Artificial intelligence assistants like ChatGPT are filling an information void left by insurers, and are increasingly influencing shoppers who research coverages and compare policies, according to research from J.D. Power.
The emergence of AI assistants could transform how shoppers and policyholders evaluate insurers and coverages, and some carriers are concerned that control of the message is slipping to third-party AI software.
One-third of consumers who researched insurance online used AI offered by Anthropic (Claude), Microsoft (Copilot), OpenAI (ChatGPT), Google (Gemini) and others to better understand the complicated policies sold by insurers, according to a survey fielded by J.D. Power in June and July.
Slow to adopt AI tools on their websites and apps, most carriers are missing opportunities to define their own products and strengthen relationships with customers who now turn elsewhere for information.
Some insurers and comparison sites, however, are meeting shoppers where they are, building plug-ins that integrate their quoting tools straight into AI platforms like ChatGPT.
“It’s probably no shock, the third-party tools perform much better than the carrier tools,” said Tony Soloman, director of customer solutions for insurance at J.D. Power. “Consumers are more and more starting to turn to AI tools, and it’s not surprising. If you go to Google for a search right now, the first thing that comes up are the Gemini AI results.”
While more consumers are using AI themselves to research and compare coverages, many remain uncomfortable with insurance companies using AI to automate policy changes, claims and other high-stakes processes.
AI assistants could have a profound impact on how shoppers and policyholders evaluate insurers and coverages, Soloman said.
An earlier J.D. Power survey published in January showed that people who used an AI tool before shopping were far more likely to switch carriers than those who did not, likely because the technology made consumers more confident about coverage definitions and other terms.
About 51% of those who used a tool switched, compared to 38% who did not.
Soloman is interested in researching whether AI tools could also have the opposite effect depending on the individual and market conditions: A shopper could upload their policy and premium information, and the bot might tell them not to bother shopping because they already have a good deal.
Some carriers and comparative shopping sites are launching tools that connect their rating engines to popular AI platforms. Plymouth Rock, Liberty Mutual, Insurify and others now offer plug-ins that shoppers can install into ChatGPT to gather quotes during an AI conversation.
Once a shopper adds the Plymouth Rock plug-in to ChatGPT, for example, they can enter their name, address and birthdate and request a homeowners insurance quote in the six states where the company offers coverage. The process is not seamless yet: After delivering the bindable quote, the app sends shoppers to the carrier’s website to finish the sale. The Liberty Mutual plug-in generates auto insurance quotes, and is now available in every state except Colorado. Like Plymouth Rock’s tool, users must proactively add the plug-in to their ChatGPT account to be able to initiate the quote.
The J.D. Power survey also revealed a surprising willingness from consumers to call just about everything “AI,” including carrier customer service tools that are inferior to the latest AI models. On insurance company websites and apps, many respondents used “AI” to describe older chatbots that only deliver pre-written answers or preset menu options and bear little resemblance to ChatGPT or other assistants that use large language models to generate conversations.
Carriers contribute to this confusion by often calling rudimentary chatbots “AI” even when they do not produce interactions like those with Claude, ChatGPT or other modern bots.
Geico’s “AI Assistant,” for example, may be artificial, but it’s hardly intelligent. It is incapable of answering coverage questions and quickly offers to connect users to a live agent. When asked whether a “homeowners policy covers contents,” the bot replied, “Sure thing. To get a quote for a renters policy, visit our Geico homepage …” Virtual assistants from State Farm and Progressive were similarly incapable of providing coverage definitions or other useful information. Both provide reference links when asked about policy basics.
Of the top 10 auto carriers that have public-facing chatbots, only Farmers’ assistant provided useful answers to questions.
Even once a policyholder logs in to a large insurer’s website or app, most virtual assistants are no better. Allstate’s “AI digital assistant,” for example, is only available to help policyholders with payments and app support. It also responds with canned responses, or preset menu options. It quickly offers to connect users with a live agent when they ask simple questions like “how much is my monthly payment?”
In contrast, ChatGPT, Claude and others provide detailed coverage descriptions and common policy choices followed by prompts for more specific information.
