Over the past year, a striking pattern has shown up in the insurance industry's approach to artificial intelligence: when carriers, brokers and technology vendors have gone looking for a large language model to put a name to, an unusually large number of the marquee announcements have gone to Anthropic's Claude.
Baldwin Group expanded its enterprise relationship with Anthropic in May, a deal the outlet noted followed broker HUB International's earlier decision to roll Claude out across its entire workforce of more than 20,000 employees, which HUB said produced an 85% productivity increase in targeted tasks and better than 90% internal user satisfaction.
Data and analytics giant Verisk has since built dedicated connectors so underwriters can query loss-cost trends and ISO filing data conversationally inside Claude, and IT services firm DXC has committed to training tens of thousands of engineers on the model as part of a multi-year partnership that names insurance as one of its priority sectors.
New consumer research suggests those buyers may have been onto something the general public has also noticed, even if for different reasons. But as the data below shows, the announcements only tell part of the story - the actual mix of models running inside US carriers looks less like a Claude sweep and more like a multi-vendor patchwork, with real money riding on getting the match right.
YouGov's BrandIndex tracking, the same always-on survey engine that produces the network's regular brand health reports, has just published the third and final installment of a three-part US series on how Americans perceive AI tools. The first two parts looked at which brands people prefer and find accurate; this one asks something closer to a customer-experience question: among people who have actually used an AI tool, are they happy they did?
On that measure, Claude comes out on top. Anthropic's assistant posted a net satisfaction score of 59.3 among current and former US users surveyed between February 1 and July 31, 2026, ahead of Apple Intelligence at 57.7, OpenAI's ChatGPT at 53.5, Google's Gemini at 51.3 and Perplexity AI at 49.0.
DeepSeek, Amazon's Alexa, Microsoft's Copilot, Apple's Siri and xAI's Grok rounded out the top ten, with Grok lowest of the group at 39.0. The score is a net figure - the share of respondents who say they're satisfied minus the share who say they're dissatisfied - and it is a noticeably different leaderboard from the one produced by pure market share.
ChatGPT remains the runaway leader when it comes to sheer preference among US AI users, a position YouGov's companion report put at roughly a third of the market. But satisfaction is a narrower and arguably more demanding test: it only counts people who have actually spent time with a product, rather than people who simply recognize the name or say they'd consider it.
Apple Intelligence and Perplexity, both of which sit toward the bottom half of the field on raw preference, land in the top five once the question shifts to how people who've actually used them feel about the experience.
The most interesting split in the data, though, is between current and former users of each brand - a proxy, in effect, for how a product wears once the novelty fades. Among people still using the tools today, satisfaction is high across the board: DeepSeek edges out Claude for the top current-user score, 76.4 to 76.0, with Apple Intelligence (75.1), Perplexity (73.4) and ChatGPT (72.5) close behind. But once people stop using a product, the picture diverges sharply.
ChatGPT's satisfaction score among lapsed users falls to -2.8, a swing of 75.3 points from its current-user figure and the widest gap of any brand tracked. Gemini shows a similar, if less extreme, pattern, dropping from 69.7 to 10.7. Claude and Apple Intelligence, by contrast, retain some of the highest residual goodwill among people who have moved on, with former-user scores of 30.4 and 33.5 respectively.
For an industry that has spent the past two years racing to bolt generative AI onto underwriting, claims and customer service workflows, that current-versus-former gap is a useful cautionary note. A model that scores well on day one but leaves a sour taste after a few months of production use is a different procurement risk than one that holds up over time - and it's a distinction that raw popularity rankings, or a flashy pilot demo, won't necessarily surface.
