In a preregistered experiment run in 2023 with 758 BCG consultants, using GPT-4 as available at the end of April 2023, the 385 who worked on 18 tasks inside the frontier of AI capabilities completed 12.2% more tasks, in about 25.1% less time and with roughly 30–34% higher human-graded quality when given GPT-4; among the 373 who worked on a task outside that frontier, those given GPT-4 were 19 percentage points less likely to reach a correct answer.

SocioLens project,

Claim LD02-HQ-C14 (v1.1)

Provenance and status

Claim type: Estimate
Publication status: Eligible for publication
Verification state: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Does not license: General consulting labour-demand effect; a single global “GPT-4 productivity effect”
Limitations: One firm; specific task set; the experiment dates from 2023 and used the GPT-4 of April 2023, so the result describes that model, not current ones
Source (as cited): Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science 37(2), 2026
Source type: Peer-reviewed journal article (recorded as “Peer-reviewed journal article”)
Institutional position: Mixed / BCG-affiliated research setting
Maturity of the evidence: Mature evidence
Recorded date: 22 October 2026
Currentness / correction: This is the current version. No correction or concern is recorded on its sources.

Sources this version cites

Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science 37(2), 2026 (Peer-reviewed journal article)
Read the original

Version

This page shows version v1.1. Its record never changes, even when newer evidence arrives; the site may add a note to the page or withhold a figure not cleared for public use.

This pinned version is also the current version.

Versions: v1.0, v1.1 · current

What changed in this version

This version supersedes version v1.0 (LD02-HQ-C14--v1.0), which stays available with its record unchanged.

Changed fields: 2 PUBLIC-FACING CLAIM; 9 SCOPE; 11 SAMPLE / DATASET; 12 MODEL / TOOL / VERSION; 14 MAIN RESULT; 18 LIMITATIONS; 25 RECORDED DATE
What changed: The claim now gives the two separate groups of participants: 385 worked on the 18 tasks inside the frontier and 373 on the task outside it, so no figure rests on all 758 at once. “Worked 25.1% faster” and “+25.1% speed” now read “about 25.1% less time”, which is what the paper measured. The claim, scope, model field and limitations now say that the experiment was run in 2023 with GPT-4 as available at the end of April 2023, separately from the 2026 publication date. The main result names the paper’s tables and gives the correctness figures behind the 19-point fall. The figures, the source, the maturity and the verification state are unchanged.
Why: An external research-integrity review of the site on 22 October 2026 (an AI reviewer, not a person) found that the claim did not say the 758 consultants were split into two groups, that “+25.1% speed” misdescribes a reduction in time spent, and that the model vintage was not distinguished from the 2026 publication date. SocioLens re-read the published article (Organization Science 37(2), 2026) on 22 October 2026. Its §3 says “Approximately half of the participants (385 subjects)” worked on the product-development tasks and “The other half (373 subjects)” on the business problem-solving task, and that all used “GPT-4 as available at the end of April 2023”; the article dates its working-paper release to 18 September 2023, so the experiment ran in 2023; Table 6 reports time spent on the first 17 questions falling by 22.5% and 27.6% in the two AI arms; Table 7 reports 84.5% correct in the control group against 60% and 70.6% in the AI arms, “an average decrease of 19 percentage points”. The check and this revision were done by an AI agent reading the article, not by a person.

Raw record: every recorded field, with its coded values

All 25 canonical Claim fields, verbatim, with the coded values derived from them. The plain-language labels above are read from these.

Claim type (coded): ESTIMATE (as printed: “ESTIMATE”)
Publication status (coded): ELIGIBLE (as printed: “ELIGIBLE”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY
Source type (coded): PEER_REVIEWED_JOURNAL_ARTICLE
Maturity (coded): MATURE

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD02-HQ-C14	—
2	PUBLIC-FACING CLAIM	In a preregistered experiment run in 2023 with 758 BCG consultants, using GPT-4 as available at the end of April 2023, the 385 who worked on 18 tasks inside the frontier of AI capabilities completed 12.2% more tasks, in about 25.1% less time and with roughly 30–34% higher human-graded quality when given GPT-4; among the 373 who worked on a task outside that frontier, those given GPT-4 were 19 percentage points less likely to reach a correct answer.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	B — Task-level productivity and quality	—
5	MATURITY	MATURE	states: MATURE
6	SOURCE	Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science 37(2), 2026	source_ids: SRC-LD02-DELLACQUA-ORGANIZATION-SCIENCE-2026
7	SOURCE TYPE	Peer-reviewed journal article	classes: PEER_REVIEWED_JOURNAL_ARTICLE
8	INSTITUTIONAL POSITION	Mixed / BCG-affiliated research setting	—
9	SCOPE	758 BCG knowledge workers in two separate groups; selected consulting tasks; experiment in 2023, published 2026	—
10	DESIGN	Preregistered randomised experiment	—
11	SAMPLE / DATASET	385 participants on 18 inside-frontier tasks; 373 on one outside-frontier business problem-solving task	—
12	MODEL / TOOL / VERSION	GPT-4 as available at the end of April 2023, default system prompt and temperature	—
13	COMPARATOR	Randomised no-AI group	—
14	MAIN RESULT	Inside the frontier (385): +12.2% tasks completed (Table 5); about 25.1% less time on the first 17 tasks (22.5% and 27.6% less in the two AI arms, Table 6); +29.9% to +33.9% human-graded quality depending on treatment arm (Table 4). Outside the frontier (373): correct answers fell from 84.5% without AI to 60% and 70.6% in the two AI arms, an average of 19 percentage points (Table 7)	—
15	INFERENTIAL TARGET	Task-specific productivity and failure across capability boundary	—
16	DOES NOT LICENSE	General consulting labour-demand effect; a single global “GPT-4 productivity effect”	—
17	VERIFICATION STATE	CONTENT VERIFIED	state: CONTENT_VERIFIED_PRIMARY · verified_by: UNKNOWN
18	LIMITATIONS	One firm; specific task set; the experiment dates from 2023 and used the GPT-4 of April 2023, so the result describes that model, not current ones	—
19	CONTRADICTORY / SUPPORTING SOURCES	Shows why positive task studies and failure results can coexist	—
20	GEOGRAPHY / LANGUAGE	Global firm / English	—
21	DISTRIBUTIONAL LEDGER	Below-average performers benefit more in reported analyses	—
22	R01 / FIELD-MAP LINKS	D2, D3, R-2	—
23	WHAT WOULD CHANGE THIS CLAIM?	Newer-model replication erases task-boundary failures	—
24	PUBLICATION STATUS	ELIGIBLE	state: ELIGIBLE
25	RECORDED DATE	2026-10-22	—

