Evidence and methodology

How to check every number on this site.

Research cutoff 2026-11-18. Publication, observation and retrieval dates are kept separately. Raw files are immutable and hashed; transformations are deterministic scripts; nothing on the site is a mock value.

Rules this site follows

  • Vendor data are labelled as vendor data and never presented as national.
  • Incompatible series are never merged into one line. The Census question change is a break, not a jump to interpolate.
  • Missing, suppressed and not-applicable values are shown as such, never as zero.
  • Forecasts and scenarios carry coral, dashed styling and a “scenario” badge.
  • Task exposure is not job loss; announced investment is not spending; construction jobs are not permanent jobs.
  • Total capex is not AI capex; market value is not household welfare.
  • OWID training-compute data are Epoch data: one evidence family, not two confirmations.
  • No composite “AI prosperity score” is published. The six dimensions (adoption; business outcomes; worker opportunity; consumer and service benefits; infrastructure and supply chains; local fiscal and community outcomes) are reported in native units first.

Pipeline

  1. pipelines/fetch_*.py, ramp.py, bls.py, sec.py, aei.py write dated raw snapshots and log URL, time, bytes and SHA-256 to data/raw/manifest.jsonl.
  2. pipelines/build_published.py validates and writes compact JSON to data/published/.
  3. scripts/export-research.ts writes the research CSVs from the same registry the site renders.
  4. npm test checks invariants (no em dashes, no merged regimes, shares within bounds, no scenario above 100%).

Downloads

Public-domain and CC BY sources only. Ramp series are excluded until its reuse terms are confirmed.

Printable research brief

Source inventory

20 sources, classified.

