What Companies Use AI For: Three Functions, and Not Much Else
Almost every figure you have read about what companies use AI for came from a survey run by a company selling AI. A few hundred executives, a vendor’s mailing list, and a result that happens to support buying more of something.
There is a better source, and it is free. The U.S. Census Bureau runs a survey covering roughly 1.2 million businesses, collected every two weeks, and in 2026 its researchers published a working paper breaking the answers down by business function and by worker task.
The picture it produces is narrower than the discourse. Adoption is not only uncommon — among the firms that have adopted, it is shallow, and it clusters in three places.
The Survey Behind the Numbers
Worth establishing why this source is different before using it.
The Business Trends and Outlook Survey is run by the Census Bureau. Its sample is approximately 1.2 million businesses, split into six panels of around 200,000, with each business reporting once every twelve weeks. Results come out every two weeks and cover all employer businesses in the U.S. economy except farms, broken out by sector, state, metro area and employment size.
No vendor list. No incentive to produce a particular answer. And free to read.
One honest caveat about comparability: the Census revised the AI question’s wording in November 2025 to cover use in any business function rather than only production. Figures published before and after that change measure different things, which is part of why older “AI adoption” numbers circulate wildly out of step with newer ones.
What Companies Use AI For, by Function
The working paper on AI diffusion breaks adopters down by where inside the business the technology actually sits.
| Business function | Share of adopting firms using AI there |
|---|---|
| Sales and Marketing | 52% |
| Strategy and Business Development | 45% |
| IT | 41% |
Three functions, and a steep drop after them.
The concentration is not an accident of how the question was asked. It reflects where the output of a language model — text, summaries, drafts, analysis of documents — maps most directly onto what the function already produces.
Sales and marketing is a function whose deliverable is largely language. What the people running it actually do, according to the BLS, is plan campaigns, decide which media to advertise in, commission and analyse market research, and advise clients. Several of those are document work. That is why it leads.
The Finding That Matters More Than the Headline
Here is the part almost nobody reports, and it reframes everything above.
Among firms that have adopted AI, 57% use it in three or fewer business functions. And at the level of individual work, 65% of firms limit use to three or fewer tasks.
Adoption is not a wave rolling through the organisation. In most adopting companies it occupies a corner.
That matters because the public conversation treats adoption as binary — a company either uses AI or does not — when the data shows a company can be counted as an adopter on the strength of one team using it for one thing. The gap between “we use AI” and “AI changed how we work” is wide, and this is the first dataset I have seen that measures it.
18% of Firms, 32% of Employment
The headline adoption figure is 18% of firms. On an employment-weighted basis it is 32%, with the paper projecting 22% within six months.
Read those two numbers together. Fewer than a fifth of businesses use AI, but nearly a third of employed people work somewhere that does. Adoption is concentrated in large employers, which means the typical worker is far more likely to encounter it than the typical business is to have adopted it.
At the extreme the concentration is sharper still: the paper reports adoption reaching 50–60% among very large firms in Information, Professional Services and Finance, and 60–70% on an employment-weighted basis within those groups.
So “most businesses don’t use AI” and “most people work near it” are both true, and which one you quote depends on what you are trying to argue.
The Sector Gap Is Enormous
Census figures for May 2026 show how unevenly this is distributed:
| Sector | Firms using AI |
|---|---|
| Information | 39.7% |
| Finance and Insurance | 33.9% |
| Retail Trade | ~14% |
A business in Information is almost three times as likely to use AI as one in Retail. Advice written for one of those worlds is close to useless in the other, which is a reasonable explanation for why so much of it feels wrong wherever you happen to be standing.
What Workers Actually Do With It
Separate from business functions, the paper measures whether workers use AI in their own tasks: 23% of firms, or 41% employment-weighted.
The leading uses are specific and unglamorous: writing, document analysis, and information search.
Not strategy. Not creative direction. Not autonomous agents running operations. Three categories of text work — producing it, reading it, and finding it.
That maps closely onto what these systems are actually good at, and it is a useful corrective to a market full of claims about transformation. It also narrows what someone needs to be good at to be useful here, which connects to how long competence in a given occupation actually takes to build — a shorter list of tasks is a shorter learning curve.
The Employment Finding, Stated Carefully
The paper reports that 66% of users rely on AI solely to augment tasks, and that AI-related employment decreases are rare, occurring in only 2% of firms.
Two percent is a small number and it deserves to be stated carefully rather than waved as proof of anything. It is self-reported, it covers a specific window, and absence of reported decreases is not the same as absence of effect — hiring that quietly does not happen leaves no trace in a question about decreases.
What it does support is narrower: among firms using AI today, deliberate headcount reduction attributed to it is uncommon, and augmentation is the dominant stated mode. That is consistent with what the measured employment data has shown elsewhere, and inconsistent with the version of this story that gets headlines.
How a Company Integrates Changes the Outcome
One more finding, and it is the most interesting in the paper.
There is a positive association between the breadth of AI integration and commercial performance. But breadth across functions is linked to employment reductions, while integration at the task level is not.
In plain terms: companies that spread AI across many departments show a different employment pattern from companies that go deep on specific tasks within existing work. Same technology, different organisational decision, different consequence for the people there.
The paper reports an association, not a cause, and it should be read that way. But it points at something the “will AI take jobs” framing misses entirely — that the answer may depend less on the capability of the tool than on how a particular management team chose to deploy it.
