AIORG-W003 Preprints

How Organizations Use AI: Evidence from ChatGPT

arXiv:2608.12236v1

preprint full text html

Recorded claims
3
coded from the inspected source
Evidence classes
1
first-party-organization
Recorded limitations
4
stated, never hidden

From the wave-0 evidence ledger

Recorded claims

Each claim states what the inspected source says, at the recorded location—bounded by its scope and graded by its confidence. Nothing here is a synthesis across works.

The study observed 1,764 organizations and 17.45 million messages during their first six months of enterprise use; its task-analysis subsample covered 973 organizations and about 8.7 million messages, and 417 organizations were matched to US public tickers.

confidence: high for reported sample; medium for external validity first-party-organization

Scope: ChatGPT Enterprise customers present in the study windows; message use, not all organizational work.

Use grew about sevenfold across the observed organizations from June 2025 to March 2026 and about fourfold among an existing-customer cohort; use spread across task categories but remained uneven across functions and workers, with greater intensity among earlier-career workers.

confidence: medium-high first-party-organization

Scope: Observed enterprise customers and the paper's task/classification scheme; adoption intensity, not causal productivity.

Matched adopters were larger and showed higher market value and higher R&D and SG&A intensity than comparison firms, but the paper treats these as correlates and emphasizes organizational complements rather than establishing that product use caused those firm attributes or outcomes.

confidence: medium first-party-organization

Scope: 417 matched US public tickers; selection into enterprise adoption remains.

Boundaries

Limitations & independence

Recorded at coding time, carried with the work forever. A claim without its limits is not evidence.

Recorded limitations

  • Organizations using ChatGPT Enterprise are selected and non-representative
  • Message telemetry omits work performed outside the product
  • Task coding and organization matching introduce measurement error
  • Correlations do not identify productivity or firm-value effects

Source independence

OpenAI-affiliated analysis of OpenAI-controlled enterprise telemetry, augmented with matched public-firm data; not independent of the product.