The roles asking for AI are not the ones you would expect

This article is a submission by VA Masters, a global virtual assistant recruitment agency. It places rigorously vetted Filipino VAs through a custom six-stage hiring process, serving businesses across e-commerce, digital marketing, real estate, healthcare, financial services, and more.

Alon Pearl is the founder of VA Masters, which he started in 2021 after hiring his own first virtual assistant to keep pace with a growing company. He now leads a team of about 30 across recruitment, finance, analytics and development. His writing draws on the 166,350 applications VA Masters processed between January 2025 and July 2026.

What 322 virtual assistant job descriptions say about where employers want AI skills 

Key takeaways

• 23.9% of roles name AI in their requirements and 10.2% name a specific AI tool. Based on 322 roles opened February to August 2026.
• The roles asking for AI are the complex ones. Technical and analytics roles require it in 44.0% of cases. Finance and accounting roles require it in 5.3%. The gap between top and bottom is roughly eight to one.
• The smallest clients are not behind. Among clients employing ten people or fewer, 26.7% of roles require AI, above the overall figure.
• Workflow automation is asked for about as often as chat AI, and it is a different thing. Zapier and n8n together appear in more roles than ChatGPT does alone.

The assumption that runs backwards

The public conversation about AI and work starts from a familiar premise. Routine, repeatable jobs go first. Judgment work is safe for now.

We went looking for that pattern in 322 virtual assistant roles opened between February and August 2026, every one of them carrying a full description written for a real hire. What we found is close to the opposite.

Before going further, one thing needs stating clearly, because it changes what these numbers mean.

We are describing where employers ask for AI skills, not where AI is capable of doing the work. Those are two different questions and this data answers only the first.

How common is it

Across all 322 roles, 23.9% mention AI in some form. That covers artificial intelligence in general terms, large language models, prompting, generative tools and machine learning. A narrower 10.2% go further and name a specific product.

So roughly one role in four now carries an expectation that the person will work with AI, and one in ten tells you exactly which tool. This is not a forecast about where hiring is going. It is a description of what is already written into the job.

The client base behind these roles is almost entirely small businesses. For the same period, 62.7% of clients employ ten people or fewer and 91.0% employ fewer than fifty.

Among roles opened by clients with ten employees or fewer, 26.7% require AI, which sits above the 23.9% across the whole sample. The smallest businesses in our data are not asking for this less often than the larger ones.

One limit belongs next to that finding rather than buried at the end: an employee count is recorded for 97 of the 322 roles, so coverage is 30.1% and we cannot rule out that the roles without the field behave differently.

Which work requires it

Based on 322 roles opened February to August 2026. Categories with fewer than 15 roles are not reported.

This is the finding that runs against the assumption. The categories where AI appears most often are the ones that already require judgment. The categories built on defined procedure sit at the bottom.

Finance and accounting is the clearest case at 5.3%. It is the most procedural work in the list and the most rule-bound, and it is the least likely of any category to ask for AI. Sales sits second from bottom at 13.0%.

Administrative and operations support deserves its own line. It is by far the largest category of virtual assistant work by volume, and at 21.8% it sits below the overall average. The work most people picture when they imagine AI taking over office tasks is not the work employers are asking for AI skills in.

There is a reading of this that I think is right, though the data cannot prove it. Where a process is already defined, an employer knows what good looks like and does not need the person to bring a tool to it.

Where the work is open-ended, the employer is buying judgment, and asking for AI is a way of asking for someone who can cover more ground.

The tools employers name

Share of all 322 roles, February to August 2026. Jasper and Perplexity each appear below 0.5%.

The video and image generation tools at the bottom, Midjourney, HeyGen, Runway and Synthesia, are concentrated almost entirely in creative and social media roles. Everywhere else the named tool is a general assistant.

What stands out is how low every number is. Even ChatGPT, the most named product of all, appears in fewer than one role in twelve.

Most employers asking for AI are not asking for a particular product. They are asking for someone who can work with these tools generally, which is a meaningfully different requirement and a much easier one to hire against.

Automation is not AI, and we counted it separately

Zapier appears in 5.6% of roles and n8n in 2.5%. Together they show up in more roles than ChatGPT does on its own, which says something real about what small businesses expect a new hire to be able to operate.

Neither is counted as AI anywhere in this analysis. They are integration tools. A posting that asks for n8n with no AI context anywhere in it is not a posting that requires AI, and counting it as one would inflate the headline figure in a way we could not defend to anyone who read the underlying job descriptions.

That rule removed eight roles from the count and moved the headline down by more than two points. We are describing the decision rather than quietly applying it, because the boundary between automation and AI is exactly where this kind of statistic usually gets stretched. Our separate analysis of the software these roles name treats both groups on their own terms.

It is not rising, and we are not going to say it is

The obvious next question is whether this share is growing. Setting aside March, which is a particularly small month, every month in the period sits between 22.4% and 29.3%. There is no upward trend in the data and we do not present one.

That is worth saying plainly, because a rising line is what everyone expects from a statistic like this and it would be easy to produce one by starting the series earlier. We cannot.

Role descriptions only reach complete coverage in our system from February 2026, so any figure extending backwards would show a rise created by our own record keeping rather than by the market.

What does appear to be shifting is specificity rather than share. The proportion of roles naming a particular tool is higher at the end of the period than at the start. We hold that loosely: the earliest month rests on a single role, and seven months is not long enough to call a direction.

What we looked for and did not find

AI in the requirements is not a marker of a bigger engagement. Roles requiring AI are not opened at more hours than roles that do not.

AI requirements do not change how long a role takes to fill. These roles take no measurably longer or shorter than the rest.

We cannot answer the question for larger clients. The eleven to fifty and fifty-plus size bands fall below our reporting threshold once employee count coverage is applied. We can say the requirement is present in full among the smallest clients. We cannot say whether it rises or falls as clients get bigger.

We report these because a relationship that turns out not to exist is still a result, and because it shows which questions were tested rather than only which ones produced an answer.

About this data

This analysis describes what passes through VA Masters, not the offshore hiring market as a whole. Every dataset of this kind reflects the business that produced it. Our client base skews toward the companies that find us, and most arrive on their own through search, through AI assistants, and through referrals from existing clients. That shapes what appears here.

One bias is worth marking explicitly rather than leaving to a general disclaimer. Clients who reach us through search and through AI assistants are likely to be more comfortable with technology than the average small business, and that may lift these figures above what a random sample would produce. We cannot quantify it from our own data.

The sample still says something real. It covers hundreds of roles across eleven reported categories and clients on four continents, and the patterns are consistent rather than driven by a handful of accounts. But treat it as a detailed view of one large pipeline, not a measurement of national demand.

A role counts as requiring AI if its description or skills field mentions AI in general terms or names a specific AI tool, with whole-word matching so that ordinary words containing those letters were not captured. Roles opened by VA Masters for its own team are excluded throughout. Full methodology sits alongside our Filipino virtual assistant workforce dataset.