In September 2025, 3.6% of US job postings on Indeed mentioned AI. By September 2026 it was 7.2%. Over the same twelve months, workers who use AI regularly on the job said it saved them a little less time each day than it had the year before.
I think the doubling in AI job ads says more about what employers hope for than what workers feel, and that gap is where the real management work sits. Both numbers come from the same research team, published on the same day, and read together they describe a hiring market that has moved faster than the work itself.
What Indeed and Glassdoor found
On 18 November, Indeed Hiring Lab and Glassdoor Economic Research published "There is no AI Jobpocalypse, but plenty of AI-nxiety", part of their 2027 US Hiring and Worklife Trends report. The headline finding on demand is plain. Indeed's AI Tracker, which counts the share of postings that contain AI terms, rose from 3.6% in September 2025 to 7.2% in September 2026, and the share of companies with at least one AI-mentioning posting was up 44% over the same period. Almost two-thirds of the job titles touched by AI are now outside traditional tech, so this is no longer a story about software teams alone.
The worker side reads differently. Indeed's Workforce Insights Survey found regular AI use at work rising from 36% to 42%, so more people are using the tools. Yet 56% of employed US respondents did not agree with the statement "AI makes or could make me more efficient". Among regular users, the estimated daily time saved by the tools fell from 2.3 to 2.1 hours. That is a small drop, and self-estimates of time saved are rough, but the direction matters: more use has not brought more felt benefit.
Glassdoor's review data shows the mood shifting over a longer stretch. In 2019, 82% of reviews that mentioned AI framed it positively; in 2026 (through 30 September) only 45% did, and a slim majority, 51%, mentioned AI only in the "cons" section. Reviewers who were critical of AI were 3.5 times as likely as the base rate to mention burnout, 3.8 times as likely to mention job insecurity and 7.1 times as likely to mention layoffs. Their approval of the CEO sat at 37%, against 89% among reviewers who were positive about AI. The authors say these gaps hold even after accounting for the review's overall star rating.
A companion post looks at who is being hired. By September 2026, senior-level postings were 19.2% above their January 2025 level, while entry-level and mid-level postings were each down around 4%. Using Indeed profiles, the team estimates that 50.4% of the class of 2025 were underemployed a year after graduating, against a rate that hovered just under 40% for most of the past decade. (Underemployed here means working in a job where 20% or fewer of that occupation's Indeed listings ask for a bachelor's degree.)
What the study does not show
This is useful work, but it is company research and should be read as such. Indeed and Glassdoor are both owned by Recruit Holdings, which describes them together as its HR Technology business, so this is one parent company's data, not two independent sources agreeing. It is not peer reviewed.
The AI Tracker measures the words employers put in job ads. It does not measure how much AI those employers use, and it says nothing direct about jobs replaced. Glassdoor reviewers choose to post, so the people writing about AI in the "cons" section are not a random sample of workers. The 2026 review figures cover only part of the year. The post page does not give the Workforce Insights Survey's sample size or field dates, and it does not define "regular users" beyond the weekly threshold it uses elsewhere.
Most of all, the links are correlations. Critical reviewers mention layoffs far more often, but that could mean layoffs make people resent AI as easily as the other way round. The authors also report that worker sentiment has no strong relationship with how exposed an occupation is to AI, which suggests the souring is about how AI arrives at a company more than what the technology can do in that job. On the entry-level squeeze, the companion post is careful: AI has been cited as a possible cause, but in its words, "while evidence for this is still mixed, entry-level workers are nevertheless visibly frustrated with AI."
The headline about no "jobpocalypse" holds, at least for now. The authors cite Bureau of Labor Statistics figures showing labour productivity growing at a 2.1% annual rate since late 2019, up from 1.5% over 2008 to 2019 and in line with the long-run average, and they say the data "do not yet show a dramatic shift."
