Personalized GPT-4 outperformed human debate opponents in the Salvi et al. experiment, but a September 2026 Author Correction says the direct personalized-vs-non-personalized GPT-4 comparison did not reach conventional significance (P=.07).

SocioLens project,

Claim LD10-HQ-C06 (v1.0)

Provenance and status

Claim type: Estimate (status)
Publication status: Eligible for publication, with qualifiers: correction visible
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: Settled causal claim that personalization itself explains the advantage
Limitations: Structured debate; correction changes interpretation
Source (as cited): Salvi et al., Nature Human Behaviour 2025 + Author Correction 3 Sep 2026
Source type: Peer-reviewed journal article, Editorial notice or correction (recorded as “Peer-reviewed article + correction”)
Institutional position: Independent academic
Maturity of the evidence: Rapidly maturing evidence
Recorded date: 25 September 2026
Currentness / correction: This is the current version. Correction or concern on its sources: Salvi et al., Nature Human Behaviour, 2025 — personalized GPT-4 persuasion experiment: corrected by Salvi et al. Author Correction, 3 September 2026.

Sources this version cites

Salvi et al., Nature Human Behaviour, 2025 — personalized GPT-4 persuasion experiment (Peer-reviewed journal article)
Read the original Correction or concern: Salvi et al., Nature Human Behaviour, 2025 — personalized GPT-4 persuasion experiment: corrected by Salvi et al. Author Correction, 3 September 2026.
Salvi et al. Author Correction, 3 September 2026 (Editorial notice or correction)
Read the original Correction or concern: Salvi et al. Author Correction, 3 September 2026: corrects Salvi et al., Nature Human Behaviour, 2025 — personalized GPT-4 persuasion experiment.

Version

This page shows version v1.0. 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 · current

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 (STATUS) (as printed: “ESTIMATE / STATUS”)
Publication status (coded): ELIGIBLE (CORRECTION_VISIBLE) (as printed: “ELIGIBLE WITH CORRECTION VISIBLE”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY
Source type (coded): PEER_REVIEWED_JOURNAL_ARTICLE, EDITORIAL_NOTICE_OR_CORRECTION
Maturity (coded): RAPIDLY_MATURING
Source correction / concern relations (coded): SRC-LD10-SALVI-2025 CORRECTED_BY SRC-LD10-SALVI-CORRECTION-2026

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD10-HQ-C06	—
2	PUBLIC-FACING CLAIM	Personalized GPT-4 outperformed human debate opponents in the Salvi et al. experiment, but a September 2026 Author Correction says the direct personalized-vs-non-personalized GPT-4 comparison did not reach conventional significance (P=.07).	—
3	CLAIM TYPE	ESTIMATE / STATUS	base_types: ESTIMATE · qualifiers: STATUS
4	RESEARCH STRAND	A — Controlled persuasion	—
5	MATURITY	RAPIDLY MATURING	states: RAPIDLY_MATURING
6	SOURCE	Salvi et al., Nature Human Behaviour 2025 + Author Correction 3 Sep 2026	source_ids: SRC-LD10-SALVI-2025, SRC-LD10-SALVI-CORRECTION-2026
7	SOURCE TYPE	Peer-reviewed article + correction	classes: PEER_REVIEWED_JOURNAL_ARTICLE, EDITORIAL_NOTICE_OR_CORRECTION
8	INSTITUTIONAL POSITION	Independent academic	—
9	SCOPE	U.S. structured online debates	—
10	DESIGN	Preregistered RCT	—
11	SAMPLE / DATASET	N=900	—
12	MODEL / TOOL / VERSION	GPT-4	—
13	COMPARATOR	Human opponents; personalized vs non-personalized GPT-4	—
14	MAIN RESULT	64.4% of non-tied GPT-4-vs-human pairs favoured GPT-4; personalization causal contrast narrowed by correction	—
15	INFERENTIAL TARGET	GPT-4 vs human persuasion and supersession status	—
16	DOES NOT LICENSE	Settled causal claim that personalization itself explains the advantage	—
17	VERIFICATION STATE	CONTENT VERIFIED	state: CONTENT_VERIFIED_PRIMARY · verified_by: UNKNOWN
18	LIMITATIONS	Structured debate; correction changes interpretation	—
19	CONTRADICTORY / SUPPORTING SOURCES	Hackenburg finds smaller personalization effect in chat	—
20	GEOGRAPHY / LANGUAGE	US / English	—
21	DISTRIBUTIONAL LEDGER	Use of personal data may matter, but causal personalization advantage is not settled here	—
22	R01 / FIELD-MAP LINKS	D7, D4a, R-1, R-2	—
23	WHAT WOULD CHANGE THIS CLAIM?	Further corrected analysis or replication	—
24	PUBLICATION STATUS	ELIGIBLE WITH CORRECTION VISIBLE	state: ELIGIBLE · qualifiers: CORRECTION_VISIBLE
25	RECORDED DATE	2026-09-25	—

Field 22 references resolved against Field Map v0.2: D7 (DOMAIN); D4a (SUB_AREA); R-1 (LENS); R-2 (LENS)

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

Pinned Sources (field 6, version-specific)

“Salvi et al., Nature Human Behaviour 2025” → Salvi et al., Nature Human Behaviour, 2025 — personalized… (SRC-LD10-SALVI-2025--v1.0) — Salvi et al., Nature Human Behaviour, 2025 — personalized GPT-4 persuasion experiment; class PEER_REVIEWED_JOURNAL_ARTICLE — status: CORRECTED_BY SRC-LD10-SALVI-CORRECTION-2026 (explicit public provenance pin)
“Author Correction 3 Sep 2026” → Salvi et al. Author Correction, 3 September 2026 (SRC-LD10-SALVI-CORRECTION-2026--v1.0) — Salvi et al. Author Correction, 3 September 2026; class EDITORIAL_NOTICE_OR_CORRECTION — status: CORRECTS SRC-LD10-SALVI-2025 (explicit public provenance pin)

