CNBC and SurveyMonkey published their Q3 2026 AI and jobs survey on August 19. One result runs against the assumption most of the AI-anxiety coverage is built on.
Start with the usage split. Three in 10 US workers use AI daily, 21% multiple times a day and 8% once. Another 34% use it occasionally, a few times a week or less. And 37% never touch it at work.
Now sort perceived job security by that same axis.
Daily users are the only net-positive group: 27% say AI has made their job feel more secure against 22% who say less. Weekly users invert it, 10% against 25%. Workers who use AI once a month or less show the widest gap at 3% against 24%. Non-users are comparatively calm, 4% against 10%.
Asked about the job market overall rather than their own desk, the shape holds and the spread widens. Daily users split 35% optimistic to 33% pessimistic. Weekly users go 16% to 35%, a 19-point gap. Monthly-or-less users go 7% to 52%, a 45-point gap. Non-users go 5% to 31%.
Anxiety doesn’t rise with exposure
If AI dread came from seeing what the tools can do, the curve would climb the whole way: more use, more fear. It doesn’t.
The floor sits in the middle. Occasional users have watched AI finish an email and compress a 30-page report into 600 words. They have not used it enough to make it part of their own output. They got the evidence of the threat without the leverage.
Daily users got the other half. Among all AI users, 68% say it has saved them time, rising to 84% for daily users. On skill confidence, 56% overall and 73% for daily users. On critical thinking, 45% overall and 60% for daily users.
They also have not turned into a dependency case. Only 6% of AI users say they go to AI first when a task lands. Most, 58%, try to work it out themselves before asking, and 30% don’t use AI for problem-solving at all.
That kills half of a popular claim. “The heavy users are the ones who should be worried about replacement” is not in this data. The heavy users are the only cohort that feels it has its footing.
One number hints at where that footing comes from. Nearly a quarter of AI users, 23%, admit they have redirected a colleague to ask AI rather than answering themselves, and among daily users it is 34%. The reasons they give, in order: pushing colleagues toward self-sufficiency (50%), saving time and avoiding interruptions (44%), AI being good enough (34%).
If your standing at work rests on being the person everyone asks, that channel is closing. Management isn’t closing it. Your coworkers are.
One in five is looking at the trades
The hardest number in the survey isn’t a sentiment reading. It’s a movement.
One in five US workers, 20%, is considering a switch into a trade job or manual labor because of AI. Unpacked: 13% think about it sometimes, 4% are actively considering or planning it, and 3% say they are currently in the process of switching.
That last 3% is not an attitude. It’s migration that has already started.
Widen the frame and 23% have changed (10%) or considered changing (13%) their target industry, rising to 31% among students and workers still in school. On skills, 32% have changed or reconsidered what to develop, 45% for those in school. Another 24% changed the jobs they apply for, and 24% changed the companies.
Where are they going? Technology, media and telecom takes 23%. Healthcare, education and social assistance takes 22%. Construction, skilled trades and utilities takes 12%. Finance, real estate and business services 8%. Manufacturing, mining and agriculture 7%. Retail, hospitality and personal services 7%. Transportation, logistics and warehousing 7%.
Set that destination list against the postings data and the mismatch shows. AI roles are 31% of US tech job postings while non-AI postings fell 60%. The 23% heading for technology are walking into an industry whose internal composition has already been swapped out, not the one they remember.
Pulling up the ladder and banning the climb
One more finding deserves its own line: 35% of workers think junior-level employees should be prohibited from using AI entirely. Another 42% would allow it with clear guidelines, 14% only under supervision and approval, and 9% would let juniors use it freely.
Read that beside two other numbers. Stanford’s August tracking put employment for 22-to-25-year-olds in highly AI-exposed occupations 19% below their less-exposed peers, and entry-level postings are down 13% since ChatGPT.
So the junior rung is disappearing, and a third of the workforce wants juniors barred from the tool. The same survey says daily use is the only usage pattern associated with feeling more secure. Put those together and the position amounts to locking new entrants out of the one channel the data shows works.
Worth noting alongside it: 55% of companies have no official AI policy at all, 34% make it optional, 6% require it, and 5% ban it outright. The argument about whether juniors may use AI has no written answer in most workplaces. Everyone is guessing.
What this means for jobs
Occasional use is the worst place to stand. The data says so plainly. That tier collects neither the productivity gain nor the calm, and runs roughly twice as pessimistic as people who never touch AI. Either move it into a daily workflow or accept that it functions as a daily reminder that something is gaining on you.
There’s a concrete test for whether your value is being hollowed out. Over the past six months, have colleagues stopped bringing you questions they used to bring? If yes, sort those questions. Knowledge that can be written down and retrieved is repricing fast. Answers that require looking at the site, the person, or the context are not.
Treat the trades path as a real option with real costs. With 20% considering and 3% mid-switch, this is migration rather than online mood. But the barriers are licensing, apprenticeship length, and a physical career arc. The cost is time, not nerve. Before anything else, price the entry path for the specific trade: how long the apprenticeship runs, which credential, and how deep the local shortage actually is.
If you manage people, that 35% is a warning, not a consensus. Banning junior staff from AI protects one thing in the short run, which is that they learn to think without it. What it gives up over the long run is the only source of security this survey found. Guidelines beat prohibitions: name which tasks must be attempted unaided first, which outputs get checked line by line, and where a human judgment has to be signed.
