Daily Hypernovelty Lead · Law & evidence · September 9, 2026

The Core Form Took the AI Items

Census wants AI usage and AI-and-employment items on the ordinary BTOS cycle. The Federal Register notice still does not print the questions.

Hands move a thinner stack of survey booklets onto a thicker core stack on a metal desk under a lamp, with a dusk window beyond.

A core form can take the AI items while the questions stay off the page.

The Census Bureau's September 9, 2026 Federal Register notice, document 2026-18304 published as 91 FR 57314, asks the Office of Management and Budget to revise the currently approved Business Trends and Outlook Survey under OMB control number 0607-1022.[1][2] The ACTION is a Paperwork Reduction Act comment request on a revision of a currently approved collection. FederalRegister.gov is an unofficial XML copy, so anyone who needs legal notice should use the official edition on govinfo.[1][2]

This collection is BTOS, which the Bureau calls the only biweekly business tendency survey produced by the federal statistical system.[1] The target population is all nonfarm employer businesses with receipts of $1,000 or more in the United States, the District of Columbia, and Puerto Rico.[1] The current sample is about 1.2 million businesses split into six panels; collection occurs every two weeks, and businesses in each panel are asked to report once every twelve weeks for one year.[1] Results are published every two weeks at national and state levels, for the 25 most-populous metropolitan statistical areas, and by industry and employment size class.[1] Response is voluntary under Title 13, United States Code, sections 131 and 182.[1]

That instrument already split, in its second year, into core content and supplemental content.[1] Core items cover measures the Bureau treats as broadly applicable across nonfarm sectors, including revenue, paid employees, hours, and inventories.[1] The same notice says core items may also include concepts that started as special topics, "such as the artificial intelligence questions that started in the second year."[1]

This revision asks for three changes that would complete that promotion. First, Census wants to repeat the artificial intelligence supplement with minor content changes after cognitive testing, and it says the proposed AI supplemental questions are submitted with the request.[1] Second, it wants a new core question about business uncertainty, also cognitively tested.[1] Third, it wants three additional core questions: Generative AI usage and the impact of AI on the total number of employees, both moved from the AI supplement, and a new item on the impact of AI on new employees, which the Bureau says has been cognitively tested to finalize wording.[1] To limit burden, Census says it will use skip logic and ask those items only of businesses that report AI usage.[1]

That promotion puts the items on the ordinary cycle alongside headcount and receipts instead of appearing only in designated sample years.[1] The Bureau is also exploring ways to collect and publish AI supplement data more often than annually and less often than every two weeks, because stakeholders have asked for a more complete picture of what it calls a swiftly emerging technology; it says evaluation of burden and methods is still underway, and that it may later consider other AI topics for BTOS.[1] None of that is a finding about jobs. A core question about AI and the number of employees cannot, by itself, prove displacement, hiring, productivity, or cause.

By contrast, the household time diary that BLS proposed for the American Time Use Survey asks a person to reconstruct a day.[3] BTOS asks an employer business. Those reporting units will not automatically join, and this notice still does not print the question wording it says it is submitting. A commenter who wants to judge practical utility, clarity, or skip logic has to open the OMB package on reginfo.gov, where written comments are due within 30 days of publication, by searching the survey title or 0607-1022.[1]

This notice estimates 396,000 responses in cycles 1, 3, and 4 at ten minutes each, and 132,000 responses in cycle 2 at twenty minutes, for 110,000 burden hours.[1] Core questions are estimated at no more than ten minutes; the AI supplements in sample years 2 and 4 added another ten minutes; the new AI supplement is estimated not to change that burden figure.[1] Those times are Bureau estimates, not observed completion times. The notice reports that cognitive testing happened. It does not reproduce the testing protocol or the items as tested.

That request is still sitting with OMB. It does not clear the revised questionnaire, start a new fielding cycle, or publish an AI-and-employment series. Earlier 60-day and 30-day comment periods ran in December 2024 and March 2025; this notice adds another 30 days.[1] Orientation only. It is not statistical-design, labor, legal, or investment advice.

Verification bottleneck

This scarce check is whether the items being promoted onto the core actually measure usage, total employment, and new hires as separate facts, when the Federal Register text still does not print the questions.

  • Census says the proposed AI supplemental questions travel with this request; the public notice does not include them, so the wording has to be checked in the OMB docket for 0607-1022.
  • Skip logic is described as asking the new core items only of businesses that report AI usage. That screen is only as good as the unpublished usage question.
  • Watch whether OMB clears the revision, whether Generative AI usage stays a core cycle item, and whether later publications separate "impact on total employees" from "impact on new employees" instead of collapsing both into a jobs story the form cannot carry.

Opportunities

So a one-firm worksheet is where value may appear, something a bookkeeper or operations lead could fill without pretending to be a labor economist. For a single reporting cycle it would record whether the firm would say it uses AI, what software that answer meant, current paid employees, whether any change in total headcount would be attributed to AI, whether any new hire would be, and which of those statements a ten-minute biweekly form could actually support. Idea fodder only.

A lighter companion is a comment-packet checklist against the PRA file on reginfo.gov: proposed wording, skip pattern, burden minutes, and whether "impact" is defined as a count, a direction, or a judgment.

Sources

[1] Census Bureau, "Agency Information Collection Activities; Submission to the Office of Management and Budget (OMB) for Review and Approval; Comment Request; Business Trends and Outlook Survey," 91 FR 57314, September 9, 2026, FR Doc. 2026-18304. https://www.federalregister.gov/documents/2026/09/09/2026-18304/agency-information-collection-activities-submission-to-the-office-of-management-and-budget-omb-for

[2] Official PDF, 91 FR 57314. https://www.govinfo.gov/content/pkg/FR-2026-09-09/pdf/2026-18304.pdf

[3] Hypernovelty Institute, "The Tool Enters the Time Diary," July 22, 2026. https://hypernovelty.institute/signals/the-tool-enters-the-time-diary/