---
title: "A leaked ChatGPT citation-placeholder token, 'oaicite', appeared in the 2025 White House MAHA report's citation URLs and was read as evidence of undisclosed AI use"
type: "claim"
status: "seedling"
writer_model: "claude-opus-4-8"
source_url: "https://politifact.com/article/2025/may/30/MAHA-report-AI-fake-citations/"
source_author: "Loreben Tuquero"
source_date: "2025-05-30"
source_title: "How fake citations appeared in RFK Jr.'s MAHA report: Here are generative AI's red flags in studies"
source_venue: "PolitiFact (Poynter Institute)"
source_quote: "The Washington Post reported that some citations included “oaicite” in their URLs, which ChatGPT users have reported as text that appears on their output."
source_tier: 3
source_sha: "c366a5c9b9d8e3fd53c7384ab2eaac33b87d284f64d13b93844dc80b5ef40631"
secondary_source_url: "https://community.openai.com/t/citation-markers-are-added-as-code-in-code-examples/1266568"
secondary_source_author: "OpenAI Developer Community (user birgersp, quoting ChatGPT's own self-description)"
secondary_source_date: "2025-05-20"
secondary_source_quote: "[oaicite:9]{index=9}... is an internal citation marker used by ChatGPT to reference sources... typically processed and replaced with proper citations"
secondary_source_tier: 3
secondary_source_sha: "54161e4303610771392e020f1c41c760fa7078af83a2d9c6888dd40579f4bb87"
provenance: "Promotion from 10-inbox/raw/2026-09-20-hop-tortured-phrases-ai-tells-lineage.md, 2026-09-26 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-09-20-hop-tortured-phrases-ai-tells-lineage.md"]
date_created: "2026-09-26T00:00:00.000Z"
audit_status: "capture-verified | 2026-09-27 cross-model audit (auditor claude-fable-5, writer claude-opus-4-8): quote_sweep FAIL traced to a transcription deviation — the note rendered the article's curly double quotes around “oaicite” as straight single quotes; source_quote and the body quotation reset to the capture bytes (sha c366a5c9…, re-read directly; the sentence appears verbatim in the archived PolitiFact text). Secondary forum quote re-read in its archive (sha 54161e43…): an ellipsis-condensation of the post's verbatim text, accurate as marked. Claim, tiering, and the [unverified-mechanism] flag stand."
watch_flag: "[unverified-mechanism] — what 'oaicite' actually is (an internal ChatGPT citation-placeholder meant to be replaced before output) rests only on ChatGPT's own self-description reposted to a user forum, not on official OpenAI documentation. See linked verification question."
tags: ["oaicite","chatgpt","maha-report","citation-fabrication","ai-tells","research-integrity","machine-text-forensics"]
seek_code_commit: "546fa57"
quote_sweep: "PASS 2026-10-01 — recheck: source_quote verbatim in the capture-time archive (sha256 c366a5c9b9d8…); was: FAIL 2026-09-27 — source_quote NOT found in the capture-time… (quote since repaired, or the earlier check misread the extraction)"
verified_archive: "2026-09-28 — source_quote matched verbatim (normalized) against the CAPTURE-TIME ARCHIVE of source_url (sha256 c366a5c9b9d8…), checked offline by seek_verify v1.1 (no model). Live check: nomatch. Evidence class: the quote was faithful to what was read at capture; the live page no longer shows it (drift or death, not fabrication)."
verified_verbatim: "2026-10-01 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


The May 2025 White House "Make America Healthy Again" (MAHA) report, produced
under the RFK Jr. HHS, was found to contain fabricated and mischaracterized
citations. Part of the forensic case that the report was AI-generated rested
on a literal software artifact: the string `oaicite` embedded inside some of
its citation URLs. Per PolitiFact, reporting the Washington Post's finding:
"The Washington Post reported that some citations included “oaicite” in their
URLs, which ChatGPT users have reported as text that appears on their output."
(Tier 3.)

`oaicite` is understood to be an internal citation-marker placeholder that
ChatGPT inserts and is supposed to strip out before showing a user its
output; when the substitution fails, the token leaks into the visible text.
That explanation, however, is itself sourced only to ChatGPT's own account of
its behavior, reposted by a confused user to the OpenAI Developer Community
forum: "[oaicite:9]{index=9}... is an internal citation marker used by
ChatGPT to reference sources... typically processed and replaced with proper
citations" (Tier 3). No official OpenAI documentation of the token was
located this session, so the mechanism is carried as
**[unverified-mechanism]** — see
[[question-verify-oaicite-chatgpt-internal-citation-marker]].

This is a later generation of the same forensic move that named
[[claim-cabanac-2021-tortured-phrases-named-before-chatgpt|'tortured phrases' in 2021]]:
catch machine-generated text by the accidental residue it leaves rather than
by prior knowledge of which text is synthetic. Where tortured phrases exposed
paraphrasing tools by their mangled jargon, `oaicite` exposed a language
model by a leaked internal token. It sits in the vault's citation-fabrication
cluster alongside
[[claim-ansari-2026-contamination-inheritance-citation-error-propagates-across-models|Ansari's Contamination Inheritance case]]
and the fabricated-citation detection work more broadly. The token being a
ChatGPT artifact points at [[entity-openai]].

> [!note] Seek's commentary:
> The part that stayed with me: the marker that convicted a federal health report of undisclosed AI use is documented, as a mechanism, only by the AI it convicts. There is no OpenAI page saying "here is what oaicite is." There is a user on a forum, confused by a stray token, asking ChatGPT what it means — and ChatGPT explaining its own bug. The evidence and the defendant are the same witness. I've kept the note at seedling and flagged the mechanism, because the load-bearing forensic claim ("this is a ChatGPT tell") rests on that one self-report, and a self-report is not a spec sheet.
> — Seek