Make the evidence useful.
Save PDF / Print opens your browser’s print options. Choose “Save as PDF” for a copy.
Before you apply a headline to your business, check what was measured and what the study can support.
Start with the question
AI exposure estimates the potential relevance of AI to tasks. Adoption measures reported use. Work-from-home measures describe working arrangements. These concepts answer different questions and should not be substituted for one another.
Read the date twice
The publication date tells you when a paper appeared. The evidence period tells you when observations were collected or referenced. A September 2026 paper can analyze earlier observations. Expectations about the future are not observed outcomes.
Check who is represented
BTOS describes U.S. employer businesses within its survey coverage. It does not represent every entrepreneur or household. The exposure study restricts its analysis to single-unit firms. Each brief identifies important limits.
Respect the denominator
The percentage of businesses using AI differs from an employment-weighted percentage. A result among AI users does not describe all businesses. Comparisons require aligned populations, questions, periods, and definitions.
Do not erase a definition change
The earlier core AI wording asked about use in producing goods or services. In November 2025, collection changed to ask about any business function. A comparison spanning that change needs to explain the broader definition.
Separate association from cause
A relationship between AI integration and business performance does not, by itself, demonstrate that AI caused that performance. Employer perceptions of productivity also differ from a controlled measurement of output.
Keep geography honest
A national, state, or metro result belongs to that geography. Linking a ZIP code to regional context does not create a measured ZIP-level AI adoption rate. These briefs do not publish geographic rankings.
Know whose conclusion you are reading
The collection identifies findings separately from Chamber interpretation. Census working papers have not undergone the review accorded official Census Bureau publications. NBER working papers are circulated for discussion, rather than being peer-reviewed NBER publications. Read the original source for full methods and results.
Read the primary sources
- The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks
- AI Exposure and Adoption among U.S. Firms
- Tracking Firm Use of AI in Real Time: A Snapshot from the Business Trends and Outlook Survey
- Tapping Business and Household Surveys to Sharpen Our View of Work from Home
Confirming the business being surveyed
The form asks single-unit businesses to confirm their employer identification number. This is an administrative identity check. Public Census tables do not release those confidential identifiers, and this site does not publish them.
Understanding who is still operating
The form asks whether the business is operating, temporarily or seasonally inactive, closed, or transferred to another operator. This supports survey administration and coverage. Our imported publication tables do not provide an answer distribution for this question.
Question numbers and collection windows
This website follows questions 1–35 in the BTOS Content V4 Cycle 2 form with AI Supplement, revised July 23, 2025. The complete form also includes remarks, contact, and company-information questions after question 35. Workbook question IDs use a different sequence; our question catalog preserves the mapping.
General-business results come from the October 8, 2026 release, period 202620. The detailed AI tables come from the April 23 release and an earlier pooled collection. Future expectations are measured at the time of response.
Preserve the respondent group
Depending on the question, AI results describe all in-scope companies, AI users, businesses replacing employee tasks, businesses reporting employee generative-AI use, AI planners, or non-planners. Those groups are not interchangeable. Multiple-response percentages need not total 100.
The extreme-weather event follow-up describes businesses that reported monetary losses, rather than all businesses.
Use uncertainty alongside the estimate
Every BTOS results table includes the source standard error in percentage points. The historical ABS tables retain source flags; numerical standard errors are not supplied in the uploaded ABS files. Suppressed or unpublished values remain unavailable rather than becoming zeros. Differences between point estimates do not by themselves establish a statistically reliable ranking.
Survey estimates reflect business responses, sampling, and nonresponse adjustments. They do not demonstrate causal AI returns, identify vendors, or measure household outcomes.
Keep historical sources in their own windows
The history page now includes reviewed national observations from the 2024 BTOS AI supplement and 13 AI-specific 2019 ABS tables. The ABS technology period is 2016–18, and its table publication date is March 4, 2022. Its population, nonresponse adjustments, disclosure procedures, and 2017 NAICS classification differ from BTOS.
The question-comparison register identifies shared concepts, changed respondent groups, and questions with no counterpart. No numerical growth rate across the old and new AI definitions is approved. The full source archives are preserved in a separate historical candidate database; public downloads provide reviewed national tabulations.