The AI Adoption Numbers: Why 18% and 78% Are Both Correct

Executive Summary

• The 18% vs. 78% adoption gap is not a contradiction — it is a measurement artifact. The Federal Reserve’s April 2026 analysis of three independent surveys reveals that every widely-cited AI adoption figure is technically accurate and practically misleading, depending on what it measures (Jeffrey S. Allen, FEDS Notes, April 3, 2026).
• Firm-level adoption (BTOS): 18%. At year-end 2025, 18% of US firms report using AI in any business function. This is the most conservative, most defensible number — weighted by number of firms, the vast majority of which are small businesses with little AI exposure.
• Employment-weighted adoption (SBU): 78%. Because large employers have higher AI adoption and employ more people, 78% of the US labor force works at firms that have adopted AI. Both numbers are from credible Federal Reserve surveys. Neither is wrong.
• Individual worker GenAI usage at work (RPS): 41%. Nationally representative quarterly survey of workers: 41% report using generative AI for their job as of November 2025, up 31% year-over-year. This is the most relevant number for mid-market CHROs planning AI training programs.
• Cognitive work leads adoption by a wide margin. Information (37% firm-level, 70% individual), financial services (30%, 63%), and professional services (33%, 62%) are the highest-adoption industries — the exact sectors where Brandon’s workshop audience works.

The Three Federal Reserve Surveys: What Each Measures

The Federal Reserve monitors AI adoption through three independent programs, each designed for different analytical purposes. Understanding which survey to cite — and for what claim — is the foundation of any honest AI briefing.

Business Trends and Outlook Survey (BTOS): 18% Firm Adoption

BTOS is the most conservative measure. Developed jointly by the Federal Reserve, Census Bureau, and Atlanta Fed, it surveys approximately 200,000 firms every 12 weeks, with roughly 20,000 responses per cycle. The universe covers 1.2 million US businesses. Responses are weighted by number of firms — which means small businesses (1–49 employees, 95% of all firms, 26.6% of employment) dominate the estimate.

The November 2025 BTOS modification broadened the AI question from “using AI to produce goods/services” to “using AI in any business functions.” Even with that expansion, firm-level adoption reached 18% at year-end 2025.

Why is this number low? Two reasons. First, the sample is 89.6% small firms — the segment least likely to have deployed AI beyond free tools. Second, approximately 10–11% of BTOS respondents answer “do not know” on the AI question, suggesting measurement error depresses the estimate.

Growth rate: 68% year-over-year (3.9 percentage points). The direction is unambiguous even if the level is conservative.

Survey of Business Uncertainty (SBU): 78% Employment-Weighted Adoption

The SBU surveys business executives at 1,032 firms monthly (AI questions added November 2025), operated by the Atlanta Fed, Chicago Fed, and Stanford. It is employment-weighted — each firm’s response is weighted by its headcount. Because firms with 250+ employees employ 56.2% of the US workforce, and those large firms have dramatically higher AI adoption rates, the employment-weighted figure reaches 78%.

The SBU also finds that 54% of the labor force works at firms using LLMs specifically.

This is the number that shows up in “AI is everywhere” headlines. It is accurate — but it measures where workers are employed, not whether those workers are personally using AI.

Real-Time Population Survey (RPS): 41% Individual Worker Usage

The RPS is the most useful number for mid-market executives. Run quarterly with 5,000–6,000 nationally representative workers, it asks directly: “Do you use generative AI for your job?” The definition is specific: tools that create text, images, audio, or video in response to prompts.

As of November 2025: 41% of workers report using generative AI at work, up 31.3% year-over-year. Of those:

An additional ~50% report using generative AI outside of work — meaning the personal exposure base is substantially larger than the workplace adoption figure.

Industry Breakdown: Where AI Has Actually Landed

The industry data reconciles the 18% (firm) and 41% (individual) figures by showing concentration in cognitive-intensive sectors:

The gap between firm-level and individual-level rates within each sector reflects an important dynamic: large firms in each industry adopt more aggressively, so the employment-weighted individual rate consistently exceeds the firm-count-weighted firm rate.

The practical implication for mid-market executives: If your firm is in professional services, financial services, or information industries, the relevant benchmark is 62–70% individual adoption — not the 18% all-industries firm average. The competitive cohort your talent compares themselves to is already at two-thirds AI usage.

