Reprinted with permission from the Auto Insurance Report by Risk Information
The world is enthralled with ChatGPT’s ability to forge homework assignments, replace junior-level software engineers and write stilted poetry. But insurance companies have largely ignored using the latest generative artificial intelligence technology for its customer service chatbots.
Instead, many have backtracked their offerings, according to Corporate Insight‘s 2025 analysis of property/casualty insurance chatbots and virtual assistants.
“That was definitely a surprise to us,” said Justin Suter, research manager at Corporate Insight.
Suter figured that insurers would implement some form of AI-based large language model (LLM) tools, like OpenAI‘s ChatGPT and Google‘s Gemini, to improve chatbot communication with more fluid responses.
Instead, insurers have continued to rely on a decision-tree model that keeps customers within the guardrails of preset prompts. Surveys show that consumers today largely don’t trust current insurance website chatbots and generally don’t like them overall.
“As soon as you start to try to use free text inputs or ask it to do something that it’s not completely pre-programmed to do, it’s pretty much an auto-failure here and you’re connected with a live agent,” Suter said.
Insurers may be hesitant to expose themselves to liabilities from hallucinating AI chatbots that have the potential to provide inaccurate information. Still, the lack of advancement among insurer website chatbots is striking during the current LLM gold rush.
Most industries have adopted some form of generative AI, with Bain & Company reporting that 95% of U.S. companies use it in some way, up 12 percentage points from last year. Insurers use generative AI in other parts of the business, but Bain found that other industries have more widely adopted the tool for customer service.
While they have largely ignored the technology with text-based chatbots, insurers are experimenting with LLM-powered voice AI in limited applications. AAA Mountain West Group is using Replicant voice AI assistants to make outbound calls to help its members with roadside requests, the company said in a recent webinar.
Rather than using generative AI to expand the functionality of chatbots, many carriers have pared back their abilities by removing the ability see the chat history. Corporate Insight’s analysis recommended that insurers retain chat history and offer users options to review or save the history as a best practice. But the number of carriers that allow access to chatbot transcripts fell to 17% this year from 64% in 2023.
“I’m puzzled by that,” Suter said. “I wouldn’t put [chat history] at the top of any priority list of how to build an excellent chatbot, but it’s a feature that’s been there and now isn’t.”
Some insurers have also removed preset menus at the top of a chat, the report found, which tend to save customers from having to type in their questions.
Insurers have done little to advance their chatbots since the last Corporate Insight analysis two years ago, though many have at least expanded decision trees. The four insurers rated as “Excellent” in this year’s Corporate Insight report – Amica, Chubb, The Hartford and State Farm – all had larger decision trees than in previous reports.
“The most important things that people [look] for are questions about billing and questions about their policy,” Suter said. “Understanding those topics … is better than nothing. It’s better than saying, ‘I don’t understand that.'”
The report also recommends that chatbots populate basic policy information and perform simple tasks within the window – both features that customers find helpful.
Consumer confidence in chatbots remains limited to these basic tasks, according to policyholder surveys from Corporate Insight. About 59% of those surveyed said they would trust a chatbot more or as much as a customer service agent to search for a feature. About 49% trusted chatbots to look for information on spending or budgeting, 48% for self-service tasks, and 46% for connecting to the appropriate service specialist.
“If you can take care of the low hanging fruit in a seamless way [with chatbots], that’s a big win,” Suter said. “You’re taking away calls from your customer service line and you’re saving money that way. That’s where insurance can focus first.”
Policyholders were less likely to trust chatbots in several other areas. Survey respondents trust human representatives more for tasks like troubleshooting issues with their account, with 62% trusting a live agent more. Additionally, 55% trust live agents to make a transaction on their behalf. Most people do not want chatbots to take direct actions that could impact their account or finances, according to the survey.
Trust in insurance chatbots may take time to build, as confidence is low among customers today, according to the report. Only 12% of customers said that they were very or extremely confident in chatbots, while 50% said they were a little confident or not confident at all.
This year’s report has the same number of companies with “Excellent” – the highest rating – as 2023’s report. The stagnation was especially surprising to Suter, because Corporate Insight had noticed growth and improvements to chatbot technology in the previous two reports and expected more companies to grow and more overall advancement within the industry.
