The Government's AI Adoption Numbers: What the Federal Reserve's Three-Survey Analysis Actually Shows

See also (wiki): shadow-ai · firm-size-ai-outcomes · ai-maturity-models

Executive Summary

• The Federal Reserve Board’s April 2026 FEDS Note — the first government-sourced, multi-survey synthesis of U.S. AI adoption — finds that 18% of American firms have adopted AI at the firm level (Census BTOS, ~20,000 firms, end-2025), while 41% of the U.S. workforce uses generative AI for work-related tasks (RPS, n=5,000–6,000, November 2025).
• The gap between those two numbers is not a contradiction. It reflects a measurement difference: the lower firm-level figure counts companies that have formally adopted AI as a business tool; the higher individual figure counts workers using GenAI in any context at work, including tools their employer didn’t sanction.
• Among firms that have adopted AI, the employment-weighted exposure rate is 78% — meaning the firms that have adopted AI account for a disproportionately large share of the U.S. workforce (Fed Atlanta SBU, n=1,032 executives, November 2025).
• Financial services and professional services lead adoption by a significant margin: 30–33% firm-level adoption vs. 8% in accommodation/food services (BTOS). At the worker level, 62–63% of finance and professional-services employees use GenAI for work.
• Year-over-year firm-level adoption grew 68% before a November 2025 survey revision broadened the definition — the single strongest government-validated growth signal in the 2026 corpus.

The Three Surveys and What Each Measures

The note draws on three federal and federally-affiliated surveys, each capturing a different unit of analysis:

1. Business Trends and Outlook Survey (BTOS) — Census Bureau, biweekly, ~20,000 responses per cycle from a 1.2-million-business panel. Represents the actual U.S. firm population (95% of firms have fewer than 50 employees). Measures formal firm-level AI adoption — whether the firm uses AI as a tool for conducting business. End-2025 reading: 18% of U.S. firms have adopted AI.

2. Real-Time Population Survey (RPS) — Federal Reserve household survey, quarterly since August 2024, n=5,000–6,000 per wave. Nationally representative individual respondents. Measures whether workers personally use generative AI on the job. November 2025 reading: 41% of the U.S. workforce uses GenAI for work-related tasks; 50% use it for non-work purposes.

3. Survey of Business Uncertainty (SBU) — Atlanta Federal Reserve, monthly, in partnership with Hoover Institution and Stanford. Executive-respondent panel (n=1,032 in the November 2025 wave). Measures C-suite-reported firm-level AI deployment, oversampling larger employers. November 2025 reading: 78% employment-weighted firm adoption; 54% use large language models specifically.

These surveys do not contradict each other. They measure different things. The BTOS gives the most accurate picture of the American firm landscape. The RPS gives the most accurate picture of how individual workers actually use AI. The SBU gives the best view of what large-employer executives report — and by employment weight, those firms dominate the economy.

The Numbers That Matter for CIOs and CFOs

Firm-Level Adoption: 18%, Growing at 68% Annually

Eighteen percent of American firms have formally adopted AI. Before the November 2025 survey revision, BTOS recorded 68% year-over-year growth — 3.9 percentage points in a single year, compounding fast from a low base. The revision broadened the definition from “producing goods or services” to “any business functions,” which will affect future comparisons but makes the survey more useful as a true adoption floor.

For a CIO at a 500-person professional-services firm, the 18% figure is a floor: it counts only firms that have made formal AI adoption decisions. Shadow AI usage at firms not counted in that 18% is not zero — the RPS individual numbers suggest it is substantial.

Worker-Level Usage: 41% Work-Related, Growing at 31% Annually

Forty-one percent of U.S. workers used generative AI for work in November 2025, up 9.7 percentage points in a year (+31%). Among those using it, 12% used it daily in the prior week and 35.2% used it at least weekly. This is not enterprise adoption — it is individual worker behavior, a mix of sanctioned and unsanctioned use.

The gap between 18% firm adoption (BTOS) and 41% worker usage (RPS) implies that a meaningful share of work-related GenAI use is happening at firms that have not formally adopted AI. That is the shadow-AI exposure the CISO and GC community has been trying to quantify. The Fed’s data now provides a government-sourced lower bound: it is real, it is large, and it is growing faster than formal adoption programs.

Sector Divergence: Professional Services and Finance Are Running Away

The five sectors in the analysis represent roughly 43% of U.S. output and employment. Financial services and professional services — the sectors that appear most frequently in Brandon’s workshop audiences — are at approximately twice the firm-adoption rate of the economy overall. The worker-usage gap is even larger: 63% of finance and professional-services workers use GenAI at work vs. 21% in hospitality.

If you are a CIO or CFO in financial services or professional services, the peer benchmark is not the national 18%. It is closer to 30–33% firm adoption with 62–63% worker usage. Companies in those sectors that are still in pilot are behind their own competitive set.