SourceClassUniverse and denominatorCoverageAccessLicenseLimitations
Anthropic Economic Index release 2026-06-26 ('Cadences')
Anthropic · interest: Anthropic (model developer; commercial interest in AI adoption)
vendor telemetrySampled Claude.ai chat and Cowork conversations (Free, Pro, Max) and 1P API traffic excluding Claude Code
÷ All sampled conversations in the geography and month (published cells only)
2026-04-01 to 2026-05-31
Global; 121 countries published (May 2026); U.S. states + D.C. + Puerto Rico (overall metrics only)
downloaded and validatedCC BYOne platform. Task labels are not users' occupations. Consumer plans and 1P API only; Claude Code reported separately. Suppressed cells mean shares sum to under 100%.
Ramp AI Index
Ramp Economics Lab · interest: Ramp (spend-management vendor)
vendor telemetry85,000+ U.S. businesses using Ramp corporate cards and bill pay (per Ramp)
÷ Ramp businesses active in the month (exact eligibility not published)
2023-07 to 2026-10
United States, businesses on Ramp; state, NAICS sector, size class, vendor breakdowns
published aggregate onlyRamp website terms; reuse terms not verifiedNot a probability sample. Misses free tools and personal accounts (Ramp says likely undercount). Paying is not productive use.
A New Look at AI's Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment (Kharazian, Simon, Stevens)
Ramp and Revelio Labs · interest: Ramp and Revelio Labs (vendors)
observational study21,559 firms linked by web domain; 15,926 never, 3,969 low, 1,664 high intensity
÷ Not-yet-treated firms in same eventual intensity group
2021-01 to 2026-02
United States, linked Ramp/Revelio firms
downloaded and validatedPublic working paperSelected sample of Ramp customers; high-intensity firms small (mean 26.7 employees) and tech-adjacent (63.8%); 3 of 11 pre-periods flagged for the headline; significant only in Information.
Business Trends and Outlook Survey (BTOS): AI / 'Super Intelligence' core questions
U.S. Census Bureau
representative surveyEmployer businesses (excluding farms) in the BTOS sample frame
÷ All responding employer businesses (weighted)
2023-09-11 to 2026-11-01
U.S., 50 states + D.C., NAICS sectors, size classes, top 25 MSAs
downloaded and validatedPublic domainLow response rates; questionnaire breaks; no collection Oct 6 to Nov 16, 2025 (funding lapse).
Current Employment Statistics (CES): employment and average hourly earnings by industry
U.S. Bureau of Labor Statistics
administrative recordNonfarm payroll jobs
÷ n/a (levels)
2019-01 to 2026-10
United States
downloaded and validatedPublic domainJobs, not people; excludes farms and self-employed. Benchmark revisions each February.
Occupational Employment and Wage Statistics (OEWS), May 2025, national
U.S. Bureau of Labor Statistics
administrative recordWage and salary workers in nonfarm establishments
÷ n/a
2025-05 to 2025-05
United States
downloaded and validatedPublic domainOnly latest year retrieved; trend analysis (e.g., radiologists 2016-2025) needs flat files.
GDP by Industry: value added, real value added indexes, contributions to growth
U.S. Bureau of Economic Analysis
administrative recordAll industries
÷ n/a
2005Q1 to 2026Q2
United States
downloaded and validatedPublic domainRevised annually; quarterly industry detail lags GDP.
Value of Private Construction Put in Place: Data center
U.S. Census Bureau
administrative recordPrivate construction projects
÷ n/a
2014-01 to 2026-09
United States
downloaded and validatedPublic domainExcludes chips and IT equipment; preliminary values revise.
Employment in Data Centers Increased by More Than 60% From 2016 to 2023
U.S. Census Bureau
administrative recordWorkers in the data-center industry (QWI; BLS for national counts)
÷ n/a
2016 to 2023
United States, states, counties
published aggregate onlyPublic domainArticle does not print the NAICS code; its national figures (306k to 501k) match the scale of NAICS 518 (data processing, hosting), which includes more than physical data-center operations.
AI data centers: A hidden driver of U.S. jobs risks becoming a missed opportunity
Stanford Institute for Economic Policy Research (SIEPR)
modeled forecastDomestic manufacturing industries supplying data-center components
÷ n/a
2026 to 2026
United States
downloaded and validatedPublic policy briefNot a count of hires; excludes substitution away from other uses of the same inputs (which would lower it); depends on a third-party capex forecast.
Data on AI Models (all, notable, frontier)
Epoch AI
compiled datasetAI models meeting Epoch inclusion criteria
÷ n/a
1950 to 2026-11-17
Global (organization country recorded)
downloaded and validatedCC BY 4.0Recent months under-collected; compute often estimated.
Frontier Data Centers
Epoch AI
compiled datasetLarge AI data centers tracked by Epoch
÷ n/a
to 2026-11
Global (77 of 96 sites in the U.S.)
downloaded and validatedCC BY 4.0Estimates, not metered power; capacity is not utilization.
Computation used to train notable AI systems
Our World in Data
compiled datasetNotable AI systems (from Epoch)
÷ n/a
Global
accessible but not yet downloadedCC BYSame evidence family as Epoch: not independent confirmation.
EDGAR XBRL company facts (19-issuer AI-economy cohort)
U.S. Securities and Exchange Commission
corporate filingDefined cohort; not all public companies
÷ n/a
FY2019 to FY2026
U.S.-listed issuers (consolidated, global operations)
downloaded and validatedPublic domainUniverse file (company_tickers_exchange.json) returned 403; total capex is not AI capex; employees mostly untagged.
Business applications (Census BFS), output per hour (BLS), real GDP (BEA)
Federal Reserve Bank of St. Louis (FRED)
administrative recordVarious
÷ n/a
United States
downloaded and validatedPublic domain source seriesRepublication; vintages follow the source agencies.
Projected Impact of Generative AI on Future Productivity Growth
Penn Wharton Budget Model
modeled forecastU.S. economy
÷ n/a
United States
published aggregate onlyPublicAssumes 23% of exposed tasks automated and near-full adoption within ~15 years.
The Simple Macroeconomics of AI (w32487)
NBER (Daron Acemoglu)
modeled forecastU.S. economy
÷ n/a
United States
published aggregate onlyWorking paperBuilt on early task-level studies; excludes new tasks/products.
The economic potential of generative AI: The next productivity frontier
McKinsey Global Institute · interest: McKinsey (consulting; sells AI implementation services)
modeled forecast63 modeled use cases
÷ n/a
Global
not yet verifiedProprietary reportPage refused automated retrieval on 2026-11-18; figure not re-verified this pass. Global, potential, not realized U.S. GDP.
Colossus permits, utility approvals and council actions (as reported)
Shelby County Health Department; TVA; MDEQ; Memphis City Council
journalismxAI / SpaceXAI facilities
÷ n/a
Memphis and Shelby County, TN; Southaven, DeSoto County, MS
not yet verifiedPublic records (accessed via reporting)Tax bills, PILOT files, MLGW contracts and permit PDFs not yet obtained.
The State of Enterprise AI 2025
OpenAI · interest: OpenAI (model developer)
customer surveyOpenAI enterprise customers + worker survey
÷ OpenAI customers
Enterprises (global)
not yet verifiedProprietaryReturned HTTP 403 to automated retrieval on 2026-11-18; not used for any figure.

Claim ledger

Every headline, with its permitted wording.

