A 2026 Nature analysis estimated that more than 110,000 scholarly publications from 2025 alone probably contain invalid references generated by artificial intelligence. As AI tools like ChatGPT, Claude, and Copilot become embedded in everyday research workflows, knowing how to cite AI-generated content in research papers is no longer optional — it is a basic requirement for academic integrity. Yet most researchers still aren't sure when citation is needed, which format to use, or what their target journal actually requires.
This guide covers everything you need: when to cite AI, how to format citations in APA, IEEE, MLA, and Chicago styles, what major journal publishers demand in 2026, and practical steps to keep your AI usage documented and audit-ready from first draft to final submission.
When do you need to cite AI in a research paper?
You must cite an AI tool whenever you quote, paraphrase, or directly incorporate its output into your manuscript. This includes text passages, data analysis results, code, images, literature summaries, and any content where the AI's output shaped what appears in your paper. If you used AI only for basic grammar correction or spell-checking, most journals treat this the same as using any word processor feature — no formal citation is required, though disclosure is increasingly recommended.
Here is a practical breakdown:
Cite when you quote or paraphrase AI-generated text, use AI to generate code or statistical analysis, rely on AI for literature summaries or conceptual frameworks, or use AI-generated images or visualizations.
Disclose but citation may not be required when you use AI for grammar and language editing, translation assistance, or brainstorming ideas that you substantially rework in your own words.
No citation needed when you use standard autocomplete, spell-check, or grammar tools built into word processors.
The key principle across every major style guide is the same: if AI meaningfully contributed to the content, cite it and disclose how it was used. When in doubt, disclose. No researcher has ever been penalized for being too transparent.
How to cite ChatGPT and AI tools in APA style
The American Psychological Association treats the AI developer as the author and the AI tool as the work being referenced. APA's approach recognizes that because AI chat outputs are not retrievable by other readers, they function more like algorithm outputs than personal communications.
APA reference format
Reference list entry:
OpenAI. (2025). ChatGPT (May 15 version) [Large language model]. https://chat.openai.com
In-text citation:
(OpenAI, 2025)
Key APA rules for AI citation
Author is the developer of the AI model (OpenAI, Anthropic, Google, etc.).
Date is the year of the version you used.
Title is the name of the model, with the specific version or date in parentheses.
Bracketed description identifies the type of tool, such as [Large language model].
URL points to the general tool URL, not to a specific chat session.
Include the full text of the prompt you used in your paper, an appendix, or supplemental materials so readers understand the context of the AI's output.
APA also recommends adding a note in the author note or method section explaining how AI was used in the research process. If multiple prompts or sessions were used, each distinct use should be cited separately.
How to cite AI-generated content in IEEE format
The Institute of Electrical and Electronics Engineers takes a different approach. Rather than treating AI as a cited source in the reference list, IEEE requires disclosure in the acknowledgments section of any submitted article. The March 2025 IEEE Editorial Style Manual makes this explicit.
IEEE disclosure requirements
Identify the AI system used (e.g., ChatGPT, Claude, GitHub Copilot).
Specify which sections of the article used AI-generated content.
Explain the level of AI involvement — did the tool draft text, generate code, produce figures, or assist with data analysis?
Basic grammar and language editing with AI is considered common practice and generally falls outside the intent of the policy.
If you do need to cite AI-generated content as a reference in IEEE format, treat it as a software tool:
[1] OpenAI, "ChatGPT," version GPT-4o, 2025. [Online]. Available: https://chat.openai.com
The critical difference from APA is that IEEE emphasizes acknowledgment-section disclosure over formal reference-list citation. Your paper must make transparent exactly how AI contributed, even if the reference list itself doesn't include an AI entry.
How to cite AI in MLA and Chicago styles
MLA format
The Modern Language Association does not treat AI as an author. Instead, MLA treats the prompt as the title of the work and the AI tool as the container.
Works Cited entry:
"Describe the role of CRISPR in gene therapy research" prompt. ChatGPT, May 15 version, OpenAI, 15 May 2025, chat.openai.com.
