One in Five: The 2026 Census Reality Check on AI Adoption

What did the Census Bureau actually measure about AI adoption?

The Census Bureau Business Trends and Outlook Survey (BTOS) is a biweekly pulse check on business conditions across the United States, collecting data every two weeks across six rotating panels of roughly 200,000 businesses each. Starting in 2023, the survey added a direct question about AI use: not about plans to adopt AI, not about attitudes toward technology, but about whether the business used AI in a business function in the prior two weeks.

The May 2026 data puts national AI adoption at 19.8% of responding businesses. That is the headline figure. Every analysis in this article builds from it.

What counts as "using AI" in the BTOS? The questions cover tools that automate tasks, generate content, analyze data, or support decisions. This includes broad-use tools like ChatGPT, Microsoft Copilot, Google Gemini, and Claude when deployed in work settings. It also includes industry-specific AI software for forecasting, quality control, or customer service.

The precision of the question is the key distinction. Businesses are not rating how sophisticated their AI use is. They are answering a simpler question: did the business use AI in a business function in the past two weeks? Yes or no. That specificity is what makes Census data different from most AI adoption surveys, which regularly mix current use with implementation plans and aspirational statements.

When a company says it plans to use AI in the next six months, that registers as intent. When it tells the Census Bureau it used AI in a business function during the prior two weeks, that is a reported behavior. The 19.8% is the reported-use number. It is lower than most AI trend reports suggest, which is exactly why it is worth paying attention to.

For context: commercial surveys have published far higher figures. McKinsey's 2025 global survey reported that 78% of organizations used AI in at least one business function. Those surveys use different definitions, different populations, and often different questions. The Census figure is not better or worse; it measures something more specific. The BTOS asks about recent reported use in a business function, not general awareness or experimentation. That narrower standard produces a lower and arguably more useful number for benchmarking.

This likely understates informal employee experimentation, because BTOS measures business-reported use in a business function, not every employee who may have tried ChatGPT or another tool on their own. What the Census number captures is the population of businesses that report AI use at the function level. That population is one in five.

Why is company size such a strong predictor of AI adoption?

Company size is one of the clearest predictors of AI adoption in the Census data.

Large businesses with 250 or more employees report AI adoption at 37%. Mid-market companies in the 100-to-249 employee range come in at 32%. Small businesses with fewer than five employees fall below 20%. The spread between the largest and smallest employers is nearly 20 percentage points.

This size pattern shows up across the sector data. The number of people a company employs tracks closely with whether it has reported AI use.

Why does size predict adoption this consistently? The short answer is capacity. Large companies have dedicated IT staff who can evaluate, configure, and maintain AI tools. They have HR departments that can onboard employees on new workflows. They have legal teams that can review data-use policies before tools go live. They have budgets that absorb a tool that does not work well immediately without triggering a cancellation.

Smaller companies have one or two people wearing multiple hats. The person who might champion AI adoption is also handling customer service, managing the books, and attending to whatever broke today. The friction cost of trying and failing with a new tool falls entirely on them.

The bottleneck is organizational bandwidth more than awareness. Business owners in 2026 know AI tools exist. What separates the 37% from the 19.8% is the capacity to absorb change, evaluate outputs, and adjust workflows.

What the size breakdown means for Indiana businesses

For Indiana leaders, this size breakdown carries a specific weight. Indiana's business economy tilts toward mid-market manufacturers, family-owned distributors, regional healthcare systems, and local service firms. Many of the most important employers in the state sit in that 100-to-249 employee band where the national adoption figure is 32%.

That 32% is not a ceiling. But reaching it requires the same things that move large companies past the national average: enough internal capacity to absorb a learning curve, someone with the organizational authority to set expectations on how AI outputs will be used, and a function where AI output quality can be evaluated relatively quickly.

The companies in that 100-to-249 band that close the distance to large-firm adoption rates are not the ones with the biggest technology budgets. They are the ones with the clearest starting function and the most deliberate approach to evaluating AI results.

57% of AI-adopting businesses use it in three or fewer functions. The majority has crossed the adoption line. Most have not crossed the integration bar.