It's also a data point that cuts two ways for carriers thinking about where to meet customers rather than just where to put their own workflows. A wave of insurers has spent 2026 experimenting with distribution built directly on top of AI chat interfaces rather than their own websites. Spanish digital insurer Tuio became the first carrier to launch inside ChatGPT's app ecosystem in February, and by midyear Liberty Mutual had become the first major US carrier to follow suit, rolling out an auto-quoting app that launched in seven states with a roadmap to cover most of the country by the end of the year. Broker platform Simply Business has taken a more cautious step, linking ChatGPT to its small-business pricing enginepurely as a discovery layer that still funnels shoppers back to its own platform to actually bind coverage. The stakes around this shift became obvious when US intermediary Insurify launched a comparable app: several major brokerage stocks - including Willis Towers Watson, Aon and Arthur J. Gallagher - dropped by approximately 8% to 11% in a single session as investors weighed whether conversational AI might sideline traditional intermediaries altogether.
None of that activity is happening on Claude, at least not yet - ChatGPT's status as the default consumer AI assistant makes it the natural place for insurers chasing top-of-funnel discovery, whatever the satisfaction numbers say about what happens after someone stops using it.
YouGov's data doesn't settle which strategy is right. But it does suggest that the two markets insurers are chasing - attention now versus goodwill later - may not always point to the same brand.
The headline deals tell one story; carrier technology stacks tell a messier one. A survey of national and regional property-casualty and life carriers by IA Capital Group, reported in May, found OpenAI's technology present in about nine out of every ten carrier stacks - frequently arriving indirectly through Microsoft Copilot, which showed up separately in 27% of stacks.
Anthropic had meaningful reach at 55%, but typically as a secondary model layered alongside OpenAI rather than a primary replacement for it. Google's Gemini, notably, didn't turn up in a single surveyed stack. The same survey found carrier AI use is still mostly cautious: 64% of deployments are internal-only, 36% route AI-drafted output through a staff member before it reaches a customer, and just 18% let AI act directly in front of policyholders.
That picture - one dominant model plus a specialist second model, rather than a single winner-take-all platform - matches what Insurance Business's own model-by-model guide for brokerages recommends rather than what any single vendor would prefer you conclude. That guide, published in June, argues brokers get the best results using GPT-5.5 or Claude for analytical, document-heavy drafting where accuracy carries compliance risk; Microsoft Copilot as the operational layer for firms already embedded in Microsoft 365; and treating Google Gemini or xAI's Grok as narrower, situational tools rather than primary platforms. In other words, the Claude deals making headlines and the ChatGPT/Copilot combination running quietly underneath most carrier operations aren't necessarily competing for the same job.
Insurance-sector AI contracts have also gotten considerably bigger, which raises the stakes of a wrong or short-lived platform bet. Average deal size for insurance AI transactions reached $23.23 million in the second quarter of 2026, the highest level recorded since late 2021, according to market tracking from ScienceSoft. Pricing structures also aren't as simple as they look on a vendor's website: Insurance Business's brokerage guide notes that Microsoft's advertised $30-per-seat Copilot Business tier actually runs to roughly $42.50 per user per month once the mandatory Microsoft 365 base subscription is factored in - two to four times the cost of a standalone Claude or ChatGPT subscription - while Anthropic's enterprise pricing for Claude isn't published at all and requires a direct sales conversation, making upfront cost comparison difficult for buyers trying to budget before they commit.
None of that is unique to insurance, but it matters more in a sector where AI workflows tend to get built around a specific vendor's document handling, context window and integrations rather than swapped out casually. A carrier that commits budget and retrains staff around one model's quirks is choosing a multi-year relationship, not a monthly subscription - which is exactly why current-user satisfaction, alongside former-user regret, is worth watching as closely as the deal announcements.
The figures above are drawn from YouGov BrandIndex US data collected between February 1 and July 31, 2026, covering 18 tracked AI brands, with the published satisfaction tables based on samples of more than 530 current and former AI tool users (more than 220 current users and more than 280 former users for the current-versus-former comparison). Notion AI and Mistral were excluded from the satisfaction results due to low sample size, and several image- and enterprise-focused tools were excluded from the broader rankings to keep the focus on consumer-facing, general-purpose assistants.