MLA recommends noting the full prompt (or at least its opening words) in quotation marks, followed by the word "prompt," then the AI tool name in italics, the version, the developer, the date you generated the content, and the URL.
Chicago style
The Chicago Manual of Style (18th edition) recommends treating AI-generated content like a personal communication cited in footnotes, since the output is not retrievable by others and does not belong in the bibliography.
Footnote format:
- ChatGPT, response to "Explain the PRISMA framework for systematic reviews," OpenAI, March 7, 2025.
For subsequent citations of the same tool, a shortened note — simply "ChatGPT" — is sufficient. Chicago also requires authors to make clear in the text or a preface how AI tools were used throughout the work.
What do major journal publishers require in 2026?
As of 2026, approximately 83% of high-impact journals have explicit AI policies, and the consensus across publishers is clear on four points: AI can assist with writing, AI cannot be listed as an author, AI-generated content must be disclosed, and human authors bear full responsibility for everything in the manuscript — including any errors AI introduces.
Here is how major publishers handle AI disclosure:
Nature and Springer Nature
Nature Portfolio requires that any use of AI tools beyond basic copy-editing be declared in the methods or acknowledgments section. AI tools cannot be listed as authors because they cannot meet accountability criteria required for authorship. Nature also prohibits AI-generated images for publication and expressly asks peer reviewers not to upload manuscripts to AI tools, protecting the confidentiality of the review process.
Elsevier
Elsevier requires authors to disclose AI usage during the submission process. AI and AI-assisted technologies cannot be listed as authors, and authors must verify the accuracy of all AI-assisted content. Elsevier's policy emphasizes that using AI does not exempt authors from responsibility for originality, accuracy, and the absence of plagiarism.
Taylor & Francis
Taylor & Francis requires a clear acknowledgment statement that includes the full name of the AI tool with version number, how it was used, and the reason for its use. This statement must appear in the methods or acknowledgments section. The publisher retains discretion over whether to publish work that has used generative AI.
PNAS
The Proceedings of the National Academy of Sciences requires that AI assistance be noted in the materials and methods section (or acknowledgments, if no methods section exists). AI software cannot be listed as an author because it does not meet authorship criteria and cannot be held accountable for data integrity.
SAGE and APA journals
Both SAGE and the American Psychological Association require disclosure of AI use and emphasize that authors should cite original sources rather than AI tools as primary references. SAGE specifically warns authors to verify AI outputs because large language models can generate false content, including fabricated citations.
The hallucinated citations crisis: why accurate AI citation matters
The urgency around proper AI citation has grown dramatically because of the hallucinated citations problem — AI tools fabricating references that look real but point to papers that do not exist.
A landmark 2026 Nature investigation, conducted in collaboration with Grounded AI, analyzed over 4,000 publications from five leading publishers. Manual checks confirmed that 65 out of the 100 most suspicious publications contained at least one invalid reference. Extrapolated across the roughly 7 million scholarly publications from 2025, this suggests that more than 110,000 papers may contain AI-generated invalid references.
The problem is accelerating. An analysis of nearly 18,000 papers accepted by three computer-science conferences found that 2.6% of 2025 papers had at least one potentially hallucinated citation — up from just 0.3% in 2024. In a controlled experiment, researchers prompted GPT-4o to generate literature reviews on mental-health topics and found that nearly 20% of the 176 references were completely fabricated, while 45% of the real references contained errors such as incorrect DOIs.
These "Frankenstein citations," as researchers call them, combine fragments of genuine publications — real author names paired with fabricated titles, or real journal names paired with nonexistent DOIs. They pass a quick visual check but fall apart under verification.
This is exactly why research teams need robust systems for tracking which references came from AI suggestions versus verified database searches. ScholarDock, a research project and reference management platform, helps teams maintain citation integrity by keeping every reference linked to its source, making it easy to flag and verify AI-suggested citations before they enter a manuscript.
