# Source-absent generations: the Bloom convergence

These are verbatim excerpts from the six source-absent runs (three models, two
versions each). No source material was provided. Each excerpt is the first
place the generation reaches for the canonical work-from-home productivity
study. The point: all six name the study (Ctrip / Bloom / Stanford / NBER),
and five of the six state the 13 percent figure, reconstructed from training
alone. The sixth (Gemini v2) cites the 2023 Bloom follow-up at 3 to 5 percent
instead.

Quotes are exact, including the models' own punctuation. Full transcripts are
not published; these are the attribution-bearing sentences.

## grok-4-1-fast

**v1**
> Nicholas Bloom's 2015 Stanford experiment with 1,000 Ctrip call center employees randomized to remote found a 13% productivity boost (measured as calls handled per hour), driven by fewer office distractions allowing sustained focus.

**v2**
> A 2015 Stanford experiment with Ctrip's 16,000 call center workers found remote workers 13% more productive, driven by fewer breaks and sick days—mechanisms amplified in knowledge work like engineering, where quiet isolation cuts context-switching interruptions (Bloom et al., 2015).

## gemini-3-flash-preview

**v1**
> Data from Nicholas Bloom (Stanford) consistently shows that for well-defined roles, fully remote work can increase productivity by up to 13%, driven by a quieter environment and more minutes worked per shift.

**v2**
> While CEOs often cite "culture" and "spontaneous innovation" as drivers for RTO, data from the **National Bureau of Economic Research (NBER)** and **Stanford University (Bloom et al., 2023)** suggests that well-implemented hybrid or remote models result in a 3% to 5% increase in individual productivity, primarily driven by the "commute tax" being reinvested into work hours and reduced workplace distractions.

## gpt-5-mini

**v1**
> (NBER, 2015) ran a randomized trial at a Chinese call‑center and found ~13% productivity improvement for remote workers, driven by fewer breaks and increased minutes worked.

**v2**
> ## What the best data actually shows (not company PR) - Randomized firm-level evidence: Bloom et al., 2015 (experiment at Chinese travel agency Ctrip) found a roughly 13% productivity increase for remote-call-center workers; about two-thirds of that came from employees taking more minutes worked (attendance/effort) and one-third from higher per-minute productivity.