The bigger picture
A survey released the same day points the same way. Marsh's 2026 HR Technology's Impact on the Workforce study, a survey of more than 6,700 employees in 10 countries fielded in July 2026, found that 71% of workers have used AI in their jobs (up from 39% in 2024) but only 37% use it regularly. The training is concentrated at the top: 78% of executives received AI training in the past year, against 44% of experienced professionals, 32% of entry-level staff and 20% of hourly workers. About one in three employees say nobody has discussed AI's impact on their role, and among those who heard something, only 21% heard it from their direct manager. Marsh sells HR consulting, so it has an interest in this finding, but the pattern matches Indeed's.
Put the two together and the shape is fairly clear. Employers are writing AI into job ads and buying tools. Workers are trying them, often without training, often without anyone explaining what changes in their job, and they report modest and slightly shrinking returns. We have covered the same gap from the spending side, where companies are spending far more on AI tools and only 13% of CIOs say AI has paid off. Adoption has spread widely, but the redesign of work behind it has not kept pace.
Why the gap is a management problem
An honest concession first: some of the drop in time saved may be measurement. As tools become routine, people stop noticing the minutes they save, and new users may be less enthusiastic than early adopters. That could explain part of the 2.3-to-2.1 slide.
Even so, I'm not convinced the explanation lies mainly in the models. When a team adds a tool without changing who does what, the saved minutes leak into checking, rewriting and extra meetings about the tool. The Glassdoor finding that critical reviewers talk about burnout and insecurity fits that reading: people are being asked to do the old job plus the new tool, with layoffs in the background. A job ad that says "AI" costs nothing to write. Deciding which tasks leave the role, who checks the output and what the person does with the time takes more thought, and most employers seem to have skipped it.
The hiring numbers add a longer-term risk. If senior postings keep rising while entry-level ones fall, teams are betting that nobody needs to learn the work from the bottom. That may save money this year, but it leaves you without people who will be senior in five years.
What to do if you build or hire
If you run a small team or build AI features, these are the steps I'd take, in this order.
Measure time saved on real tasks, not by survey. Pick three recurring tasks (a weekly report, a support reply, a proposal draft). Time them for two weeks without AI and two weeks with it, including the time spent checking and fixing output. My rule of thumb (a judgement, not a figure from the research): if the net saving is under about 15 minutes per task, the tool is not yet earning its place there.
Write down what leaves the role. For every AI tool you introduce, list the tasks the person stops doing and what they do with the time instead. If you cannot fill in the second column, expect the saving to vanish into busywork.
Train the people doing the repetitive work first. Marsh's figures show training going mostly to executives. The hours are saved at the front line, so that is where I'd spend the first training budget: I'd expect an hour of hands-on practice on the person's own tasks to do more than a general course.
Have the manager explain the change. One conversation per person covering what changes in the job, what does not, and whether roles are at risk. Going by the Glassdoor data, people left to guess tend to assume the worst.
Read AI-in-job-ad numbers as intent. When you see that AI mentions doubled, ask what the employer actually changed. In your own ads, describe the workflow ("you'll review AI-drafted replies and handle escalations") rather than adding "AI" to the title.
Keep one junior seat. If you hire this year, keep at least one entry-level role and pair it with a senior person. Someone has to learn the work, and AI-drafted output needs people who know what good looks like.
If you build AI products, the same data is a brief for your onboarding: show users the time saved on their own tasks inside the product, and make the redesign of the workflow part of setup. We have written before about how AI saves more time when the people doing the work help redesign it, and this report adds weight to that.
The doubling in AI job ads tells you what employers want from AI, and the slightly shrinking time savings tell you what they have done so far to get it. The work between the two is mostly decisions about tasks, training and conversations, and a small team can make those decisions this month.
Sources
There is no AI Jobpocalypse, but plenty of AI-nxiety, Indeed Hiring Lab and Glassdoor Economic Research
Experience Wanted. But Where Do You Get It?, Indeed Hiring Lab and Glassdoor Economic Research
Business overview, Recruit Holdings

