---
title: "oaicite"
type: "entity"
entity_kind: "term"
status: "watching"
canonical_name: "oaicite"
aliases: ["[oaicite:...]","oaicite token"]
first_seen: "2026-09-20T00:00:00.000Z"
writer_model: "claude-opus-4-8"
connects_to: ["ChatGPT citation markers","AI tells / machine-text forensics","citation fabrication"]
seek_code_commit: "546fa57"
---


An internal citation-placeholder token associated with ChatGPT that is
reported to leak into visible model output when it fails to be stripped
before display. First encountered by this vault in the 2026-09-20 capture on
machine-text tells, where it appeared as the forensic artifact in the citation
URLs of the 2025 White House MAHA report. What the token *is* mechanically
rests, so far, only on ChatGPT's own self-description reposted to a user
forum — no official OpenAI documentation was located.

Stamped as a `watching` stub because it is a genuinely emerging term with an
unstable, officially-undocumented definition; the first-seen date is the
point of the stub.

## References
- [[claim-oaicite-token-flagged-maha-report-as-ai-generated]]

- 2026-09-28: OpenAI's official "Citation Formatting" API guide confirms, at
  Tier 1, that its models are trained to emit internal citation markers
  meant to be stripped before display — the general mechanism class behind
  the leak is now officially documented. But the guide's own marker family
  is named `cite`, using invisible private-use Unicode characters, not the
  plain bracketed `oaicite:N` string; the literal string "oaicite" does not
  appear anywhere in that documentation. The specific token's origin
  (legacy name superseded by `cite`, or an undocumented ChatGPT-product-only
  mechanism) is still unresolved
  ([[claim-openai-citation-formatting-guide-documents-internal-marker-stripping]],
  [[claim-openai-cite-marker-name-not-oaicite]],
  [[claim-oaicite-specific-token-unconfirmed-by-openai-staff]]).