Insurance claims adjusters have turned out to be the harshest critics of AI on Glassdoor: 98% of their reviews that mention the technology are negative. The sentiment number arrives next to a payroll number that explains most of it. In 2024 the US Bureau of Labor Statistics projected the profession would shrink by 18,900 people, or 5%, over the following decade. The same agency recorded a 21% drop in employment in the field between May 2025 and May 2026. The decade-long forecast was overtaken inside a single year.
The complaints themselves are specific rather than atmospheric. Adjusters writing on Glassdoor object to being made to use AI in place of their own judgement, to it being pushed on everyone regardless of the work, and to it arriving as badly built applications. The recurring target is management: executives fixated on AI, imposing systems on staff and on customers that get things wrong often enough to matter. When the software fails, people clean up after it.
Ahmad Jackson worked in the claims department of a large insurer about a year ago. His employer deployed AI for first notice of loss — the system was meant to open a case and gather details as soon as a customer got in touch, handling routine matters and passing complicated ones to staff. What Jackson and his colleagues got instead was a stream of misclassified claims that had to be rerouted to the right departments, and case summaries in which the model hallucinated. When he passed that material on to customers or their lawyers by mistake, the anger landed on him. He left soon afterwards for another insurer. In his account, AI errs regularly and ends up creating work for adjusters rather than removing it.
Jeffrey Conrad, who heads a claims department in Mobile, Alabama, says his staff are worn down by the sheer quantity of AI being pushed at them. He calls it AI fatigue.
Chris Martin, senior economist at Glassdoor, did not expect this profession to come out so negative. He rechecked the results and concluded that the industry is going through a serious restructuring.
The labor data supports that reading. BLS named technology as one of the main drivers of the decline. Glassdoor separately recorded a 50% fall in entry-level job postings since 2025. Meanwhile Liberate and Pace, both building AI for insurers, have raised millions of dollars on the promise of reinventing the business, and carriers are routing more claims through software.
What that software does is now fairly concrete. A customer filing a claim may deal with a text assistant instead of a person. A damaged house may be assessed from photographs and video rather than a site visit. Models summarize hundreds of pages of medical records and pull out the key details. A system can take a customer's uploaded photos, video, documents and receipts, calculate a payout and send it within seconds.
Lemonade has been aiming at exactly that since it was founded in 2015, promising to replace bureaucracy with software assistants and machine learning. By the end of last year its AI assistant Jim was handling 96% of first notices of loss, and automation covered roughly 55% of all claims. Those two numbers deserve to be read against each other. First notice of loss is the intake form — structured, repetitive, the easiest thing in the workflow to automate — and ten years of building for it as a founding mission produced automation across a little over half of claims overall. The frontier here moves slower than the funding rounds suggest.
More traditional carriers position themselves differently. State Farm stresses that claims handling needs a combination of human and digital expertise, and spokesperson Justin Tomczak told WIRED the company wants to give agents and employees better tools so they can spend more time helping customers. Lemonade spokesperson Paul Staats told WIRED that AI will affect many professions and will threaten the positions of current employees in insurance and across the economy, that automation lets workers direct empathy, attention and experience to the hardest cases, and that the industry should take the Glassdoor report seriously. Some adjusters suspect they are training the systems that will eventually take their jobs.
The 98% figure is softer than it looks, and Glassdoor's own data says why: when employees sense layoffs coming, their reviews of AI get more negative, and skepticism deepens when a product they consider poor is pushed at them or their customers. Martin puts it plainly — people are looking at an absence of jobs and prospects. Some of that 98% is fear wearing the clothes of a technology review. What survives the discount is the operational complaint, which is separable from the job anxiety: misrouted claims, hallucinated summaries, and the fact that the error surfaces in front of a customer attached to a human name.
The number I would build the story around is not 98% or even 21%. It is the 50% collapse in entry-level postings. Automating first notice of loss removes precisely the work junior adjusters learned on, and the industry's own defense of the transition — that people will be freed up for the difficult cases — requires a continuous supply of people who got good at difficult cases by working through thousands of easy ones. Cutting the bottom rung is not a side effect of this deployment; it is the deployment. Nobody in the industry has explained where the senior adjusters of 2035 are supposed to come from.
The other thing nobody is discussing is who is accountable when the model is wrong. Sandy Avina, a former adjuster now consulting for the industry, says staff do not trust AI output much: a smudge in an attorney's document can produce a hallucination and a wrong payout, and so can a missed detail in a summary of a medical report. Customers generally have no idea that AI caused the confusion or the misinformation — they conclude the adjuster made the mistake. That is a liability question with an insurance company on one end and a named employee on the other, and it is absent from every vendor pitch and every corporate statement in this story.
Conrad understands why people see adjusters as soulless vultures circling someone else's catastrophe. His answer is that a good adjuster has to show empathy and fight to get the customer the largest payout available, and that he knows this from the other side. Twenty-five years ago, before he entered the industry, his house burned down and had to be written off entirely. Had someone told him then that a person would simply photograph the wreckage and AI would generate an estimate from the images, he says he could not have accepted it. What he needed at that moment was a human being who would check that he and his family were unharmed and safe. AI can assist, in his view, but using it as a substitute for human presence is a dangerous strategy.
That is the part the automation numbers cannot capture. A claim is a transaction on the insurer's books and the worst week of someone's life on the customer's, and the first contact — the moment Lemonade has automated at 96% — is where those two versions of the event meet.