IDClaimTypeStatusPermitted public wordingCounterevidence
C01Computer and mathematical tasks are the largest occupational task group in U.S. Claude.ai conversations.observed descriptive resultverifiedIn Anthropic's sample of U.S. Claude.ai conversations (May 2026), 21.1% were classified to computer and mathematical tasks, the largest of 22 groups. This describes activity on one platform, not users' occupations or the effects of all AI.API traffic leans further to code (28.2% globally), consumer chat less so; non-chat AI (embedded, industrial) is invisible here.
C02A widely shared figure says 21.7% of Claude.ai conversations are computer and mathematical.observed descriptive resultnot reproducedUse the release value with its period (21.1%, U.S., May 2026) rather than 21.7%.n/a
C0356.8% of U.S. businesses on Ramp paid for AI in October 2026, up 1.0 point from September and 13.3 points year over year.observed descriptive resultverifiedAmong businesses that use Ramp, 56.8% paid for an AI product in October 2026. This is not a national adoption rate.Census BTOS (representative) shows 24.2% using AI in late Oct 2026 under different wording and denominator.
C04Measured business AI use in the Census BTOS roughly doubled when the question was broadened in November 2025.observed descriptive resultverifiedWhen Census broadened the AI question in November 2025, measured use jumped from 10.0% to 17.3%. The two series are shown separately and never joined.n/a
C05High-intensity AI adopters grew headcount about 10% over two years after adoption.associationverified with caveatsIn a linked sample of Ramp customers, the heaviest AI spenders grew headcount about 10% relative to later adopters over two years, driven by Information firms. Lighter adopters showed no change. This cannot be scaled to national employment.Low-intensity adopters (the majority of adopters) show no change; Ramp's own freelancer research documents substitution.
C06Data-center construction supports about 304,000 U.S. manufacturing jobs.modeled scenarioverifiedSIEPR models about 304,000 U.S. manufacturing jobs supported by data-center construction at 2026 investment levels (range ~150,000 to 450,000+ if investment is 50% lower or higher). It is a modeled estimate, not a count of new hires.Same brief: GPUs are 48% of capex but 4% of domestic jobs (import dependence).
C07Employment in data centers grew from 306,000 to 501,000 between 2016 and 2023.observed descriptive resultverified with caveatsEmployment in the industry Census uses for data centers rose from 306,000 (2016) to 501,000 (2023). That industry includes hosting and data processing, and it has edged down since 2023 on BLS monthly data.On BLS CES NAICS 518, employment peaked near 485k in early 2023 and was 454.8k in Oct 2026 (preliminary), while data-center construction spending rose to a $96.5B annual rate.
C08Census: 6.6% of businesses say AI increased headcount in the last six months; 2.5% say it decreased.observed descriptive resultverifiedIn late October 2026, 87.3% of U.S. businesses reported no change in headcount from AI over six months; 6.6% reported an increase and 2.5% a decrease (self-reported, one survey cycle).n/a
C09Penn Wharton projects AI raises U.S. GDP and productivity levels 1.5% by 2035.forecastverifiedPenn Wharton's model (Sept 2025) projects a 1.5% higher U.S. GDP level by 2035. A projection, not an observation.Acemoglu: under 0.66% TFP over 10 years.
C10Acemoglu estimates no more than 0.66% TFP gain over 10 years.forecastverifiedAcemoglu (2024) bounds the 10-year TFP gain at 0.66% (under 0.53% after adjusting for task difficulty).PWBM and others project larger effects.
C11Generative AI could add $2.6 to $4.4 trillion annually.forecastunverifiedMcKinsey (2023) estimated global annual value potential of $2.6 to $4.4 trillion across modeled use cases (not re-verified this pass; never shown as U.S. GDP).n/a
C12xAI received no tax incentives for Colossus in Memphis.corporate commitmentreported, not independently verifiedNo Memphis/Shelby County incentive agreement for Colossus was identified in the records reviewed so far; the Chamber has said there were none. Tax records have not been obtained.Mississippi says the Southaven project uses the state's standard data-center incentives.
C13Colossus 2 will add 200 to 400 jobs.corporate commitmentreported, not independently verifiedxAI said Colossus 2 would add 200 to 400 jobs (company claim, not verified).n/a
C14Data-center construction spending reached a $96.5 billion annual rate in September 2026.observed descriptive resultverifiedPrivate data-center construction ran at a $96.5 billion annual rate in September 2026 (preliminary), up from $1.6 billion in January 2014.n/a

Missing-data register

What is not here yet, and why.

GapWhyNext step
OEWS occupation history 2016-2024 (radiologists, nurses, teachers trends)bls.gov flat files return HTTP 403 to automated clients without a contact addressRun pipelines with a contact address you approve, or download oesmYYnat.zip manually into data/raw/bls_oews/
Full SEC issuer universe (Layer A)sec.gov company_tickers_exchange.json returned 403Retry with approved contact header; or use EDGAR full-index files
AI-specific revenue for any issuerNot tagged in XBRLHuman review of 10-K segment notes and MD&A with excerpt capture
Memphis tax, PILOT, MLGW contract, permit PDFsNot retrievedShelby County Trustee/Assessor, EDGE, MLGW board minutes, SCHD permit file
OpenAI enterprise and economic reportsHTTP 403 to automated retrievalManual review; record figures with denominators
McKinsey figure re-verificationSite refused automated retrievalManual check
Ramp token data by model and token type (input/output/cached)Only provider-level shares published; cached-token chart values not in payloadRequest from Ramp Economics Lab
Ramp reuse termsNot reviewedConfirm before offering Ramp CSVs
Sector outcome studies (health, transport, retail) with measured outcomesNot yet assembledNBER, AHRQ, CMS, BTS literature pass
EIA electricity sales and prices near data centersEIA API requires a keyRegister a free EIA key (your call) or use bulk files
Net national AI-created jobsNo credible economy-wide identificationNot currently established by the evidence assembled.