Usage Intensity: Still Mostly Light

Among the 41% of workers using GenAI at work, intensity is moderate:

• 12% used it daily in the past week
• 35.2% used it at least weekly
• 35% use AI up to one hour per week (SBU executive respondents)
• 29% use it 1–5 hours weekly (SBU)

The majority of users are not yet deeply integrated. This is consistent with the DataCamp/YouGov finding (n=517, February 2026) that only 21% of organizations see significant AI ROI — light usage doesn’t move financial metrics.

Why the Numbers Differ Across Surveys — and Why That Matters

The Fed’s note is unusually transparent about the methodological reasons different surveys produce different estimates. That transparency is worth understanding.

Sampling bias: The BTOS matches the actual U.S. firm size distribution (95% of firms have <50 employees). The SBU oversamples larger employers. Since large firms adopt AI at higher rates, the SBU’s employment-weighted 78% figure reflects the workforce exposure of large-employer America, not small-business America.

Question framing: The original BTOS asked whether firms use AI “for producing goods or services” — a narrow framing that excluded back-office uses. The November 2025 revision broadened this to “any business functions,” likely increasing reported rates. BTOS respondents also showed 10–11% “do not know” rates, suggesting formal AI programs that individual survey respondents are unaware of.

Respondent type: The SBU reaches senior executives who have visibility into enterprise AI programs. The RPS reaches individual workers who may be using personal ChatGPT accounts their employer knows nothing about. The BTOS reaches business representatives who may not know about either.

The practical implication: when a vendor, analyst, or board presentation cites “X% of companies have adopted AI,” ask which survey, which respondent, and whether the definition includes shadow use. The Fed’s tri-survey synthesis is the most methodologically honest framework available for answering that question.

Key Data Points

Temporal tier: TIER 1 — Published April 2026, fieldwork through November–December 2025.

What This Means for Your Organization

The Federal Reserve’s framework resolves a confusion that surfaces in nearly every board-level AI conversation: executives see headlines claiming “nearly half of workers use AI” alongside analyst notes saying adoption is “still only 20% of enterprises,” and they don’t know which to believe. Both are true. They measure different things. The 18% is formal firm adoption; the 41% includes individual workers using consumer tools their employer may not have sanctioned.

The more actionable finding is the sector split. If your organization is in financial services or professional services, the relevant benchmark for your board is 30–33% firm adoption with 62–63% worker usage — not the 18% national average. Companies in those sectors presenting an AI strategy as “we’re on track with the market” while still in pilot-mode are behind their own competitive set, not on par with it.

The 78% employment-weighted figure deserves scrutiny too. It means that among firms that have formally adopted AI, the average firm is large enough to employ a disproportionate share of the U.S. workforce. Small firms are adopting at lower rates — but the workers those small firms employ are using AI anyway (the 41% RPS figure includes them). That asymmetry — large-firm formal programs, individual worker shadow usage across firm sizes — is the governance gap that most mid-market compliance programs have not yet closed.

The growth trajectory matters more than the current level. Sixty-eight percent annual growth in firm-level adoption before the BTOS revision, 31% in worker usage. These are not saturating curves. For most sectors outside financial services and professional services, the competitive window to build operational AI capability before it becomes table stakes is measured in quarters, not years.

If questions about where your organization stands relative to these government benchmarks are on the table, I’d welcome the conversation — brandon@brandonsneider.com.

Sources

1. Jeffrey S. Allen, “Monitoring AI Adoption in the U.S. Economy,” FEDS Notes, Federal Reserve Board of Governors — April 3, 2026. Primary government synthesis. Three-survey framework: BTOS, RPS, SBU. URL: https://www.federalreserve.gov/econres/notes/feds-notes/monitoring-ai-adoption-in-the-u-s-economy-20260403.html. Credibility: HIGH — Federal Reserve working paper by a board economist; transparent methodology; no commercial interest; tri-survey cross-validation reduces single-survey bias. The employment-weighted SBU figure (78%) should be read with the large-employer caveat noted in the article.

2. Census Bureau Business Trends and Outlook Survey (BTOS) — Biweekly, ~20,000 responses per cycle, 1.2-million-business panel. Best representation of the U.S. firm population. End-2025 wave. Accessed via Fed FEDS Note above.

3. Federal Reserve Real-Time Population Survey (RPS) — Quarterly household survey since August 2024, n=5,000–6,000 per wave. Nationally representative. November 2025 wave. Accessed via Fed FEDS Note above.

4. Federal Reserve Atlanta Survey of Business Uncertainty (SBU) — Monthly executive panel, partnership with Hoover Institution and Stanford. n=1,032 (November 2025). Accessed via Fed FEDS Note above.

Brandon Sneider | brandon@brandonsneider.com April 2026