Why Survey Results Vary: The Five Measurement Gaps

The Federal Reserve analysis names five structural reasons AI adoption estimates range from 18% to 88%, depending on source. Every executive team gets confused by this range; here is the explanation:

1. Sampling distribution. BTOS mirrors the firm-count population (95% small businesses). McKinsey’s State of AI surveys large enterprise by design. Neither misrepresents its data — they are measuring different populations.

2. Question framing. “Using AI to produce goods/services” (older BTOS framing) produces lower estimates than “using AI in any business function” (current framing), which produces lower estimates than “88% are using AI in at least one function” (McKinsey’s broad framing). Broader questions produce higher numbers.

3. Information asymmetries. BTOS respondents report 10–11% “do not know” rates. SBU respondents are business executives with organizational visibility. Executives report higher adoption because they are more likely to know what tools their organization has procured — even if individual workers are not using them.

4. Social desirability bias. In 2023–2024, some companies underreported AI adoption for security optics. By 2026, the pressure is reversed — executives face criticism for lagging. Self-reported AI adoption figures may be inflating under efficiency-investment scrutiny.

5. Materiality thresholds. Is a team using ChatGPT Free to draft emails “AI-adopted”? It depends on the survey. Conservative surveys require meaningful deployment; broad surveys count any usage.

What This Means for Your Organization

The measurement debate has a practical consequence: your board will see conflicting AI adoption figures and conclude the data is unreliable. It is not unreliable — it is measuring different things. The correct framing is:

• For competitive benchmarking: Use the RPS individual-usage rate for your industry. If you are in professional services, 62% of workers at your competitors are using generative AI at work. That is the peer comparison your talent evaluates themselves against.
• For budget justification: Use the BTOS firm-level growth rate (68% year-over-year) as the direction signal. Adoption is accelerating regardless of the starting-point disagreement.
• For workforce planning: Use 20%+ of BTOS firms expecting to adopt AI in the first half of 2026 as the near-term pressure indicator. Firms not currently using AI are joining rapidly.
• For HR training strategy: The 41% individual usage rate, growing at 31% annually, implies that by end of 2026 the majority of the US knowledge-work workforce will use generative AI at work. Training programs designed for a minority of early adopters are already miscalibrated.

The Federal Reserve data is the most credible independent triangulation point in the AI adoption literature — no vendor interest, no consulting engagement incentive, three independent methodologies. For executives tired of vendor surveys claiming impossible adoption rates, this is the honest baseline.

If reconciling these numbers against your organization’s own AI deployment data is something your leadership team is working through, I welcome the conversation — brandon@brandonsneider.com.

Key Data Points

Source credibility: HIGH. Federal Reserve FEDS Notes represent independent government research with no commercial interest in AI adoption levels. Three independent survey methodologies triangulated against each other. The methodology is transparent about limitations and disagreements between surveys — a standard no vendor research meets. Cross-reference: McKinsey State of AI (n=1,993, Nov 2025): 88% use AI in at least one function — this uses a much broader enterprise-focused sample with less conservative question framing. Federal Atlanta/NBER CFO Survey (n=748, Mar 2026): 1.8% perceived productivity gain, 0.6% measured. Both are consistent with the Fed FEDS picture of widespread but shallow adoption.

Sources

1. Jeffrey S. Allen, “Monitoring AI Adoption in the US Economy,” FEDS Notes, Federal Reserve Board of Governors, April 3, 2026. Primary source. Independent government research; no commercial interest. URL: https://www.federalreserve.gov/econres/notes/feds-notes/monitoring-ai-adoption-in-the-u-s-economy-20260403.html — TIER 1 (April 2026)

2. Business Trends and Outlook Survey (BTOS). Joint Federal Reserve / Census Bureau / Atlanta Fed; ~200,000 firms surveyed per 12-week cycle, ~20,000 responses. Year-end 2025. Source: Government statistical survey; HIGH credibility.

3. Real-Time Population Survey (RPS). Quarterly, n=5,000–6,000, nationally representative workforce, November 2025. Source: Independent workforce survey; HIGH credibility.

4. Survey of Business Uncertainty (SBU). Monthly, n=1,032 business executives, employment-weighted, Atlanta Fed / Chicago Fed / Stanford, November 2025. Source: Academic-government partnership; HIGH credibility.

Brandon Sneider | brandon@brandonsneider.com April 2026