This page provides detailed guidance on the use of generative AI tools in content deposited on Zenodo. It supplements our page on unsuitable content, which lists AI-generated content without a verifiable basis in research conducted by humans as not suitable for Zenodo.
Content on Zenodo is part of the scholarly record, and depositors are responsible for meeting the AI-use standards that journals, conferences, and other scholarly venues require. Zenodo's approach draws on guidance from COPE and on policies adopted by major scholarly publishers, and we expect depositors to follow the same standards.
All content on Zenodo must be grounded in research work conducted by its authors. AI may be used as a tool, but not as the source of the research itself. What matters is whether genuine human-conducted research underlies the output, not whether AI was used along the way.
Author accountability
By depositing a record on Zenodo, the depositor, together with the listed authors and creators, takes full responsibility for all of its contents, regardless of how those contents were produced. If an AI tool contributes inappropriate language, plagiarized material, biased content, fabricated references, factual errors, or otherwise misleading output, and that output ends up in the deposited record, responsibility rests with the depositor and the listed creators, not with the tool. This principle applies equally to text, code, figures, data, and any other content type that Zenodo accepts.
AI authorship
AI tools cannot be listed as authors/creators or contributors of a Zenodo record. Authorship implies accountability, the ability to consent to publication, and responsibility for the integrity of the work. AI tools cannot meet these requirements. This is consistent with the position of COPE and all major scholarly publishers.
Disclosure
Depositors are expected to disclose substantive use of AI tools in the description of their record, for example which tools were used, for what purpose, and which parts of the work were affected. Disclosure helps readers and reviewers assess the work, and journals, funders, and research institutions increasingly expect it. AI tools are disclosed the same way other tools and software used in research are. Specific disclosure conventions vary by field, and depositors should follow the standards of their discipline and target publication venue.
This expectation applies to any generative AI tool whose output contributes substantively to the deposited record, whether that output is text, code, images, data, or any other form.
Where undisclosed AI use has produced problems listed elsewhere in this policy (e.g. fabricated results, hallucinated references), those problems are handled under the relevant section. Non-disclosure on its own is not grounds for removal, but it reduces the transparency of the record and makes issues such as fabrication or plagiarism harder for readers to detect.
Fabricated results
Using AI to generate research results (text, images, figures, data, code, or other evidence) and presenting them as real observations, experiments, or analyses is not allowed. If AI is used to generate code for producing a figure or visualization from real data, the author(s) are responsible for verifying the correctness of this code and the resulting figure, and this should be disclosed. If AI-generated content is part of the research methodology itself (e.g. research on generative models, synthetic data generation with documented methods), this should be clearly described in the submission.
Examples
Using AI tools to assist with legitimate research is not grounds for removal. Examples of acceptable use:
- Editing and improving manuscript language or readability.
- Code generation, where the depositor reviews and takes responsibility for the output.
- Data analysis and visualisation assistance.
- Generating figures or diagrams from the author's own data or specifications, where the author verifies the output is accurate and faithful.
- Literature search and summarisation, where the depositor verifies claims against primary sources.
- Translation of research outputs.
- Structuring or reorganising the author's own notes and drafts.
Examples of submissions we would consider AI-generated content without a research basis:
- A paper produced by prompting an LLM and submitting the output with no additional research work.
- Survey or review papers that are essentially AI-generated annotated bibliographies and/or lack original analysis or critical engagement by the author(s).
- AI-generated "research findings" that present speculative claims as empirical results.
- AI-generated images or figures presented as experimental photographs, microscopy images, or other empirical observations.
- Synthetic datasets presented as real experimental data without documented methodology.
- Raw AI outputs uploaded as primary research artefacts.
- Bulk submissions of AI-generated content across multiple records.
Indicators
Common signs that a submission may not meet the research-basis requirement:
- Core ideas, arguments, or findings that originate from AI output rather than from the uploader's own research, empirical work, or engagement with existing literature.
- Artifacts typical of unreviewed AI output: fabricated references, internal contradictions, or claims the author clearly has not checked.
- Speculative or theoretical claims presented as established findings without empirical evidence, peer review, or prior publication in recognized research venues.
Please contact us on our support line if you are in doubt whether a specific use of Zenodo is within scope.