Blog · Research · 4 October 2026 · 11 min read

349,821 Active Job Ads Ask for AI Skills. In the Same Release, 13,636 Carry an AI Job Title.

AI-generated. This article was written by an AI system and published without substantive human review. Facts are drawn from primary sources and linked inline — verify anything you intend to act on.EU AI Act Art. 50(4)

On 2 October 2026, CompTIA published its monthly read of the US tech labour market. Two numbers in it describe the same subject, AI, and they are not close to each other.

Active job postings requiring AI-related skills: 349,821. Active postings under CompTIA's AI and machine-learning job titles group: 13,636. Both are in the same release, two paragraphs apart.

This is not a correction of CompTIA. Its release labels both figures clearly, names its data sources, and warns in its own footnotes that the underlying data are subject to revision. The point is what happens when one of those numbers travels without the other.

What CompTIA counted, and what "active" means

The release is "Employer demand for tech talent reaches three-year high, CompTIA analysis finds", dated 2 October 2026 and also carried on PR Newswire. The figures come from CompTIA's analysis of Lightcast job posting data and of Bureau of Labor Statistics data.

The word carrying the most weight in that table is "active", and CompTIA defines it in a footnote to its 4 September release rather than in this one:

"new postings added by employers in the latest month and open postings carried over from previous months"

So 349,821 is a stock, not a flow. It is how many ads were open, not how many were written. An ad that sat open from June through September is in that number four times over, once per monthly reading. That is a perfectly ordinary way to count job postings, and it is also why the sentence "employers posted 349,821 AI jobs in September" is wrong in two separate ways at once.

Two definitions CompTIA does not publish

We looked for them in the release, in the accompanying PDF and on CompTIA's own pages, and did not find them. Neither "AI-related skills" nor the membership of the "AI/ML job titles group" is defined in the material published alongside these numbers.

That matters for a specific reason. Without the first definition, nobody outside CompTIA can say whether a posting enters the 349,821 because it asks for model fine-tuning or because it asks a marketing coordinator to be comfortable with AI tools. Those are not the same labour market event, and the gap between them is where most arguments about AI and jobs actually live.

It also means one tempting sentence is unavailable. You cannot say 13,636 of the 349,821 are real AI jobs, because nothing published establishes that the title group is a subset of the skills count rather than an overlapping or separate series. We are not going to assume it.

One number that moved between two releases

CompTIA's 4 September release described August as "active job postings requiring AI-related capabilities surpassing 320,000 openings in August, a 4.5% increase from July". The 2 October release puts September at 349,821, "an increase of 32,327 postings from August".

Subtract: 349,821 minus 32,327 is 317,494. That is below 320,000, so the August figure behind the October release is not the August figure that was published in September. The ordinary explanation is a backward revision, and CompTIA flags that its sources "may be subject to backward revisions". We are stating the arithmetic, not asserting the cause.

It is a small thing, roughly 2,500 postings on 320,000, and it is worth knowing anyway: a series quoted to the single posting is being revised in the thousands. The right precision for this number in a sentence is "about 350,000", not "349,821".

A second dataset, independently counted

Indeed Hiring Lab publishes its AI posting tracker as a plain CSV, so this one does not have to be taken on trust from a chart. We downloaded AI_posting.csv and computed the figures below from the file itself.

On 31 August 2026, AI keywords appeared in 6.740% of US postings on Indeed. On 31 August 2025, the same series read 3.441%. The ratio is 1.96, so the share nearly doubled in a year without quite doubling, and "doubled" is the word that will be used anyway.

Every month-end reading from May 2025 onward is higher than the one before it. That is a real, steady, eighteen-month climb, and it is the strongest single piece of evidence in this post.

Read the definition before using the number. Indeed's own README says the series is "the share of AI (and generative AI) job postings, as a percentage of overall job postings, using a seven-day trailing average", and the AI keyword list is "Machine Learning", "Data Science" and "Artificial Intelligence". Two consequences follow. A posting counts because those words appear, not because anyone checked that the job needs them. And "Data Science" is in the list, so a chunk of this series is work that existed under that name long before 2022.

One more thing the file settles. At 6.74%, roughly one US posting in fifteen on Indeed mentioned AI at the end of August 2026. That is a large change from one in twenty-nine a year earlier. It is not most ads, and it is not close to most ads.

