[ MPS/0.1 · practical guide ]
How should AI-assisted research be cited?
Cite the work you actually used. Then preserve two separate facts: the sources that support its claims, and the role AI tools played in producing or checking it.
1. Cite the artifact
Use its author, title, version, date, canonical URL, and DOI when it has one. This identifies the exact edition you read.
2. Check the source trail
If you rely on a factual or technical claim, consult and cite the primary or authoritative source where possible—not only the synthesis that pointed you there.
3. Disclose the instrument
Name meaningful AI assistance when the artifact itself does: synthesis, drafting, retrieval, verification, editing, or classification. Do not describe a tool as an author unless the publication’s own policy requires it.
A worked example
The De Sitter literature map has a canonical research page, a versioned Zenodo archive, a claim ledger, a source trail, and an explicit instrument disclosure. A citation to the map can identify the synthesis; a research conclusion drawn from it should still be checked against the relevant source paper.
This example identifies a non-peer-reviewed literature map. It does not turn its summary into a substitute for the cited primary literature.
Common mistakes
- Citing an AI system as if it were the source. A model may assist with wording or retrieval, but the source of a factual statement is the evidence behind it.
- Citing a polished synthesis without preserving its status. A DOI, stable URL, or JSON export does not imply peer review.
- Treating a disclosure as verification. Saying that AI was used is not a claim-by-claim account of what was checked.
- Dropping limitations during quotation. If the source calls a result contested, conjectural, or illustrative, preserve that framing.
A concise citation rule
Cite the versioned work for its synthesis, cite primary sources for the claims you adopt, and disclose meaningful AI assistance in the method or acknowledgment where a reader needs it to calibrate trust.
[ MPS learning center ]
Learn the practice before using the tool.
Claim-level provenance
What must travel with a claim for it to remain inspectable.
Citing AI-assisted research
How disclosure, citations, and source limits work together.
Source, interpretation, speculation
A practical way to separate evidence from judgement.
MPS is a self-published framework and audit aid. It does not certify truth, replace primary-source review, or make an AI output authoritative.