Artificial intelligence was supposed to make insurance claims handling more efficient. Instead, we may be discovering one of its unintended consequences: AI can make it incredibly easy for everybody involved in a claim to create more work for everybody else.
A fascinating WIRED article, 1 provided to me by San Francisco attorney Dan Veroff, reports that insurance claims adjusters may be among workers most unhappy with artificial intelligence. According to research from Glassdoor, 98 percent of claims adjuster reviews that mentioned AI were negative. Adjusters complained about AI making mistakes, misclassifying claims, generating inaccurate summaries, and forcing humans to clean up the mess afterward.
One adjuster quoted by WIRED stated that AI was “implementing more work onto the adjusters.” Another claims executive described what he called “AI fatigue.”
Independent and company adjusters are telling me, following speeches and in comments to LinkedIn posts, that some policyholders, public adjusters, contractors, and attorneys are feeding every disagreement into artificial intelligence and then sending back enormous emails, demands, arguments, document requests, legal theories, and follow-up questions. What previously might have been a thoughtful three-paragraph letter can now become twelve pages in about thirty seconds.
The person sending it may have spent five minutes creating it. The adjuster receiving it may have to spend two hours figuring out whether there is one important issue buried on page nine.
There is already a term for this phenomenon in the workplace: “workslop.” Researchers associated with the Stanford Social Media Lab and BetterUp have studied AI-generated work that appears polished but actually shifts the burden of thinking, checking, correcting, and completing the work onto the person receiving it. Their research found that 40 percent of surveyed workers reported receiving this type of AI-generated work, and that dealing with each incident consumed nearly two hours on average. Managers reported receiving it even more frequently than individual contributors.
So, are we now creating “claimslop”?
In a Reddit online forum for insurance adjusters, one adjuster recently complained about receiving extraordinarily long AI-written emails from a claimant that contained contradictions and incorrect information about state claim-handling requirements. Another adjuster described receiving approximately twenty AI-generated emails regarding a single claim. Another adjuster described receiving an obviously AI-generated response to a coverage denial filled with legal terminology and cases that did not even apply in the state where the claim arose.
The legal profession is seeing it too. The Financial Times recently reported that lawyers are receiving increasing amounts of AI-generated correspondence, legal theories, questions, and proposed strategies from clients. I see it in my own practice as well. The problem is that somebody still has to read all of it, determine what is right, separate the useful from the nonsense, and respond appropriately. Instead of reducing professional work, AI can sometimes multiply it.
If artificial intelligence finds an overlooked policy provision, compares hundreds of Xactimate line items, identifies an inconsistency in the carrier’s position, organizes thousands of pages of documents, or helps a public adjuster explain a complicated coverage or construction issue more clearly, that is exactly how the technology should be used.
But AI should make an argument better. It should not merely make it longer.
Before AI, writing a fifteen-page letter required someone to spend hours crafting it. People generally focused on their strongest points because researching and writing the weaker ones took time. Artificial intelligence has largely eliminated that economic restraint.
Now somebody can ask an AI program to “identify every possible argument,” “make this more comprehensive,” “cite every applicable regulation,” and “prepare detailed follow-up questions,” and suddenly an adjuster has a miniature law review article sitting in the inbox.
The computer generating it does not care whether anybody has time to read it. The recipient does.
I do not think an insurance adjuster can safely assume that page eleven is nonsense simply because pages one through ten were nonsense. There may be a legitimate coverage issue, overlooked damage, factual mistake, or statutory obligation buried somewhere inside. So, somebody still has to read it.
Insurance companies should not become too sanctimonious about this problem. They are aggressively deploying artificial intelligence themselves. The WIRED and Glassdoor reports demonstrate that adjusters are complaining that management is forcing AI into their workflows, sometimes creating mistakes that those same adjusters must correct.
Imagine where this can lead. The policyholder uses AI to write the demand. The public adjuster uses AI to expand the demand. The carrier uses AI to summarize the demand. The adjuster checks the AI summary. The carrier’s AI helps draft the response. The policyholder feeds that response back into AI, which discovers eighteen new “issues.” Those eighteen issues go back to the adjuster, whose AI summarizes them again.
Everybody becomes more paper-productive, and nobody seems to have enough time to adjust the claim.
I am an enthusiastic user of artificial intelligence, and I think insurance professionals who refuse to learn how to use it will eventually place themselves at a significant disadvantage. But using AI well requires human judgment.
The best public adjusters, attorneys, contractors, and policyholders should use AI to identify what matters and then communicate it clearly and precisely. They should not use artificial intelligence as an infinite correspondence machine.
Artificial intelligence can help us think better, investigate faster, and communicate more clearly. But if we use it simply to manufacture unlimited words for somebody else to deal with, we have not eliminated busywork. We have automated its production.
Thought For The Day
“Tools are made by humans, are being deployed by humans, and should be used properly by humans.”
—Fei-Fei Li, Stanford Institute for Human-Centered Artificial Intelligence
1 Kate Taylor. “You Know Who Really Hates AI? Insurance Claims Adjusters.” Wired (Aug. 31, 2026). Available online at https://www.wired.com/story/insurance-claims-adjusters-really-hate-ai/