Eleven of 18 large P/C carriers analyzed by consulting firm Corporate Insight have a virtual assistant on their websites and apps. But none are AI-powered and instead offer rules-based, predetermined responses, according to a June report.
Nevertheless, many consumers reported using essentially nonexistent “AI tools” to research coverages on insurance company websites and apps.
Insurers have lagged in adopting the latest AI technology, likely because the industry is both slow to innovate and cautious about the risks of AI offering incorrect advice. But other financial services industries with similar liability risks, like personal finance, have managed to integrate modern AI tools into their web experience. Charles Schwab, for example, uses an AI-augmented search function, said Justin Suter, director of market engagement and thought leadership at Corporate Insight.
The central role of human agents in distributing many insurance products has also likely hampered the growth of AI tools among insurers, according to Corporate Insight’s report.
But consumers are gravitating to the AI tools whether or not insurers provide them. About 29% of consumers in the J.D. Power survey said they researched insurance for personal needs on the internet. Of that group, about one-third said they used third-party AI software like ChatGPT, Claude, Gemini or Copilot. Another third said they used an AI tool on an insurer’s website or app, and the remaining third said they did not use AI.
Few policyholders said they used an AI tool to file a claim. Of those that did, more than half had to follow up with a human contact. Less than half of claimants feel the insurer claims AI tools “definitely” saved time.
The top reason given by people who did not use AI tools was that they were “not familiar enough,” at 36%, followed by the “use of non-AI tools out of habit,” at 30%. About 29% said they did not trust results, 25% were concerned about data privacy, 21% were concerned with security, 11% had ethical concerns, 8% had environmental concerns and 3% said they had no access.
The third-party AI assistants like ChatGPT received higher marks from consumers than the tools provided by carriers, reflecting how primitive carrier tools are in comparison.
More survey respondents said AI like ChatGPT “definitely” helped them save time compared to tools offered by insurers in four out of six categories. Third-party tools were bigger time savers for understanding insurance terminology, researching available insurance policies, and getting recommendations about which insurers and which endorsements they should consider.
Insurer tools were named as time savers more often in two categories: getting recommendations for coverage amounts and deductibles and for choosing specific coverages.
Suter said that older chatbots that don’t use large language models like ChatGPT and its ilk could frustrate policyholders and deter them from using better tools provided by insurers in the future, especially if the older technology is mislabeled as “AI” today. Failing to accurately label bots and virtual assistants as AI can also frustrate consumers in the opposite direction. If a bot that uses large language models for conversational responses doesn’t disclose that it is artificial, it could anger consumers who feel tricked.
About 58% of people surveyed by Corporate Insight disconnected from a support chat when they realized they were talking to an AI who they had thought was a human, Suter said.
“Now maybe they’re not disclosing it as AI because it isn’t really AI,” he said. “Maybe it is just a rules-based chatbot, and that’s why they don’t have that disclosure there. But because of all the noise in the news, people are assuming that you are using AI.”
Two-thirds of consumers assume that every brand they interact with is already deploying AI in some way in the experience, he added.
California, Colorado and several other states have recently proposed or passed laws requiring AI bots to disclose that they are not human.
The new J.D. Power report also showed that most policyholders are uneasy about how insurers use AI to automate business. Policyholders were most uncomfortable about insurers using AI to make changes to their current policies. That was followed by processing claims, using AI for customer service and personalizing policy or coverage recommendations.
Policyholders were more comfortable with insurers using AI to manage billing and for calculating premiums.
Unease with insurers’ AI-initiated policy changes was consistent across baby boomers, Generation X, millennials and Generation Z.
Gen Z and millennial policyholders reported using more AI tools than their older Gen X and baby boomer counterparts. The four generations reported similar usage levels across some categories, like for using the tools to research recommended coverage amounts and deductibles.
Insurers are concerned about the growing influence of third-party AI assistants with policyholders, Soloman said.
“Some of that decision making, some of that control is now out of your hands,” he said. “Because what used to flow through a call center or your website or your agents is now going someplace that you have really no control over.”