As of May 2026, 19.8% of US businesses reported using AI in any part of their operations. That figure comes from the Census Bureau’s Business Trends and Outlook Survey, which asks a nationally representative sample of firms every two weeks. Not a vendor survey. Not a consulting projection. The government asking businesses what they actually do.
Roughly four in five American businesses are not using AI at all.
I want to sit with that number for a moment, because it is very hard to reconcile with the articles you have been reading. Including, I should say, one on a site I run, which lists eight industries “completely transformed by AI” and is the reason I went looking for real data in the first place.
Why this one is here
This blog is about strategy, and strategy is mostly about seeing the situation accurately before you commit money to it.
Building AI systems is my day job. At Omnilogic Labs we design and run AI automation for banks and enterprises. Doing that work continuously, for clients who have to justify the spend internally, gives me a fairly unromantic view of where this technology actually is. I sell AI for a living and I am telling you the adoption numbers are far lower than the coverage suggests. That should probably carry some weight.
The gap between the headlines and the survey data is not a curiosity. It is a strategic opening, and I will come back to that at the end.
What “AI adoption” actually means here
Worth being precise, because this is where most statistics in this area fall apart.
The Census Bureau asks businesses whether they used AI in producing goods or services in the last two weeks. In November 2025 they widened the question from production-specific uses to AI used in any business function, which made the number go up and made it more honest.
So 19.8% is a generous measure. It counts a marketing team using a chatbot to draft copy. It counts one department running one tool. It does not require the company to have changed anything structural.
Even on that generous definition, four out of five say no.
Compare that to the framing in most coverage, where entire sectors have been “completely transformed.” Both things cannot be true.
Where adoption is genuinely high
Two sectors stand clearly above everyone else, and they have something in common.
Information: 39.7%. Software, media, telecoms, publishing. Roughly double the national rate.
Finance and insurance: 33.9%. Banks, insurers, asset managers. Also well above.
The pattern is not that these industries are more forward-thinking. It’s that their raw material is already digital. An insurer’s product is a document and a risk model. A software company’s product is text. There is no physical step to re-engineer, no sensor to install, no union to consult. The distance between “we should try this” and “it is running” is short.
That is the actual predictor of AI adoption, and it is more useful than any list of industries. Ask how much of your core work already exists as text, code, or structured data. That fraction is roughly the fraction of your business AI can touch this year.
Where it is much lower than you have been told
Retail trade sits around 14%, below the national average.
I want to flag that specifically, because retail and e-commerce appear near the top of nearly every “industries transformed by AI” list, including the one I am rewriting. Recommendation engines have been in the press for a decade. The survey says most retailers are not using AI in any business function.
Accommodation and food services is around 8%, the lowest of the major sectors. Also frequently described as being revolutionized.
Both make sense once you drop the assumption that press coverage tracks reality. These are low-margin businesses with physical operations, seasonal labor, and thin technical staff. The barrier is not skepticism about AI. It is that nobody in the building has time to implement anything.
What about healthcare, manufacturing, and the rest of the usual list?
The standard “industries transformed by AI” article names roughly the same eight every time: healthcare, financial services, retail, manufacturing, transportation and logistics, energy, legal, and agriculture. Two of those hold up against the survey data. Most do not, and the reason is consistent.
Legal and professional services hold up. Professional, scientific and technical services report adoption above 33%, in the same band as finance. That is entirely predictable once you accept the digital-raw-material thesis. Legal work is reading documents and producing documents. There is no physical process in the way.
Healthcare, manufacturing, energy, transportation and agriculture are physical-first. Their core work involves patients, machines, turbines, trucks and soil. AI can help around the edges of all of them, and there are real deployments worth reading about. But “completely transformed” is not a description of an industry where the central activity happens in the physical world and the regulatory approval cycle is measured in years.
