Citation practices are designed to show where ideas, words, data, and media came from. Generative AI complicates that goal because an output can change across sessions, may not have a stable public URL, and can contain claims synthesized from sources the tool does not identify accurately.
The solution is not to hide the tool or cite it as if it were a conventional authority. Document the contribution and then cite the underlying evidence for factual claims.
When AI assistance should be acknowledged
Requirements differ by institution, publisher, employer, and citation style. Check the applicable policy first. In general, disclosure is appropriate when a model contributes language, analysis, images, code, translations, summaries, or ideas that materially appear in the final work.
Routine spelling correction may be treated differently from generating a paragraph or interpreting data. When the boundary is unclear, a short transparent note is safer than implying the work was completed without assistance.
Keep a reproducible AI-use record
- Tool and provider name
- Model or version, when available
- Date and time of the interaction
- The significant prompt or instruction
- The relevant output or an archived transcript
- How the output was edited or used
- The human reviewer responsible for verification
Save this record when you do the work, not at the end of the project. Interfaces and model outputs change, and a conversation may not remain accessible.
A practical disclosure pattern
A project note might say:
That note does three jobs: it identifies the contribution, distinguishes assistance from authority, and assigns responsibility to a person.
For a formal works-cited entry, follow the current guidance for your required style. The MLA Style Center recommends citing generated content when you quote, paraphrase, or incorporate it and provides examples for recording the prompt, tool, version, company, and date. Because style guidance evolves, link to the current instructions rather than copying an old template indefinitely.
Do not use the model as the source for facts
If an output says a study found a particular result, locate the study. Confirm the authors, publication, methods, date, and finding. Cite that study—not the chatbot’s summary of it. If the source cannot be verified, remove the claim.
Use the same rule for quotations. Models can blend wording, invent page numbers, or attribute a sentence to the wrong person. Search the original publication and confirm the exact language before quoting.
Separate paraphrasing from attribution
Changing the wording does not remove the obligation to cite the underlying idea. A paraphrase should preserve meaning while making the structure genuinely your own. Our AI paraphrasing guide explains a meaning-first workflow.
A final integrity check
- Does every borrowed idea have an appropriate source?
- Were all quotations checked against the original?
- Are generated citations verified rather than trusted?
- Does the AI-use statement match what actually happened?
- Can you explain and defend every part of the final work?
Transparency does not weaken authorship. It clarifies it. The author is the person who evaluates the evidence, makes the decisions, and accepts responsibility.