A Glassdoor analysis covered by WIRED drops a bombshell: 98% of insurance claims adjusters' comments about AI are negative. Ninety-eight percent. The figure is staggering enough to read as a technology failure. It isn't. AI implementation in companies fails when the tool arrives before the why, and that 98% says more about those who impose the tool than about the tool itself.
TL;DR: The no-nonsense summary
- 98% negative: insurance adjusters hate AI because it generates more work than it removes, not because the technology fails.
- The Pentagon, flipped: ChatGPT and Grok for 1.7 million employees. Same AI, bounded implementation with a clear narrative. Opposite result.
- Inverted U-curve: too little AI frustrates, too much AI scares. The sweet spot is moderate, chosen adoption, not imposed.
- Key stat: +11.5% productivity in companies using AI for a year, but -4% net headcount. The gain is real. How you distribute it decides everything.
What Is Behind the 98% AI Rejection Rate?
The data comes from the insurance claims adjustment sector in the US. Adjusters aren't complaining that AI exists. They're complaining that tools have been dumped on them that misclassify claims, generate "hallucinated" summaries, and force them to spend more time correcting errors than doing their actual job. In short, AI MULTIPLIES their workload.
And it's not just perception. Entry-level job postings in the sector have fallen 50% since 2025. Adjuster employment dropped 21% between May 2025 and May 2026. The equation workers see is straightforward: a broken tool gets forced on them without anyone asking, and on top of that, fewer people are being hired.
Who wouldn't be furious?
On social media the conversation is unanimous. @jorgesabella1 nails it in a recent video with over 1,200 views: "They push you to use it more and more, but you sign off on the mistake. You can't blame the AI, and you can't fire it either." That line captures the problem better than any consulting report.
The Pentagon Integrates ChatGPT and Grok: Same AI, Different Outcome
While adjusters are cursing their tools, the Pentagon has just rolled out AI to 1.7 million people without stirring up resentment. Its platform GenAI.mil offers customized versions of ChatGPT and Grok for unclassified tasks: administration, logistics, planning, policy drafting.
Here the technology is beside the point. What changes everything is the perimeter and the narrative. The Pentagon bounded usage to specific tasks, adapted the tools to its security requirements, and presents it as a productivity improvement. Nobody was told "use this or you're out." They were told "this takes the tedious part off your plate."
Is it a perfect deployment? From the outside it's impossible to say. But the contrast with the insurance sector is devastating: same technology, opposite result. The variable that changes is how you put it in front of people. The language model is irrelevant.
Are You Implementing AI for Your Business or for the Slide Deck?
This is the question that separates a useful AI adoption from a million-dollar failure. According to a Morgan Stanley survey, companies that have been using AI for at least a year report an 11.5% productivity increase, but also a net 4% reduction in headcount. Both figures go together. If you only look at the first, you're looking at the slide deck. If you look at both, you're looking at the business.
A study published by the NIH confirms it: when a company has the radical idea of asking the team before changing their tools, satisfaction and performance go up. Turns out people like being asked. Surprising. And while 75% of organizations report growing AI adoption, only 44% have clear guidelines for their teams. That 31% gap has a name: anxiety.
I'd wager most failed implementations share a pattern: the person who decides to bring in AI isn't the one who's going to use it. And the person who's going to use it finds out when their tool changes on a Tuesday morning. In our experience automating content with AI, the leap from "this is a threat" to "this saves me three hours" happens when the team chooses the tool and defines how to use it. Impose it, and they'll push back.
The Inverted U-Curve Executives Ignore
The relationship between how much AI you introduce and how happy your team is isn't linear. It follows an inverted U-curve: too little AI frustrates because the tools aren't there. Too much AI scares because it smells like replacement. The sweet spot is moderate adoption with the team on board, not the "all-in on AI" that looks so good in the CEO's presentation.
In Spain, 61% of companies have already implemented AI solutions, 7 points above the European average. 81% report productivity gains. But the missing figure is how many actually asked their teams whether the tool was working for them. I'd wager that number drops considerably. And that's where your own 98% starts brewing.
The Arrogance of Imposing AI Without Asking
@josevizner explains it on TikTok with over 112,000 views and 5,600 likes: "The worker the employer lays off is also their customer. When they run out of salary, they stop spending." It's not a new argument. But applied to AI adoption it carries a twist few see: you don't just lose customers. You lose the process knowledge no model will replicate without quality data. And that data is generated by the people you're letting go.
That 98% has a clear culprit: leadership. Those who blow up an entire team's trust. If as an executive you choose the AI model that scales best but ignore the team that has to use it, what you have is a slide deck with an expiration date. And the one who pays the price is the person signing off on errors every morning, not the one who signed the purchase order.