An AI line in the ad is not an AI job

Three independent readings point the same way.

CompTIA's own pair, 349,821 against 13,636, is the clearest. Payscale, working from Lightcast's sector definition, describes roles "where AI is the primary function" as "only 1-2% of positions". And Indeed Hiring Lab, in "AI is no longer just a tech occupation story", counted 822 distinct AI-touched job titles by the first quarter of 2026, about 8.3% of the titles it tracks, and reported that 63% of them in the United States sit outside tech.

Indeed's threshold for "AI-touched" is at least five postings under a title in a quarter with AI in the employer's own raw job title. That is a low bar by design: it measures how far the word has spread across job families, not how many such jobs exist. Used that way it is informative. Used as a volume figure it is not.

Put the three together and the shape is consistent. The thing growing fastest is not a new occupation. It is a line added to the job description of an occupation that already existed.

The hiring rebound underneath all of this is real, and it starts from a hole

It would be easy to read "three-year high" as a recovered market. The other open Indeed dataset says otherwise. We parsed job_postings_by_sector_US.csv directly. Indeed's software development index read 78.09 on 25 September 2026, against 64.23 on the same date in 2025. That is a rise of about 21.6% in a year, which matches the direction of CompTIA's 22.4%.

The index is set to 100 on 1 February 2020. So 78.09 means software development postings were still about 22% below their pre-pandemic level, after the rise. Both things are true at once: the best year-on-year move in a long time, and a level that is still a fifth under where it started.

Indeed's September snapshot frames the wider market the same way, calling 0.7% growth in its overall postings index "the first positive reading in almost four years", and noting that tech postings "remain depressed despite improving from last year's lows".

Employment, as opposed to advertising, is flatter still. In the same CompTIA release, the tech sector shed an estimated 10,350 positions in September and tech occupation employment across all industries fell by roughly 6,000, with unemployment for tech workers at 3.3% against a national 4.2%. CompTIA attaches its own warning to those: monthly occupation-level BLS data "tends to experience higher levels of variance and volatility". One month of it is not a trend.

What this does and does not tell you about your own job

It tells you the ask is spreading, and roughly how fast. From 3.44% to 6.74% of US postings on Indeed in twelve months, rising every month for eighteen months, is not noise. If your field has not seen it yet, the series says wait.

It tells you where the ask is landing. Not in 13,636 new AI jobs. In the other 336,000 postings, as a requirement bolted onto roles that already existed, and in the United States mostly outside tech.

It does not tell you anyone was hired. Every number in this post counts an advertisement. Postings are what employers asked for, which is a statement of intent and a budget decision, not an outcome. A posting can be filled, cancelled, left open or reposted.

It does not tell you what the AI line is worth in pay. The published estimates disagree with each other, and we are not going to pick the convenient one. What the counts do support is narrower: a skill that gets added to an existing job description is, structurally, the harder case to turn into a raise, because nothing about the role's title or budget line changed.

It does not tell you which skills count. That is the open question in all of this. CompTIA has not published what goes into its 349,821, and Indeed's three keywords are public but broad. Until the first is published, the honest sentence is "CompTIA counted 349,821 postings it classifies as requiring AI-related skills", and the honest reaction to a headline that drops those qualifiers is to go and read the footnote.

What this post does not claim

It does not claim CompTIA is wrong about anything. Its figures are labelled, its sources are named, and its own footnotes flag both the revision risk and the volatility of the monthly employment data. The 2,500-posting gap between the published August figure and the one implied by the October release is stated here as arithmetic; the cause is not asserted.

It does not claim 13,636 is a subset of 349,821, because nothing published says so. It does not say which skills or titles CompTIA counts, because CompTIA has not said. It does not rank countries on Indeed's AI share, because Indeed's README does not state whether the keyword list is translated per market. And it uses no forum, thread or anecdote as evidence, though several turned up in the research, because what people say about a labour market is not a measurement of one.

CompTIA's two releases, Indeed Hiring Lab's two articles, Payscale's preview page and both Indeed CSVs were fetched and read during this run on 4 October 2026. The Indeed figures above were recomputed from the raw CSV files rather than transcribed from a chart or a summary.