I would put it this way. The genuinely exciting healthcare AI stories are about diagnostic imaging, which is a text-and-pixels problem sitting inside a physical industry. That is the pattern. AI is arriving in the document-shaped parts of physical industries, not in the industries as a whole. An article that names a sector rather than a function is making a claim it cannot support.
A note on what I can and cannot tell you here. The Census release breaks out some sectors by name and not others, so I am not going to give you a precise figure for healthcare or manufacturing, because I do not have one I can point you to. Most articles in this category would simply make one up. If you need those numbers, the BTOS data is public and you can cut it yourself.
The divide that matters more than industry
Sector turns out to be less predictive than size.
| Firm size | Using AI |
|---|---|
| 250+ employees | 37% |
| 100–249 employees | 32% |
| Fewer than 20 employees | Under 20% |
A large company is roughly twice as likely to be using AI as a small one, regardless of what business it is in. That is a resourcing story: large firms have people whose job includes evaluating tools, and budget that does not require the owner’s personal approval.
This is the single most important line in the data for most readers of this blog. If you run a business with fewer than 20 people, you are not behind your industry. You are exactly where four out of five of your peers are. What you are behind is the largest firms in your market, and that gap is widening.
Why the coverage and the data disagree so badly
Four reasons, and they are worth recognizing because you will keep encountering all of them.
Vendor surveys are not neutral. Most adoption statistics in circulation come from companies selling AI, surveying their own customers or an audience that opted into an AI newsletter. That population is not the economy.
“Transformation” has no threshold. A sector gets called transformed when a few visible companies do something interesting. Nobody defines what share of firms must be doing what before the word applies.
The exciting cases are unrepresentative. A hospital piloting diagnostic imaging AI is a genuinely good story. It is also one hospital, and the article rarely mentions what share of hospitals are doing anything comparable.
Articles cite articles. This is the mechanism that keeps stale claims alive. The retail example is instructive: the claim that retail is being reshaped by AI has been repeated so many times that it now requires effort to notice the survey data says otherwise.
What I would actually tell a client
Stop asking whether your industry is being transformed. Ask what fraction of your own work already exists as text or structured data, because that is what determines whether AI can help you this year. A law firm and a software company have very different reputations and very similar answers.
If you are small, the opportunity is unusually good right now. Four in five of your competitors are doing nothing. The tools are cheap and no longer require a data science team. This is the widest gap between “available” and “adopted” I have seen in twenty years of enterprise software, and gaps like this close.
If you are large, your risk is different. You are probably already using AI somewhere, without a clear picture of where, by whom, or on what data. Adoption in most large companies happened bottom-up and unmanaged. Finding out what is already running is usually more urgent than starting something new.
Be suspicious of any number without a source. Including the ones in this post, which is why every figure here links to the survey it came from. If a statistic about AI adoption has no citation you can click, treat it as marketing.
The part that matters
The most useful thing in this data is not which sectors are ahead. It is the size of the gap between what is possible and what is actually happening.
Nearly everything written about AI describes the frontier: the best model, the most advanced deployment, the company doing something remarkable. Meanwhile four out of five businesses have not started, and the most common reason is not fear or philosophy. It is that nobody had time.
That is a slower and less exciting story than transformation. It is also where essentially all of the remaining value is.
Sources
- US Census Bureau, “Large Firms With at Least 20 Employees Biggest AI Users” (May 2026), the source for every adoption figure in this post: census.gov
- Business Trends and Outlook Survey, the underlying instrument: census.gov/hfp/btos
- Census working paper, “The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks”: census.gov
- Federal Reserve, “Monitoring AI Adoption in the U.S. Economy” (April 2026): federalreserve.gov
The Census Bureau widened its AI question in November 2025 from production-specific use to any business function, so figures before and after that change are not directly comparable. All figures above are as of 3 May 2026.
A note on how this gets made: I run my own content pipeline, the same kind of system I build for clients. It drafts, I edit and fact-check every line, and I sign it. Claiming expertise in production AI and then hiding that I use it would be a strange way to make the point.
— Juan