Best practices for documenting AI use in your research workflow
Proper AI citation is not just about formatting a reference correctly at the end — it requires a documentation workflow that starts when you first open an AI tool and continues through submission.
1. Log every AI interaction as it happens
Keep a running record of which AI tool you used, the date, the prompt, and a summary of the output. This doesn't need to be elaborate — a simple spreadsheet or project note works. What matters is that you can reconstruct exactly what AI contributed to your work.
2. Save prompts and outputs
Many AI tools don't maintain permanent, shareable logs. Copy and paste the prompt and the relevant portions of the AI's response into your research notes immediately. APA and MLA both recommend including prompts in your paper or supplemental materials.
3. Verify every AI-suggested reference
Never include a citation that an AI tool suggested without independently confirming that the paper exists, the authors are correct, the journal is real, and the DOI resolves to the actual publication. Given that studies show citation error rates in the broader literature already range from 25% to 54% even without AI involvement — and only about 20% of authors read the original paper they cite — adding unverified AI references into the mix creates serious integrity risks.
4. Separate AI-assisted sections in your draft
As you write, flag which paragraphs or sections received AI input. This makes it straightforward to write accurate disclosure statements later and ensures nothing slips through during revisions.
5. Use a centralized research management platform
Tracking AI usage across multiple collaborators, drafts, and projects is where most teams struggle. ScholarDock brings your references, project notes, and collaborative workspace into a single platform, making it simple to annotate which sources were AI-discovered versus manually sourced, keep shared reference libraries verified and up to date, and maintain audit-ready documentation of your entire research workflow across team members.
6. Write your disclosure statement before submission
Don't leave the AI disclosure for last. Draft it alongside your methods section so you capture the details accurately while they are fresh. Review your target journal's specific AI policy — requirements vary significantly between publishers.
How to choose the right citation format for your paper
The citation style you use depends on your target journal or institution, not personal preference. Here is a quick decision framework:
Check your target journal's author guidelines first. Many journals now include specific AI citation instructions that may differ from the general style guide.
Follow the style guide your journal requires — APA, IEEE, MLA, Chicago, Vancouver, or another format.
When the journal has no AI-specific guidance, default to the general principles of your required style and add an explicit disclosure in the acknowledgments section.
For multi-author collaborative papers, ensure every team member documents their AI usage consistently. Establish a shared protocol at the start of the project, not at submission time.
Research teams managing multiple ongoing manuscripts across different journals and styles benefit from having a centralized system that tracks citation formats and disclosure requirements per project. ScholarDock's project-based workspace structure lets teams maintain separate reference libraries and documentation standards for each manuscript while keeping everything accessible from one dashboard.
What happens if you don't disclose AI use?
Failing to disclose AI use in a research paper carries growing consequences. Journals are investing heavily in detection tools — publisher Frontiers reports that around 5% of submissions now show potential reference-related issues flagged through automated integrity checks. Taylor & Francis editors have reported rejecting up to 25% of submissions in a single month due to fake references.
If undisclosed AI use is detected after publication, consequences can range from published corrections and expressions of concern to full retractions. Beyond the paper itself, researchers risk damage to their professional reputation, loss of trust from collaborators and funding agencies, and potential investigation by their institution's research integrity office.
The safest approach is simple: document everything, disclose transparently, and verify every reference. The academic community is rapidly normalizing AI use in research — the stigma is not in using AI, but in hiding it.
Moving forward with confidence
Citing AI-generated content in research papers is a skill every researcher now needs. The rules are still evolving, but the core principles are stable: be transparent, cite the tool, disclose how it was used, and verify everything.
Start by understanding your target journal's AI policy. Use the citation format your style guide requires. Build a documentation habit that captures AI interactions as they happen, not retroactively at submission time. And verify every reference — human-sourced or AI-suggested — before it enters your manuscript.
If your research team is juggling multiple projects, collaborators, and citation styles while trying to keep AI documentation organized, ScholarDock brings your entire research workflow — references, projects, notes, and team collaboration — into one connected workspace where nothing gets lost and every source is traceable.