. A Journal of Marketing framework sets out five things firms must get right, and six KPIs.
When Italy blocked ChatGPT for four weeks, hotel reviews written on desktop fell 2% in number and 9% in length. AI help is quietly reshaping reviews.
Among shoppers still browsing, 80% chatted with a plain chatbot and only 51% with a personalized one. Shoppers with a clear goal reversed.
Bruno Schivinski is Associate Professor of Marketing at Gdańsk University of Technology. He studies how consumers engage with brands in digital and social media.
If this issue saved you a meeting or sharpened a campaign, buy me a coffee. It keeps Marketing in 3 free, weekly and ad-free.
