AI at Work: Why Low Turnover Doesn't Prove Employee Support
I think employers risk mistaking limited worker mobility for successful AI adoption. New Indeed and Glassdoor research raises an uncomfortable possibility: people notice the problems but cannot find an acceptable next job.

On November 18, Indeed and Glassdoor published a striking divergence: 21% of Glassdoor reviewers rated their career opportunities poorly in September 2026, up from 17% in April 2022. Over that interval, the US quits rate fell from 3.3% to 2.1%. [1]
I think this deserves attention from anyone judging workplace AI by how few employees leave. Staying can be a rational response to limited alternatives. Continued employment alone tells us little about whether people welcome the changes to their jobs.
The problem with the boiling-frog metaphor
The boiling-frog metaphor captures the danger of accepting gradual deterioration. But I object to its implication that people stay because they fail to notice what is happening. An employee can recognize worsening conditions, understand the consequences and still lack an acceptable next job.
The AI reviews suggest that at least some workers are paying close attention. Among current employees' Glassdoor reviews classified for AI sentiment in 2026 through September, 51% mentioned AI only in the cons section. [2] [3]
That is a selected group of reviewers, not a representative vote by American workers. The classification also depends on where AI appears in a review. It cannot measure the severity of every complaint or establish that an implementation failed. [3]
Nevertheless, I would take the complaints seriously. The researchers describe comments about inadequate tools and poorer experiences for employees and customers, alongside concerns about employment. These are workers' accounts, rather than independently audited results. [2]
My concern is the gap between recognizing a problem and being able to act. Telling someone to wake up to AI disruption offers little when they already understand it. For a worker who cannot secure another position, staying may be the most sensible available choice. That does not make the situation acceptable.
The evidence cannot identify who is trapped
The report estimates that monthly quits are roughly 349,000 below what the historical relationship with worker satisfaction would suggest. This is a modeled gap, not a count of people trapped in jobs by AI. [1]
Its separate analysis of search behavior tracks applications started within 90 days of an employee review. The methodology explicitly says it does not observe completed applications off Glassdoor, successful hiring or actual departures. [3]
Those limits matter. Starting an application shows activity; it cannot tell us whether someone received an offer, rejected an unsuitable position or abandoned the search. Nor can national quits tell us specifically how many employees stayed at companies introducing AI.
The dissatisfaction figures and AI complaints also cannot be combined into a single biography. These sources do not establish that the same individuals disliked workplace AI and were unable to leave, or that AI caused weak hiring.
I still see a serious warning. An employer claiming that retention proves approval would need evidence about employees' alternatives and reasons for staying. A low departure count cannot supply that explanation by itself. Uncertainty about the precise mechanism is no reason to treat continued attendance as an endorsement.
A quiet workforce can give management the wrong signal
Fortune reports that Glassdoor economist Daniel Zhao encounters bosses who take strong retention as proof they are doing well. [4]
I worry about a cycle in which weak hiring suppresses departures, low departures reassure management, and each additional workplace requirement becomes easier to impose. That is a plausible mechanism, not a causal finding established by this research.
Consider a hypothetical team required to use an AI tool that employees find unreliable. Everyone complies, nobody resigns, and management declares the transition successful. The same observed behavior could occur if workers liked the tool or if they disliked it but needed to keep their jobs. Retention cannot distinguish those explanations.
My preferred evaluation would ask employees whether they would choose the redesigned role again if a comparable alternative existed. It would also examine whether they can influence changing duties or move into another role within the organization. Answers would still describe perceptions and intentions, but they would address employee choice directly.
I would expect management to investigate specific complaints about the work, including complaints from people who continue using the system. Compliance with a requirement is weak evidence of enthusiasm for it.
Under a standard that recognizes dissatisfaction only when somebody resigns, the employee must bear the cost of leaving before the problem counts. I regard that as an unreasonable way to judge whether workplace change is benefiting people.
Watch for opportunities becoming usable
There are grounds for cautious optimism. The labor-market chapter reports low layoffs and September's first annual increase in Indeed's Job Posting Index since 2022. But the index can also rise when advertisements remain open longer. More active listings do not establish more completed hires. [1]
The AI chapter finds little evidence of widespread job losses caused solely by AI. It also reports that senior positions accounted for 71% of the recent increase in software-development postings. [2] That offers encouragement while leaving substantial uncertainty about who can benefit.
For someone considering a move, I would follow a small set of plausible employers over time, distinguishing fresh vacancies from listings that linger. Track whether applications progress toward actual hiring decisions. This cannot measure the whole market, but it provides evidence relevant to your circumstances instead of assuming that a national improvement has reached your experience level or location.
I would also periodically reconsider the reason for staying. A role chosen because it offers worthwhile development is different from one retained because applications elsewhere have gone nowhere. Both decisions can be rational. Naming the constraint makes it easier to recognize when circumstances change.
My conditional forecast is that stronger hiring would expose some dissatisfaction currently hidden by low departures. Employers could then lose people they assumed were content. If hiring remains weak, that reckoning could be delayed while employees continue accepting changes they dislike.
I would judge career security partly by whether people have usable alternatives and influence over their work. A workforce that stays because it has good reasons to stay is a much stronger sign of successful AI adoption.
The sources behind this perspective
- A Restless Labor Market Antsy to Move Again - Indeed Hiring Lab
hiringlab.indeed.com · Nov 18, 2026
- There is no AI Jobpocalypse, but plenty of AI-nxiety - Indeed Hiring Lab
hiringlab.indeed.com · Nov 18, 2026
- Indeed & Glassdoor’s 2027 U.S. Hiring and Worklife Trends Report: Consolidated Methodology - Indeed Hiring Lab
hiringlab.indeed.com · Nov 18, 2026
- Workers are skipping 349,000 quits a month. Indeed and Glassdoor say that could flip in 2027 | Fortune
fortune.com · Nov 18, 2026