Field 22 references resolved against Field Map v0.2: D2 (DOMAIN); D3 (DOMAIN); R-2 (LENS)

The sources as recorded, with their record IDs and coded source types:

Pinned Sources (field 6, version-specific)

“Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science 37(2), 2026” → Dell’Acqua et al. (2026), “Navigating the Jagged…” (SRC-LD02-DELLACQUA-ORGANIZATION-SCIENCE-2026--v1.0) — Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science 37(2), 2026; class PEER_REVIEWED_JOURNAL_ARTICLE (explicit public provenance pin)

Cited by Living Debate versions

Aggregate null or distributional shift? Employment (v0.3) LD-02

Cite this page

SocioLens project. (2026, October 22). In a preregistered experiment run in 2023 with 758 BCG consultants, using GPT-4 as available at the end of April 2023, the 385 who worked on 18 tasks inside the frontier of AI capabilities completed 12.2% more tasks, in about 25.1% less time and with roughly 30–34% higher human-graded quality when given GPT-4; among the 373 who worked on a task outside that frontier, those given GPT-4 were 19 percentage points less likely to reach a correct answer [Claim LD02-HQ-C14, version v1.1]. SocioLens. https://sociolens-web.pages.dev/claims/ld02-hq-c14/v1.1/

@misc{sociolens_ld02_hq_c14_v1_1,
  author = {{SocioLens project}},
  title = {In a preregistered experiment run in 2023 with 758 BCG consultants, using GPT-4 as available at the end of April 2023, the 385 who worked on 18 tasks inside the frontier of AI capabilities completed 12.2\% more tasks, in about 25.1\% less time and with roughly 30–34\% higher human-graded quality when given GPT-4; among the 373 who worked on a task outside that frontier, those given GPT-4 were 19 percentage points less likely to reach a correct answer.},
  year = {2026},
  date = {2026-10-22},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld02-hq-c14/v1.1/}},
  note = {Claim LD02-HQ-C14, version v1.1}
}

The link points to this version, whose record never changes. For a finding, cite the original study as well: SocioLens summarizes research and does not replace it.

Research-of-record provenance

Materialized by sociolens-materializer 0.6.0 at 2026-09-28T20:00:00.000Z; normalization contract SocioLens — Claim Value Vocabulary v1.0.2 — Pre-Spike Materialization Contract (v1.0.2). Content seal SHA-256 441c753863116f550e1b2cbd02fe6c63d8532d6c7b581b0673926b25000f8397.

Repository-authored inputs (not Drive exports)

SocioLens claim revision — LD02-HQ-C14 v1.1 (v1.1) — written in the SocioLens repository by Project Lead (Claude, AI) on 2026-10-22; no Drive identity exists. Authority: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and docs/PROJECT.md "Research revisions" (a published claim changes only through a new version that states what changed and why). Block “ENTRY 14 — LD02-HQ-C14” SHA-256 813c046fc81ff746fd78d299dd99a51ac83836e0373da8ed421a387863b26d52; fixture fixtures/authored/claim-revisions/LD02-HQ-C14--v1.1.

Builder decisions (not Drive data)

VERSION = v1.1 — basis: Claim version taken from the repository-authored revision fixture version (SocioLens claim revision — LD02-HQ-C14 v1.1); a new version of LD02-HQ-C14 v1.0, which stays routable (AGENTS.md research integrity rule 3)
CURRENT_POINTER = true — basis: repository-authored revision "SocioLens claim revision — LD02-HQ-C14 v1.1" (Project Lead (Claude, AI), 2026-10-22); basis: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and docs/PROJECT.md "Research revisions" (a published claim changes only through a new version that states what changed and why)
FIELD_6_SOURCE_REFERENCE = SRC-LD02-DELLACQUA-ORGANIZATION-SCIENCE-2026--v1.0 — basis: field 6 of ENTRY 14 — LD02-HQ-C14 cites "Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science 37(2), 2026"; the field-6 text of this ClaimVersion's repository-authored ledger block; the same Source record LD02-HQ-C14 v1.0 cites.
SOURCE_PROVENANCE_PIN = SRC-LD02-DELLACQUA-ORGANIZATION-SCIENCE-2026--v1.0 — basis: explicit version-pinned public provenance of this ClaimVersion (Amendment v1.2 §6): the pinned field-6 Source record "Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science 37(2), 2026".