What This Means If You Sell or Work
Four things follow from the numbers rather than from speculation.
The demand is concentrated, so aim at it. Sales and marketing, strategy, and IT account for the bulk of functional adoption. Anyone offering services around this is better off being specific to one of those three than generic across all of them — and the categories of AI-assisted work people already pay for line up closely with where the adoption actually sits.
Shallow adoption is an opening, not a closed door. With 57% of adopters using AI in three or fewer functions and 65% limiting it to three or fewer tasks, the companies that already said yes have barely started. Expanding an existing deployment is a different, easier sale than converting a non-adopter.
Know which sector you are talking to. A 39.7% Information firm and a 14% Retail firm are not at the same point in the same conversation, and pitching them identically wastes both.
If you are small, you are the norm. The overwhelming majority of U.S. businesses have no employees at all, and the one-person operation is the default structure rather than the exception. Adoption statistics drawn from employer firms describe a world most independent operators are not in.
What the Data Does Not Say
Three limits worth naming.
It is self-reported. Firms answer a survey about their own behaviour. That is the best instrument available at this scale, and it is still what companies say rather than what they do.
It is a snapshot, and the question changed. The November 2025 revision means older figures are not comparable, and adoption is moving — the paper itself projects 22% within six months.
It covers employer businesses. Farms are excluded, and so are the tens of millions of U.S. businesses with no employees. Nothing here describes solo operators.
Frequently Asked Questions
What do companies use AI for most? Sales and marketing leads at 52% of adopting firms, followed by strategy and business development at 45% and IT at 41%. At the individual worker level, the leading uses are writing, document analysis and information search.
How many companies actually use AI? 18% of firms in the Census working paper’s data, rising to 32% on an employment-weighted basis, with 22% expected within six months. The gap between those two numbers reflects that adoption is concentrated among large employers.
Is AI causing job losses? This dataset says AI-related employment decreases occurred in 2% of firms, with 66% of users relying on it solely to augment tasks. That is self-reported and covers a specific window, so it is evidence rather than proof — but it does not support the mass-displacement version of the story.
Which industries use AI the most? Information at 39.7% and Finance and Insurance at 33.9%, against roughly 14% in Retail Trade. Among very large firms in Information, Professional Services and Finance, adoption reaches 50–60%.
If a company “uses AI”, how much does it really use it? Often very little. 57% of adopting firms use it in three or fewer business functions and 65% limit worker use to three or fewer tasks. Being counted as an adopter can mean one team using it for one thing.
Why do other surveys report much higher adoption? Many are run by vendors with small samples and a commercial interest in the result. This survey covers about 1.2 million businesses and is run by the Census Bureau. Question wording also matters — the Census broadened its AI question in November 2025, so figures across that boundary are not comparable.
Does using AI more broadly make a company perform better? The paper reports a positive association between integration breadth and commercial performance. It also finds that breadth across functions is linked to employment reductions while task-level integration is not. Both are associations, not demonstrated causes.
Does any of this apply to a one-person business? Not directly. The survey covers employer businesses, which excludes the large majority of U.S. firms that have no employees at all.
Conclusion
The honest answer to what companies use AI for is narrower than almost anything written about it. Three functions carry most of the adoption — sales and marketing, strategy, and IT — and three categories of text work carry most of the individual use.
The more useful finding is about depth rather than breadth. Most adopting firms use AI in three or fewer functions and limit workers to three or fewer tasks. The line between “uses AI” and “has been changed by AI” is not one most companies have crossed, and treating adoption as a binary hides that entirely.
And the structural point underneath: 18% of firms, 32% of employment. Whether this technology feels ubiquitous or absent depends almost entirely on how large your employer is and which sector it sits in — which is why the same article about AI at work reads as obvious to one person and as fiction to another.
None of that is a forecast. It is a map of where the thing currently is, built from the largest business survey available, and it costs nothing to read.
A Note from Lucas
I went looking for this paper because I was tired of citing adoption statistics that came from companies selling the thing being adopted, and the difference in sample size alone — 1.2 million businesses against the few hundred executives most surveys use — made the search worth it. The number that genuinely surprised me was not 18% or 52% but the shallowness: most companies that adopted AI use it in three functions or fewer, which means the gap between announcing adoption and actually changing how work happens is enormous. I had assumed the opposite, that firms which started would spread it quickly. My recommendation if you sell anything in this space is to stop trying to convert the holdouts and look at the companies that already started and barely use it — the data says that is where the room is.
About the Author
Lucas Palma is the founder and editor of Info Is The Key. He writes about online business, entrepreneurship, and the practical mechanics of building income independently, working from primary sources and official data rather than recycled advice.
Last Updated: October 2026 | Reading Time: 10 minutes
Editorial note: This article summarises U.S. Census Bureau survey data and a Census working paper as accessed in October 2026. Survey responses are self-reported, the AI question’s wording was revised in November 2025 so figures across that date are not directly comparable, and the survey covers employer businesses excluding farms. Associations reported in the working paper are not demonstrated causal effects. Figures are United States data and will differ elsewhere.
Sources
- U.S. Census Bureau — The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks (CES-WP-26-25)
- U.S. Census Bureau — Business Trends and Outlook Survey
- U.S. Census Bureau — AI use among businesses, by sector and firm size
- U.S. Bureau of Labor Statistics — Advertising, Promotions, and Marketing Managers