Cited by Living Debate versions

Persuasion, Intermediation & the Information Order (v0.1) LD-10
Raw record: where this debate places the claim, with its coded facets
HQ placement: SC-LD10-02; facets: evidence_status=CORRECTED, source_maturity=PEER_REVIEWED, claim_focus=PERSONALIZATION, public_caution=CORRECTION_VISIBLE

Cite this page

SocioLens project. (2026, September 25). Personalized GPT-4 outperformed human debate opponents in the Salvi et al. experiment, but a September 2026 Author Correction says the direct personalized-vs-non-personalized GPT-4 comparison did not reach conventional significance (P=.07) [Claim LD10-HQ-C06, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld10-hq-c06/v1.0/

@misc{sociolens_ld10_hq_c06_v1_0,
  author = {{SocioLens project}},
  title = {Personalized GPT-4 outperformed human debate opponents in the Salvi et al. experiment, but a September 2026 Author Correction says the direct personalized-vs-non-personalized GPT-4 comparison did not reach conventional significance (P=.07).},
  year = {2026},
  date = {2026-09-25},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld10-hq-c06/v1.0/}},
  note = {Claim LD10-HQ-C06, version v1.0}
}

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 34f7d99963ba5c9e73eab51682bc2b8e67534ed93d2bdf7bd01d7c7b5dd0c8bc.

Frozen Drive-derived inputs

R07 — LD-10 Canonical Claim Ledger v1.0 — LOCKED (v1.0) — Drive ID 1TwCjF7zxWTDs4bDvwQfEb5hbMAMuMu3vmK53kcOdo54; revision: NOT_EXPOSED_BY_DRIVE_TEXT_EXPORT; Drive modified 2026-09-25T21:56:50.285Z; block “ENTRY 06 — LD10-HQ-C06” SHA-256 706199ceae86af0883af0330b6cf5c26627e08fdad22ba9b13638b8794068a56; SocioLens — Programme Status & Authority Register v0.6 status “CURRENT.”; frozen fixture fixtures/frozen/r07-ld10-claim-ledger-v1.0.
SocioLens — Proof Source Registry v0.4 (v0.4) — Drive ID 1_-GLmj6PfetWDQH2KldMv-61vmSmzdAyyxgZZZCFnbE; revision: NOT_EXPOSED_BY_DRIVE_TEXT_EXPORT; Drive modified 2026-09-27T19:04:00.374Z; block “REGISTRY ENTRY SRC-LD10-SALVI-2025” SHA-256 ac104c95a6c5489a37bde54a1fec6b986204f79a5c3f3b4c5170f306e03ec792; SocioLens — Programme Status & Authority Register v0.6 status “CURRENT PROOF SOURCE REGISTRY.”; frozen fixture fixtures/frozen/proof-source-registry-v0.4.
SocioLens — Proof Source Registry v0.4 (v0.4) — Drive ID 1_-GLmj6PfetWDQH2KldMv-61vmSmzdAyyxgZZZCFnbE; revision: NOT_EXPOSED_BY_DRIVE_TEXT_EXPORT; Drive modified 2026-09-27T19:04:00.374Z; block “REGISTRY ENTRY SRC-LD10-SALVI-CORRECTION-2026” SHA-256 96113b14894c823d675e407042b3ca6250049a300a0a2483ac1d6513a7780cb5; SocioLens — Programme Status & Authority Register v0.6 status “CURRENT PROOF SOURCE REGISTRY.”; frozen fixture fixtures/frozen/proof-source-registry-v0.4.

Builder decisions (not Drive data)

VERSION = v1.0 — basis: Claim version taken from the canonical ledger artifact version (R07 — LD-10 Canonical Claim Ledger v1.0); Implementation Architecture Selection §5 logical_id/version convention.
CURRENT_POINTER = true — basis: SocioLens — Programme Status & Authority Register v0.6 status for "R07 — LD-10 Canonical Claim Ledger v1.0 — LOCKED": "CURRENT."
FIELD_6_SOURCE_REFERENCE = SRC-LD10-SALVI-2025--v1.0 — basis: field 6 of ENTRY 06 — LD10-HQ-C06 cites "Salvi et al., Nature Human Behaviour 2025"; Builder identification: the cited work is Proof Source Registry v0.4 identity SRC-LD10-SALVI-2025 (canonical label "Salvi et al., Nature Human Behaviour, 2025 — personalized GPT-4 persuasion experiment"); SocioLens record v1.0 is its sole record.
FIELD_6_SOURCE_REFERENCE = SRC-LD10-SALVI-CORRECTION-2026--v1.0 — basis: field 6 of ENTRY 06 — LD10-HQ-C06 cites "Author Correction 3 Sep 2026"; Builder identification: the cited work is Proof Source Registry v0.4 identity SRC-LD10-SALVI-CORRECTION-2026 (canonical label "Salvi et al. Author Correction, 3 September 2026"); SocioLens record v1.0 is its sole record.
SOURCE_PROVENANCE_PIN = SRC-LD10-SALVI-2025--v1.0 — basis: Issue #5 proof rendering (Architecture Proof Protocol v0.1 AP-08 / Amendment v1.2 §6): explicit version-pinned public provenance of this ClaimVersion — the pinned field-6 Source record.
SOURCE_PROVENANCE_PIN = SRC-LD10-SALVI-CORRECTION-2026--v1.0 — basis: Issue #5 proof rendering (Architecture Proof Protocol v0.1 AP-08 / Amendment v1.2 §6): explicit version-pinned public provenance of this ClaimVersion — the pinned field-6 Source record.