The ability to seamlessly transfer to a live agent is one of the most important functions for a chatbot.
Suter said it’s likely that insurers took a step back due to both the conservative innovation culture in the industry in addition to growing legal and compliance concerns.
Notable legal troubles have emerged for businesses that give AI chat tools too much agency. In 2024, Air Canada‘s chatbot promised a discount to a customer that the airline wouldn’t honor. While Air Canada argued that the chatbot is responsible for its own actions, Canadian courts held the airline liable and ordered it to pay for the discount the chatbot promised the customer along with his court fees.
The potential for costly chatbot mistakes is likely on the minds of hesitant insurance carriers, Suter said, where the potential losses or damages could be much larger than an airline ticket.
Still, chatbots and other AI-based tools will be hard to ignore as they play a growing role in American life.
Insurance industry chatbots are already falling behind those used in other finance sectors like banking. Bank of America‘s Erika chatbot can understand text requests, provide account information, and redirect customers to essential areas of the website.
“That’s the challenge that insurers are up against, and that’s where the bar for consumer expectations is being set,” Suter said. Policyholders are far more likely to interact with a bank or airline’s chatbot than an insurer’s chatbot, he added. “And that’s a bar [insurers] are not meeting at the moment.”
One of the most important chatbot functions is the ability for a smooth handoff to live agents when needed. Most customers (64%) said that they would wait less than two minutes in interacting with a chatbot before requesting to talk with a live agent, the report found. A large portion of policyholders don’t like current insurance chatbots at all. Corporate Insight found that 66% will disconnect from support chat as soon as they realize they aren’t talking to a live person. “Customers have an expectation that this is going to speed up the process, not that it’s an unnecessary aspect of their search for help,” Suter said.
Rather than going all in on chatbots, some carriers have embraced other text-based solutions that use some of the same language processing technology but are specifically intended to funnel conversations to a real person.
Drips, which offers conversations-as-a-service (CaaS) technology, works with insurers on outbound outreach, texting customers and potential customers to set up calls with live agents. For example, if a potential customer had requested a quote but didn’t follow up, it will send a message, hoping to get them on the phone with a real person.
Drips isn’t a chatbot, but it does use natural language understanding (NLU) to interpret what people say in their text response. This service can automatically schedule phone calls, responding to customers with pre-written prompts to put them into different categories for live agents.
“The AI comes in for understanding what they’re saying back and then understanding the intent,” according to Mark Birkenstock, manager for the compliance marketing campaign at Drips. “We reached out to someone about their bill and the AI determined they’re saying that they’ll pay it next week when their paycheck comes in, or they already did pay. It can go from there to route appropriate responses.”
Drips uses “polite persistence” to follow up with customers, said Maddy Hubbard, who manages the company’s property/casualty marketing campaign. People are often busy or forget to make a payment, she said, but an outreach AI service like Drips can help remind people.
Following up is also a best practice for insurers using chatbots or virtual assistants, Suter said. Questions like “was this helpful?” and “do you have more questions?” at the end of a chatbot interaction can help the company refine its chatbot offerings to effectively answer all customer questions. However, only 50% of the tools Corporate Insight analyzed perform these follow-up questions.
“Consumers appreciate those throw-in follow ups,” he said, “and it does make it feel more like a real conversation and not like something that’s truly just this canned bot responding to you.”
While carriers were relatively late to introduce chatbots, Suter said, several vendors now offer virtual assistants and chatbots specifically for the industry. These include PolyAI, Nurix and Regal, in addition to the chatbot platforms offered by tech giants, like IBM‘s Watson, Google’s Dialogflow and Amazon‘s Lex.
As generative AI evolves, insurers face the challenge of balancing consumer expectations formed by other industries with the risk of dispensing incorrect information. Birkenstock at Drips warns that even a small fraction of messages that give misinformation or make bad promises can bring lawsuits and bad publicity. Tolerance for errors is low in insurance.
“We think that insurers should be open to technology with due diligence in mind,” Birkenstock said. “It’s a balance of taking the proper steps but not being completely stuck in the old ways.”
Justin Suter is the director of thought leadership at Corporate Insight.