Artificial intelligence chatbots are becoming the new standard for managing products, complaints and refunds – critical customer satisfaction interactions, brand perception and sales. According to a recent survey, 85% of customer service leaders are already exploring a customer -oriented AI solution. If your company does not have one yet, it will probably have soon.
However, customers are coming out of interactions with chatbots feeling dissatisfied. Research shows that 77% of people find chatbots frustrating, and 88% still prefer to talk to a real person.
Most companies try to solve this by developing better AI models for their chatbots, thinking that if the models sound human enough, the problem will disappear. But this assumption is wrong.
Continues after advertising
Even when chatbots reach almost perfect realism (in recent blind turing tests, 73% have confused ChatgPT-4.5 with one person), customers still don’t like them. This is because the root problem is not technological, it is psychological.
In a study on chatbots interactions, customers who had identical conversations with people and chatbots reported being 8.5% less satisfied after interacting with Chatbot. But when the benefits of chatbot (such as “instant answers” or “available 24/7”) were explicitly highlighted and delivered, people were 37% more satisfied with chatbot than with slower and less available human agents.
The next chatbots jump will not just come from better AI models. It will come from the recognition that with AI, the customer experience is no longer an engineering challenge. It is now a challenge of human psychology. Your chatbot is not just the customer’s assistant – it’s your brand ambassador and often determines whether people decide to buy from you again or not. Make sure they are designed to make customers happy and boost their sales.
Continues after advertising
We recently analyzed the latest research on chatbots and ia-human interactions and turned it into a practical guide for effective AI chatbots that promote satisfaction, positive brand perceptions and sales.
Here are six of our favorite ways to improve your chatbot:
Label your chatbot as “in constant learning”
People don’t just want smart chatbots. They want chatbots that get smarter. Research shows that customers are up to 17% more likely to follow the suggestions of a chatbot when it is labeled as “in constant learning” compared to a static algorithm – or even a human agent. Without this contextualization, customers can be suspicious of the board, assuming that it is outdated or frozen in time. Avoid static language as “last update in 2023”. Instead, emphasize that your chatbot is continually improving with new data and feedback. It’s like working with a new enthusiastic contractor: people forgive small mistakes and enjoy that each interaction will be better than the previous one.
Continues after advertising
Prove the accuracy of your chatbot
Customers trust results, not technical explanations. Give proof that your AI chatbot really works. Showing examples of success in the real world can increase confidence and expectation of accuracy by up to 22%. Display metrics such as “attended 4,500 consultations today, with 94% accuracy” or “98% of customers found it useful”. Think about it as a product testimonial: “This dishwasher has cleaned more than one million dishes without problems” is much more persuasive than a detailed explanation of how the rinse cycle works. With AI, proof is worth more than promise.
Subtle compliments make recommendations more persuasive. Listening to “Yes, this bag matches your fashion style” is pleasant and makes people up to 12.5% more likely to accept product recommendations. But there is an important caveat: if the compliment seems very human, it may have the opposite effect. Excessively personal language generates suspicion, making customers feel manipulated. Keep compliments based on facts, objectives and aligned with the expected tone of an AI.
Behave like a person to reduce fraud
Customers act differently when interacting with chatbots, and not always for the better. Research shows that customers are up to 35% more likely to act unethical (for example, lie to get a refund) when dealing with a chatbot compared to a human agent. Interacting with a machine seems less personal and reduces the feeling of guilt. To combat this, set your chatbot to behave more like a person in these situations. Friendly language, empathic phrases and natural interjections like “Oh, no!” Or “I understand” they help create an emotional connection, increasing guilt and reducing unethical behaviors by almost 20%.
Continues after advertising
When customers are annoyed or in a hurry, they want speed and clarity, not empathy. Research has shown that annoyed customers were 23% less satisfied with interacting with a humanized chatbot compared to a more direct and mechanical. In times of stress, such as reschedule a canceled flight, humanized answers do not help. They raise expectations and may increase frustration when they are not met. Instead, schedule your chatbot to respond quickly, clearly and directly to detect anger or urgency. In such cases, it is better not to look “human” at all.
Delivered positive decisions in a humanized way
When your chatbot brings good news – perhaps approving a loan, reimbursement or free upgrade – make it look warm and human. Customers evaluated companies 8% better when positive results were communicated by a humanized chatbot. People tend to assign positive results to themselves and their qualities (“It’s because I’m great!”) And negative results to external factors. A humanized chatbot takes advantage and amplifies this emotional impulse. “Hi, I’m Ava – great news! You have qualified for our premium program” is much more rewarding than a simple “you have been approved.”
c.2025 Harvard Business Review Distributed by New York Times Licensing.
