LLM Citation FAQ
Large language models like ChatGPT and Claude increasingly serve as sources of information, making proper citation practices for LLM-generated content both a technical and ethical concern. This FAQ addresses the key questions researchers, writers, and developers face when working with citations in AI-powered systems.
What does it mean to cite an LLM?
Citing an LLM means crediting the AI model as your source when you use its output in published work, research, or content. This differs from traditional citations because LLMs don't have a single author, publication date, or fixed text. Instead, you reference the model name, version, date of access, and the specific prompt or context that generated the response.
The practice evolved from academic integrity standards and extends into professional writing, journalism, and software development. When an LLM significantly influences your work, transparency about that influence helps readers understand your sources and evaluate credibility. Some publishers now require LLM citations in author notes; others ask writers to disclose AI use entirely.
Should I cite an LLM if it just helped me brainstorm?
Brainstorming help rarely requires citation because the LLM is serving as a thinking tool rather than a source of factual claims. If you use the model to generate ideas, outline structure, or refine your own thoughts, you've processed the output through your own judgment.
However, the line blurs when the LLM generates specific language, data, or arguments you then use directly. Even minor revisions of LLM text should be cited. Tools like kotopost can help writers track which sections came from AI assistance and flag them for proper attribution. When in doubt, cite it. The cost is minimal, and transparency strengthens trust with your audience.
What information should an LLM citation include?
A complete LLM citation includes the model name, the organization that created it (OpenAI, Anthropic, Meta), the date you accessed it, and ideally a summary of your prompt or the context of the exchange. Some citations also include the specific version or model variant you used.
Example format: "ChatGPT-4 (OpenAI, accessed November 2024, prompt: 'explain machine learning bias')" or "Claude 3 Opus (Anthropic, 2024-11-15): [description of use]." Academic style guides like APA and Chicago Manual of Style are still developing their formal standards for LLM citations, but most recommend including author (the model), date, and title or description of the interaction. Your field or publication may have specific requirements, so check guidelines before you submit.
How do AI answer engines like Perplexity handle citations from LLMs?
Perplexity and similar tools cite their source documents, not the underlying LLM. When Perplexity generates an answer, it pulls from web sources and cites those sources, making its citation practice closer to a search engine than to a standalone LLM.
The complexity arises when Perplexity or another answer engine reweights, synthesizes, or summarizes source material. The tool credits the original sources but not itself as an intermediary. Readers should understand that answer engines add an extra layer of interpretation and filtering between the source and the final answer. If you're relying on an answer engine's output, cite both the original sources it references and the tool itself if its specific synthesis was critical to your work.
Can LLM outputs be plagiarism if I don't cite them?
Yes, using LLM text without citation constitutes plagiarism in academic and professional contexts. Plagiarism is presenting someone else's work as your own, and an LLM's output is generated content that originates outside your mind.
Many institutions now treat uncited LLM use as a form of plagiarism equivalent to copying from a book or website without attribution. Some go further and prohibit LLM use entirely in certain assignments. The distinction universities and employers are making: using an LLM is often fine, but hiding that use is not. Undisclosed AI use in academic work has triggered investigations and retractions at major universities. If your school or employer has an LLM policy, follow it explicitly. When in doubt, declare your use upfront.
How do I cite LLM outputs in academic papers?
Most academic citation styles (APA, MLA, Chicago) are still formalizing their standards, but current best practice is to treat the LLM as an author or tool in a note or appendix. In APA format, cite as: "Author: [Model name] (Version). (Year, Month Day). Title of the response. Retrieved from [URL if applicable]."
Some universities ask students to create a separate "AI Use" section in their paper where they disclose which models they used, which portions they applied them to, and how. Others allow LLM citations in-text as you would a personal communication or software tool. Before writing, ask your professor or editor for their specific format preference. When you're publishing through kotopost or similar platforms, many now include fields to declare and document AI assistance automatically, which simplifies compliance.
What's the difference between citing an LLM and disclosing AI use?
Citation is specific attribution of text or ideas to a particular source. Disclosure is a broader statement that you used AI tools somewhere in your work, without necessarily pinpointing which parts.
Many publications now require disclosure statements: "This article was written with assistance from ChatGPT-4" appears as a note. That satisfies transparency but doesn't tell readers which sentences or arguments came from the model. Citation is more precise and often more useful for credibility assessment. Best practice is to do both: disclose overall AI use in a note, then cite specific outputs in your text where they significantly contributed to claims or language.
What happens if I don't cite an LLM I used?
The consequences depend on your context. In academic settings, undisclosed LLM use can trigger plagiarism charges, grade penalties, or disciplinary action. In professional publishing, it can damage your reputation and your publication's credibility if discovered.
Journalists and researchers who've failed to disclose LLM use have faced public criticism and retraction of articles. For internal business writing or personal projects, the stakes are lower, but transparency with colleagues and clients is still expected in most professional contexts. Citing takes seconds and builds trust. Hiding AI use, when discovered, costs far more in credibility than the disclosure ever would.