What does 57% using AI in three or fewer functions tell us?

The Census data includes not just whether businesses use AI but where they use it. Among businesses that report current AI adoption, 57% apply AI tools in three or fewer operational functions.

Pause on that number. The majority of AI-adopting businesses are running AI in a narrow slice of their operations. One team uses a chatbot. The finance department has automated one report. The IT team runs a single AI-assisted monitoring process. The rest of the business operates the same way it did two years ago.

This is what AI adoption looks like at Level 2 and Level 3 of the 7 Levels of AI Proficiency. At Level 2 (Ensign, also called Prompt Engineer or Practitioner), individuals use AI tools regularly and can direct them toward useful outputs. At Level 3 (Lieutenant, also called Critical Thinker), people apply judgment to AI outputs, catch errors, and use AI-generated results in real decisions.

Most of the 19.8% of businesses that report using AI sit at Level 2 or Level 3. They have crossed the adoption threshold. They have not crossed the integration threshold.

The integration threshold is what Level 4 (Commander, also called Context Engineer or Builder) represents in the 7 Levels of AI Proficiency framework. A Commander has moved from tool use in one function to systematic deployment across functions. Sales, operations, finance, and customer service are sharing AI-generated inputs. The tools are configured to the company's specific context, not running on factory settings applied to a generic use case.

The competitive difference between a Level 3 company and a Level 4 company rarely comes from the tools they use. Both might use the same AI products. The difference is in how they feed context into those tools and how they connect outputs across teams.

A Level 3 company asks an AI tool: "Write a proposal for this client." A Level 4 company asks: "Here is our standard pricing model, here are the notes from the last three calls with this client, here is the proposal we wrote for a comparable client last year, and here is our current delivery capacity. Write a proposal." The output from the second prompt is more accurate, more relevant, and more likely to advance the sale.

That difference in prompt construction is a workflow design question. It requires someone with the organizational authority to document and distribute the company's context for AI use. That is the Commander function.

57% of current AI users have not built that function yet. They have individual AI users. They do not yet have an AI-integrated operation.

Which industries are furthest ahead on AI adoption?

The Census sector breakdown shows meaningful variation around the 19.8% national baseline.

The Information sector leads at 39.7% adoption. Finance and Insurance follows at 33.9%. Professional, Scientific, and Technical Services runs around 30%. Wholesale Trade comes in at approximately 25%.

Retail Trade lags significantly, with adoption near 14%. Construction is lower still.

Two conditions predict high sector-level adoption. First, information intensity: industries that already process digital data, use software tools, and depend on analytical outputs find AI easier to add to existing workflows. The Information sector was already digital-first. AI extends what they were already doing.

Second, margin structure. Industries with higher margins can absorb the transition cost of building AI capability. A consulting firm can spend 40 hours configuring a new AI workflow without threatening the quarter. A restaurant operating on 5% margins cannot.

Manufacturing occupies an interesting position in this data. Many manufacturers work with sensor output, quality metrics, and supply chain variables that AI handles well. The analytics potential is clear. But capturing that potential requires connecting data streams to AI tools, which often means integrating systems that were not built with any external access in mind.

The technology fits the problem. The infrastructure often does not. A manufacturer running a production line on software installed in 2014 cannot simply pipe that data into a modern AI tool without an integration project. That project has a cost and a timeline that sits outside most mid-market manufacturers' current IT roadmaps.

For Indiana, where manufacturing employs more workers by payroll than any other sector, this infrastructure constraint is the central AI adoption challenge. The opportunity is not abstract. The path to it runs through legacy systems, data quality work, and investment decisions that compete directly with other capital spending.

The relevant question for an Indiana manufacturer is not whether AI can help with forecasting, quality inspection, or supply chain optimization. It can. The question is whether the data that AI needs is accessible in a usable form and whether the organization has someone who can oversee the connection between operational data and AI tool inputs. That is a solvable problem. It requires a different kind of work than buying a software subscription.

What does this mean for Indiana business leaders?

Indiana's workforce concentrates in manufacturing, logistics, healthcare, and regional services. These are sectors where the Census data shows adoption running below the 19.8% national baseline, not above it.

The practical read for a CEO running a mid-market Indiana company: the peer group most relevant to you is likely adopting AI at a rate between 20% and 30%, with real constraints on how deep that adoption goes. The companies running at 37% or above are large enterprises, metro-market professional services firms, and technology-intensive operations.

This does not describe a crisis. It describes a specific starting position.

The Census data points toward three functions where AI adoption is highest nationally: Sales and Marketing at 52%, Strategy and Business Development at 45%, and IT at 41%. These numbers reflect where AI tools have produced measurable, evaluable outputs quickly enough to earn continued use. Marketing teams can see whether an AI-generated campaign draft is good or not. Strategy teams can verify whether a competitive analysis missed an obvious data point. IT teams can tell whether a monitoring alert was accurate.

Each of those functions has a built-in feedback loop. You can check the output. That is the distinguishing feature of a good starting function for AI deployment: the speed of evaluation, not the size of the opportunity.

A mid-market Indiana company starting AI adoption from zero does not need to match the Information sector's 39.7% rate. The target is something more achievable: pick one function with a fast feedback loop, build the habit of evaluating AI outputs critically, and establish the organizational expectation that AI-generated results are checked before they go anywhere important.

That is the Commander path in the 7 Levels of AI Proficiency. Not every function at once. One function, done well enough to teach the organization how to evaluate AI outputs. Then the next.

Among Indiana companies, closing the distance to large-firm adoption rates over the next two years runs through the evaluation habit, not the tool budget.

What should you do this week?

Three steps, in sequence.

1. Count your current functions.

   Ask your leadership team which functions are already using AI tools, even informally. You want a current-state count, not an aspiration. The Census data shows most AI-adopting companies use AI in one to three functions. Knowing where you are is the starting point. If you find three functions, you are at or above the median for adopters. If you find none, you have a clear first-mover opportunity in your peer group.

2. Pick one expansion function based on feedback speed.

   Based on the Census data, Sales and Marketing, Strategy and Business Development, and IT show the highest adoption nationally. The reason is the feedback loop in each: you can evaluate the output relatively quickly. Pick the function in your business that has the fastest feedback cycle on AI outputs. Start there, not with the biggest problem AI could theoretically solve.

3. Document your context before you deploy.

   Before asking AI to help with any new function, write down what a good output looks like, what information the tool would need to do the job well, and how you will know if the output is wrong. This documentation is what separates Level 3 use (prompting AI for outputs) from Level 4 use (configuring AI with organizational context). It takes two hours. It makes every subsequent AI interaction in that function more accurate.

If you want to know where your team currently sits across the 7 Levels of AI Proficiency, the free assessment at assess.launchready.ai places individuals in about ten minutes. It covers technical skill, judgment, and organizational context-setting across all seven levels. The Commander profile at Level 4 is the target for most mid-market leadership teams in 2026.

Related reading: Level 4: Commander.

Sources

1. U.S. Census Bureau. AI Use at U.S. Businesses (America Counts), May 2026.
2. U.S. Census Bureau. The Microstructure of AI Diffusion (CES Working Paper 26-25), 2026.
3. U.S. Census Bureau. Business Trends and Outlook Survey (BTOS) Methodology, 2026.
4. McKinsey & Company. The State of AI: How Organizations Are Rewiring to Capture Value, 2025.
5. LaunchReady.ai. The 7 Levels of AI Proficiency Framework, 2026.
6. LaunchReady.ai. The AI Performance Divide: Why 74% of Value Goes to 20% of Companies.
7. LaunchReady.ai. How to Measure AI Readiness in a Team.

Harrison Painter

AI Business Strategist. Founder, LaunchReady.ai and AI Law Tracker.

Harrison helps teams build AI systems that cut cost and grow revenue. Nearly 20 years of business experience. 2.8M YouTube views. Founder of LaunchReady.ai and the 7 Levels of AI framework.

Connect on LinkedIn